Individualized Pedestrian and Micromobility Routing Incorporating Static and Dynamic Parameters

Full text

(1)LiU-ITN-TEK-A--21/044-SE. Individualized Pedestrian and Micromobility Routing Incorporating Static and Dynamic Parameters Adam Grachek 2021-06-21. Department of Science and Technology Linköping University SE-601 74 Norrköping , Sw eden. Institutionen för teknik och naturvetenskap Linköpings universitet 601 74 Norrköping.

(2) LiU-ITN-TEK-A--21/044-SE. Individualized Pedestrian and Micromobility Routing Incorporating Static and Dynamic Parameters The thesis work carried out in Transportsystem at Tekniska högskolan at Linköpings universitet. Adam Grachek Norrköping 2021-06-21. Department of Science and Technology Linköping University SE-601 74 Norrköping , Sw eden. Institutionen för teknik och naturvetenskap Linköpings universitet 601 74 Norrköping.

(3) Upphovsrätt Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare – under en längre tid från publiceringsdatum under förutsättning att inga extraordinära omständigheter uppstår. Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och tillgängligheten finns det lösningar av teknisk och administrativ art. Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller i sådant sammanhang som är kränkande för upphovsmannens litterära eller konstnärliga anseende eller egenart. För ytterligare information om Linköping University Electronic Press se förlagets hemsida http://www.ep.liu.se/ Copyright The publishers will keep this document online on the Internet - or its possible replacement - for a considerable time from the date of publication barring exceptional circumstances. The online availability of the document implies a permanent permission for anyone to read, to download, to print out single copies for your own use and to use it unchanged for any non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional on the consent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement. For additional information about the Linköping University Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its WWW home page: http://www.ep.liu.se/. © Adam Grachek.

(4) Abstract As increased demand is placed on the transportation sector from growing global urbanization, enhancing the appeal for alternative transportation in the form of pedestrian and micromobility modes is one solution for alleviating urban transportation networks, subsequently decreasing congestion, carbon emissions, and negative health outcomes associated with air pollution. Despite the clear benefits these mobility options provide, many individuals might stray away from pedestrian and micromobility modes due to lacking personalization in mainstream routing applications. Without user input into the routes provided, these commuters could misinterpret these modes to be more burdensome, polluted, difficult to navigate, and time consuming, in comparison to transportation in the form of personal automobiles and public transportation. Additionally, pedestrian groups such as the disabled, elderly, and individuals with children and prams, are unable to specify route features that are tailored to their needs in conventional routing applications. With these issues highlighted, this project seeks to demonstrate routing optimization that would allow these pedestrian and micromobility user groups to select and prioritize different route features according to their preferences. Through the creation of a routing demonstrator that considers both static and dynamic parameters in the form of pavement quality, elevation climb, travel time, and air quality, along with user-specified weights for their prioritization of each of these parameters, a number of routes have been created and mapped to qualitatively compare against routes representing only a shortest path. These parameters have been incorporated utilizing a series of cost assignment matrices and the development of a routing mechanism utilizing Dijkstra’s shortest path algorithm. By comparing these user modified routes against routes of only a shortest path, it was found that preferences for different features could be accommodated to match user specifications for the different route characteristics represented by these parameters. Potential improvements to this routing demonstrator are also discussed for future considerations of work..

(5) Acknowledgements During this project I have received help from many individuals with consideration of theory, content, technical aspects, and general guidance. Above all I would like to thank my advisor Vangelis Angelakis for his consistent support and knowledge in not only helping me identify my thesis topic, but also advising me on details regarding project direction. I would also like to thank my examiner David Gundlegård in helping ensure my project maintained a consistent goal and direction early on in the project formulation. Other notable mentions I would like to thank include Rasmus Ringdahl for his support with aspects related to the Project: Testbed Kungsgatan funded by Norrköpings fond för forskning och utveckling. Additionally, I would like to thank Hampus Nilsson and Erik Johansson for their opposition contribution to my project. Norrköping June of 2021 Adam J. Grachek.

(6) Table of Contents 1. Introduction .................................................................................................................................................... 1 Aim & Purpose ...................................................................................................................................... 2 Research Questions ............................................................................................................................... 3 Methodology .......................................................................................................................................... 3 Limitations ............................................................................................................................................. 4 Outline ................................................................................................................................................... 4 2 Literature Review ........................................................................................................................................... 5 2.1 Cyclist Route Planning with Varying Parameters .......................................................................................... 5 2.2 Cyclist Route Planning with Qualitative Routing Features............................................................................ 5 2.3 The Fietsersbond Route Planner .................................................................................................................... 6 2.4 Multiple Attribute Cyclist Route Planning ..................................................................................................... 6 2.5 Routing for Reduction of Air Pollution Exposure to Cyclists Montreal ......................................................... 7 2.6 Land Use Regression Model for Estimating NO2 Concentrations .................................................................. 7 2.7 Simulation of Air Pollutants in a Street Canyon in Stockholm ....................................................................... 7 2.8 Travel Time, Speed, and Delay Calculations for Cyclists .............................................................................. 8 2.9 Understanding Human Movement in Built Environments .............................................................................. 8 2.10 Comparison of Wheelchair Routing Algorithms............................................................................................. 9 2.11 Contributions from the Literature Review to the Thesis Project .................................................................... 9 3 Routing Demonstrator .................................................................................................................................. 10 3.1 Link Distance and Node Coordinates ........................................................................................................... 15 3.2 Pavement Quality and Elevation Climb Parameter ..................................................................................... 16 3.3 Travel Time Parameter ................................................................................................................................ 20 3.4 PM2.5 and PM10 Air Pollution Avoidance Parameter ................................................................................... 24 3.5 Routing Algorithm ........................................................................................................................................ 26 4 Results .......................................................................................................................................................... 28 5 Discussion .................................................................................................................................................... 42 6 Conclusions & Future Work ......................................................................................................................... 44 6.1 Improvements to Pavement Quality Categorizations ................................................................................... 44 6.2 Improvements to the Elevation Climb Parameter ........................................................................................ 45 6.3 Improvements to the Air Pollution Avoidance Parameter ............................................................................ 45 6.4 Improvements to the Travel Time Parameter ............................................................................................... 46 6.5 Incorporation of Other Parameters .............................................................................................................. 46 6.6 Route Selection Satisfaction Surveying and Sensitivity Analysis .................................................................. 47 6.7 Expansion into a Working Mobile Application ............................................................................................... 47 7 Copyright...................................................................................................................................................... 48 8 References .................................................................................................................................................... 48 9 Appendix ...................................................................................................................................................... 49 1.1 1.2 1.3 1.4 1.5.

(7) 1 Introduction As increasing urbanization around the world presents new challenges and stressors on cities, transportation in particular is one area that will face considerable issues in meeting the demands of an ever growing urban population. With an increased demand for transportation services, the issues associated with transit systems are expected to increase with a larger population as well (Rodrigue, et. al, 2013). As cities continue to urbanize, the transportation system can expect to see greater congestion, increased carbon emissions, and adverse health outcomes from inhalation of air pollutants from this sector (Rodrigue, et. al, 2013). One alternative to relieve stress placed on the transportation system in a city is to increase the number of trips taken using conventional methods such as walking, cycling, or other micromobility modes. For purposes of this project, micromobility as a mode can be defined as mobility devices containing these three characteristics, as defined by the Pedestrian and Bicycle Information Center: 1. Motorized completely without the assistance of human interference and/or utilize propulsion with the influence of humans, by for example pedaling or kicking 2. Contain low speed, typically less than 20 miles/hour (~32 km/hour) 3. Are small in size, typically less than 3 feet (~ 0.9 meters) and less than 100 lbs. (~45 kg), such that these devices can fit on a bike lane or sidewalk (Pedestrian and Bicycle Information Center, 2019) Additionally, for purposes of this project wheelchairs and other mobility devices for those who are disabled are not included as a micromobility mode. Individuals using these devices are considered to be pedestrians within this project. To promote these mobility options as an alternative to automobile dependency and public transportation, cities can pursue a variety of methods to increase the appeal of these transit forms. Typical methods for increasing these forms of transportation include the creation of micromobility and pedestrian prioritized infrastructure, programs to boost cycling or walking to work, and educational campaigns, to name just a few (C40 Cities Climate Leadership Group & C40 Knowledge Hub, 2019) Another potential method to increase the appeal of these modes is through better routing mechanisms that incorporate individualized preference for pedestrians and micromobility users. The standard commuter routing applications that are used today include Google Maps and Apple Maps, but when seeking routes as a micromobility user or pedestrian, the choices for alternatives are often limited to a few potential routes and lack the option to include or exclude features that are of interest to the commuter. Furthermore, both of these applications fail to provide options for commuters that require special considerations in the route choice, such as the elderly, disabled individuals, or even parents with prams and children. Including parameters such as pavement quality, level of elevation climb, air quality, and travel time when considering 1.

(8) the overall route selected for a commuter, could potentially increase the overall comfort and enjoyment of a route, and therefore drive more people towards these transportation options. Additionally, incorporating the ability to either dial-up or dial-down these different routing parameters would make routing inclusive of all potential features a commuter might want to consider when choosing their path. While this same implementation of prioritization of preferences for routing characteristics could also be included for other transit modes such as automobiles and public transportation, it is arguable that pedestrians and micromobility users are more impacted by the actual characteristics of the route, influenced by the comfort, feeling, and required level of effort the route demands. Allowing for user prioritization of different routing preferences for pedestrians and micromobility users would both expand the route choice freedom within a network and increase the personalization provided to each commuter. In the case of pedestrians such as the elderly and disabled, the option to modify routes based on the level of elevation climb and the pavement quality is critical to their ability to traverse the route provided, and as of now is completely excluded from mainstream routing applications such as Google Maps and Apple Maps. With the importance and potential of pedestrian and micromobility user prioritization of preferences for routing highlighted, it is important to note that this implementation of routing parameters requires local knowledge and understanding of each link of a route on a more microscopic level. Incorporating the additional characteristics experienced by pedestrians and micromobility users into a routing algorithm can only be done with reliable understanding of each of the link features in the network, and the often varying and changing conditions that are presented as time progresses. To showcase the implementation of routing with user prioritization of preferences, the following parameters could be included into an origindestination routing demonstrator for micromobility and pedestrian commuters: pavement quality, elevation climb, air pollution avoidance, and travel time. These parameters were chosen specifically because pavement quality, elevation climb, and travel time could be represented as static parameters in the demonstrator using a variety of features that are easily identified on mobility links. Air pollution avoidance was chosen because of its potential to represent a dynamic parameter with ease of access to real-time PM2.5 and PM10 air pollution data collected as part of the Project: Testbed Kungsgatan funded by Norrköpings fond för forskning och utveckling (Linköping University, n.d.). The location of air pollution sensors was decided by researchers within the Project: Testbed Kungsgatan funded by Norrköpings fond för forskning och utveckling, and this decision was outside of the control of this thesis project (Linköping University, n.d.).. 1.1 Aim & Purpose The aim of this project is to demonstrate optimization with respect to multiple characteristics of pedestrian and micromobility routing through user prioritization of preferences for air pollution (PM2.5 and PM10) avoidance, pavement quality, elevation climb, and travel time. Furthermore, the difference between these individualized routes created and the shortest path routes without user preferences within the defined test site will be analyzed. 2.

(9) The purpose of this implementation is to demonstrate the incorporation of user preferences in pedestrian and micromobility routing for both static parameters (pavement quality, elevation climb, and travel time) and dynamic parameters (PM2.5 and PM10 air pollution).. 1.2 Research Questions Achieving this aim and purpose will require the following research questions to be considered: •. How should the air pollution collected from street sensors in Norrköping be modeled in the area?. •. What areas of the city should be considered for air pollution avoidance given a set number of sensors?. •. Can pedestrian and micromobility preferences for routing be accommodated and what impact will they have in comparison to a shortest path route?. •. How should the optimization problem be formulated? What algorithms should be considered for implementation?. 1.3 Methodology To go about answering these proposed research questions, a methodology consisting of three parts will be conducted. First a literature review will be conducted to understand the depth of knowledge that currently exists in the relevant areas of air pollution monitoring and modeling, user preference for routing, travel time, and routing optimization. Secondly, a routing demonstrator will be created utilizing a shortest path algorithm modified to user-preference. Once this demonstrator has been established, comparisons between the routes created for the user-preference modified algorithm and that of a shortest path algorithm will be analyzed. Formulated in the literature review, relevant sources related to route planning, route optimization, and air pollution modeling and avoidance were identified through catalog searches on the Linköping University library and in Google Scholar. In addition to sources that provide background to this study area, relevant concepts applied to the routing demonstrator have been summarized for reference. To create the routing demonstrator, many components and theories from other literature works have been incorporated to achieve the intended outcome of accommodated user preference in micromobility and pedestrian routing. The creation of this routing demonstrator consisted of five main components, namely, the allocation of the area for consideration in the study, the creation of the network including relevant links and nodes, assignment of weights to relevant links for parameters, creation of the routing algorithm, and initiating the algorithm and plotting routing outcomes. Notable tools utilized to achieve the creation of this routing demonstrator include MATLAB, MATLAB Mapping Toolbox, Google Maps and Google Street View, and a Dijkstra’s shortest path routing algorithm MATLAB script file with its subsequent functions created by Dimas Aryo in MATLAB File Exchange (Aryo, 2012). Comparison and analysis of the different routing algorithms produced by a Dijkstra’s shortest path algorithm and a shortest path algorithm modified with user-preference for different parameters was then conducted. This comparison between the routing outcomes produced was 3.

(10) a qualitative analysis that focused on identifying and contrasting the routes by features in each link. These four parameters chosen were compared against travel distance rather than travel time, as travel time was considered to be a routing parameter that users would choose to prioritize in their routes. Travel distance was therefore decided to be the standard shortest path algorithm to compare against the results obtained from the demonstrator created.. 1.4 Limitations Within this project aim and purpose, the following limitations can be directly seen: •. The routing algorithm will only be applied to a specific set of areas within Norrköping. •. The air pollution data utilized within the project relies on two sensors on Kungsgatan in Norrköping, and therefore cannot be applied to the city overall, only specific areas. •. The project will not consider intermodal transit, such as the combination of walking and micromobility with other forms of transportation. •. IoT sensors will only be utilized for the collection of air pollution data in real-time, and no other sensor data will be utilized such as traffic congestion and public transport ridership frequency. •. Static values for travel time will be used rather than dynamic real-time data. 1.5 Outline The subsequent sections within this report are the following. Section 2 contains a literature review of relevant research and studies on this topic. Section 3 describes the process result of creating the routing demonstrator of the project, including the allocation of the study area, creation of the routing network with relevant links and nodes, assignment of weights for dynamic and static parameters, creation of the routing algorithm, and initiation of the algorithm and plotting. Section 4 depicts the results of this routing demonstrator and contains qualitative analysis to compare differences between routing with user prioritization and that of shortest path optimization. Section 5 contains a discussion of the impact of this demonstrator implementation for routing. Finally, section 6 concludes the results and analysis of this study with final remarks and consideration for future work.. 4.

(11) 2 Literature Review Prior to the start of work within this project aim, a thorough literature review was conducted to understand the research contributions made to this subject area. Contained within this section is a summary of relevant research considerations, methods, and model formulations related to this thesis project.. 2.1. Cyclist Route Planning with Varying Parameters. In Bicycle Route Planning Using Multiple Criteria GIS Analysis by Jurica Đerek and Marjan Sikora from the University of Split, researchers created a cyclist routing algorithm that considered five different parameters to decide the favorability of a given route (Đerek & Sikora, 2019). These parameters included the length of a road segment, the type of road, the path slope, distance between the path and an emergency unit, and the distance to the nearest drinking water station (Đerek & Sikora, 2019). Categorizations of road type within this study were labeled as either main road, local road, side road, asphalt, and path (Đerek & Sikora, 2019). To begin this study, the researchers first created a network of all roads in the area of Imotski, Croatia (Đerek & Sikora, 2019). The ArcGIS Network Analysis extension was then used to initiate route generation with desired stops and these five parameters according to a traveling salesman problem (Đerek & Sikora, 2019). To utilize the determined parameters in route generation, this study assigned a weighting value to each road segment by first normalizing each of the parameters, so they contribute to route calculation equally (Đerek & Sikora, 2019). This normalization process was done using a min-max normalization technique as seen below in equation 1. 𝑥𝑛𝑜𝑟𝑚 = 𝑥. 𝑥−𝑥𝑚𝑖𝑛. (1). 𝑚𝑎𝑥 −𝑥𝑚𝑖𝑛. (Đerek & Sikora, 2019) This normalization of study parameters resulted in all weights for each parameter being contained within the values of 0 and 1 (Đerek & Sikora, 2019). The result of the model formulation in this study was able to produce cycle routes that are tailored to cyclist experience, including beginner routes with smaller and flatter paths, advanced routes with longer and more intense slope inclines, and a more moderate level in between the two extremes (Đerek & Sikora, 2019).. 2.2. Cyclist Route Planning with Qualitative Routing Features. A similar study titled The Shortest Path to Happiness: Recommending Beautiful, Quiet, and Happy Routes in the City by Quercia, et. al., from Yahoo Labs and the University of Torino, sought to incorporate other parameters related to aspects of emotional enjoyment in addition to shortest path optimization in cyclist routing (Quercia, et. al, 2014). The methodology for creating such a routing algorithm consisted of four different steps. First a network of nodes was implemented representing all of the potential locations in the city of study (Quercia, et. al, 2014). Once a network was established, the appeal of various links in the network was 5.

(12) crowdsourced along three categorizations of beautiful, quiet, and happy (Quercia, et. al, 2014). These links in the network were then assigned weights according to these three categorizations (Quercia, et. al, 2014). The final step in the algorithm was to generate a route between two nodes that decreased the length of the route as much as possible, while also increasing the appeal of these three enjoyment parameters (Quercia, et. al, 2014). When looking at the optimization model used to create the route, the algorithm first identifies an exhaustive number (at least 106) of shortest paths between two nodes utilizing Eppstein’s shortest path algorithm (Quercia, et. al, 2014). A subset of these paths is then taken, and the average parameter ranking is calculated, and the path with the best average ranking (i.e. the lowest cost) is recorded (Quercia, et. al, 2014). This process of averaging parameter values and recording the best path is ended after improvements to the ranking are improved less than a certain threshold value according to cost-benefit trade off using Marginal Value Theorem (Quercia, et. al, 2014). After these three steps, the shortest path is chosen for which has the best ranking among the parameter categorizations (Quercia, et. al, 2014).. 2.3. The Fietsersbond Route Planner. One cyclist route planner that is currently utilized in the Netherlands is the Fietsersbond Route Planner, which allows users to select an origin and destination, along with other static preferences and parameters relevant to their journey (Fietsersbond, n.d.). Preferences included within the route planning are related to avoiding features, for example unpaved routes, ferries, and short cuts (Fietsersbond, n.d). Other route type priorities that can be selected are qualities such as road type including categorizations like neutral roads, bike lanes, cycling along a road stretch, and pedestrian shortcuts to name a few (Fietsersbond, n.d). With each of these categorizations, users can dial up and down what preference they have for each of these features (Fietsersbond, n.d). This same prioritization of features also exists for categorizations such as pavement quality and closeness to specific features (Fietsersbond, n.d).. 2.4. Multiple Attribute Cyclist Route Planning. In one thesis project from KTH, an interview was conducted with Fietsersbond which describes the route planner weight categorization of links within the network (Plynning, 2016). For each of the parameter weights included in the network, for example weights for pavement quality, this weight is multiplied against the length of a segment to reflect the impact of that particular feature on the route selection (Plynning, 2016). The authors of this thesis project from KTH utilized this process as the primary method to apply weights for different parameters to the route calculation (Plynning, 2016). For example, in one cost calculation a prioritization of bicycle paths over motorized roads was applied by multiplying the weight of the road type by the length of the road to either increase or decrease its favorability for route selection in the network (Plynning, 2016).. 6.

(13) 2.5. Routing for Reduction of Air Pollution Exposure to Cyclists Montreal. In the case study analysis A Web-Based Route Planning Tool to Reduce Cyclists' Exposures to Traffic Pollution: A Case Study in Montreal, Canada, researchers from McGill University and Health Canada estimated and measured NO2 concentrations utilizing a land use regression model (Discussed in section 2.6), and then overlayed these concentrations onto road segments within the city and computed the average NO2 concentration per segment (Hatzopoulou, et. al, 2013). From this concentration assignment for each link, a shortest path algorithm from the ArcGIS Network Analyst function was used to find the most optimal route, by multiplying the road segment length by the average concentration (Hatzopoulou, et. al, 2013). The authors of this case study then took Origin-Destination pair data and formulated over 2,300 routes that optimized for the shortest path and routes that optimized for air pollution avoidance (Hatzopoulou, et. al, 2013). The comparison between these route calculations in this study found that for a majority of routes, less exposure to NO2 could be achieved with an alternate route despite the average difference being a modest 5% decrease in NO2 exposures overall (Hatzopoulou, et. al, 2013).. 2.6. Land Use Regression Model for Estimating NO2 Concentrations. The land use regression model utilized in Hatzopoulou, et. al, 2013, was created by some of the same researchers, outlined in the paper A Prediction-Based Approach to Modelling Temporal and Spatial Variability of Traffic-Related Air Pollution in Montreal, Canada (Crouse, et. al, 2009). Utilizing a series of sampling campaigns to collect three different measurements on integrated NO2 concentrations over two-week intervals (from December, May, and August), the researchers created four different models for varying seasons in the year (Crouse, et. al, 2009). With these NO2 measurements, these four land use regression models with various spatial variables such as road density and land use were developed to estimate the NO2 concentrations in areas without sampling locations (Crouse, et. al, 2009). From these estimation models, the concentrations of NO2 were visually and numerically overlayed on a map of Montreal (Crouse, et. al, 2009). These concentrations were utilized to find the average concentration of NO2 per link in the previous study discussed in Section 2.5.. 2.7. Simulation of Air Pollutants in a Street Canyon in Stockholm. In a study titled Simulation of NOx and ultrafine particles in a street canyon in Stockholm, Sweden by researchers from the Swedish Meteorological and Hydrological Institute and the ITM Air Pollution Laboratory at Stockholm University evaluated the concentration and movement of ultrafine particles and NOx in an urban street canyon utilizing a computational fluid dynamic (CFD) model (Gidhagen, et. al, 2004). The overall objective was to discover how the concentration of particle numbers are distributed inside of an urban street canyon, discussing aerosol processes on a spatial scale, and identifying the minimal nucleation mode (Gidhagen, et. al, 2004). Localized to Hornsgatan in Stockholm, street conditions such as the slope and average number of weekday vehicles and vehicle type were recorded (Gidhagen, et. al, 2004). Sensors on both 7.

(14) sides of the street recorded NOx concentrations hourly, while wind speed and direction were recorded by a rooftop sensor 500 meters from the test site (Gidhagen, et. al, 2004). Data for precipitation and temperature were recorded from a Stockholm meteorological tower, where rainy days were noted for exclusion from the study (Gidhagen, et. al, 2004). Utilizing a CFD model, simulation of the gases and particles in the test site was done with 90 x 92 x 29 cells and a volume of 625 x 442.3 x 350 m3 with hourly intervals, incorporating 36 varying directions for wind with constant wind speed (Gidhagen, et. al, 2004). Additional modifications to their model were made to consider the effect of traffic produced turbulence from vehicles passing on the wind velocity (Gidhagen, et. al, 2004). In addition to this CFD model, the researchers also created a monodisperse aerosol dynamic model to more accurately simulate the movement generated by vehicle turbulence in the street canyon (Gidhagen, et. al, 2004).. 2.8. Travel Time, Speed, and Delay Calculations for Cyclists. In the research study titled Speed, travel time and delay for intersections and road segments in the Montreal network using cyclist Smartphone GPS data, researchers Jillian Strauss and Luis Miranda-Moreno utilized filtered GPS data from smartphones to formulate cyclist characteristics across whole road segments and individual intersections for parameters such as travel time, speed, and delay (Strauss & Miranda-Moreno, 2017). Through the results obtained from GPS data, it was found that cyclist speeds was higher on main roads as compared to side links, and when cyclist-specific infrastructure was present (Strauss & Miranda-Moreno, 2017). When considering the delay created from intersections, researchers computed this as the difference in travel time from traversing through an intersection at regular speed against that of travel time with a delay present (Strauss & Miranda-Moreno, 2017). Interestingly, one of the main points of the study found that the most important aspects affecting the speed of cyclists are cyclist behavior and personal preference, and the physical design and characteristics of the urban infrastructure (Strauss & Miranda-Moreno, 2017).. 2.9. Understanding Human Movement in Built Environments. In one study titled Human Movement Behaviour in Urban Spaces: Implications for the Design and Modelling of Effective Pedestrian Environments, researchers observed and studied the walking movement of over 2,000 people on a microscopic level in uncluttered walking scenarios, viewing the movement characteristics such as position on the segment, walking speed, and distance between fellow pedestrians (Willis, et. al, 2004). Additionally, researchers then tied the observations of these characteristics back to the physical and circumstantial scenarios such as group dynamics, that arose as they traversed (Willis, et. al, 2004). From the research conducted, it was found that the average walking speeds of individuals in the sample was 1.47 m/s, with differences in speed also appearing dependent on gender, age, time of day, and group size (Willis, et. al, 2004). From these characteristics, it was found that on average men walked faster than women, walking speed on average decreased with increasing age, average walking speed was higher for individuals as opposed to groups, and average walking 8.

(15) speed was higher at morning and evening “rush hour” and slower during the day (Willis, et. al, 2004).. 2.10 Comparison of Wheelchair Routing Algorithms In the research paper titled Disabled, but at What Cost? An Examination of Wheelchair Routing Algorithms, researchers analyzed the differences between three routing services, and in particular two wheelchair routing services called OpenRouteService and Routino (Tannert & Schöning, 2018). Within the study it was found that the routes provided to users with wheelchairs was generally longer and incorporated more complex routes than those for ablebodied pedestrians (Tannert & Schöning, 2018). For the two wheelchair routing algorithms observed in this study, parameters such as path obstacles like stairs and slope incline are taken into consideration (Tannert & Schöning, 2018). For the Routino wheelchair algorithm, the routing speed was 4 km/h while the OpenRouteService wheelchair routing speed was 8 km/h (Tannert & Schöning, 2018).. 2.11 Contributions from the Literature Review to the Thesis Project In some instances, contributions from parts of the literature review influenced different design aspects of the routing demonstrator created in this project. These contributions are discussed and summarized in Table 1 below. Table 1: Contributions to Project Methods from Literature Review Summarized. Literature Review • Section 2.1. • •. Section 2.2. Section 2.3. • •. Exemplified static parameter prioritization in cyclist routing. •. Exemplified a method for amplifying the weights of links in accordance with the distance the link traverses to obtain more realistic route choices. •. Routing speed for cyclists utilized in travel time calculations. •. Routing speed for average pedestrians utilized in travel time calculations. •. Routing speed for decreased speed pedestrians utilized in travel time calculations. Section 2.4 Section 2.8 Section 2.9 Section 2.10. Contribution to Thesis Project Provided a process to utilize multiple parameters containing weights that contribute to the overall routing algorithm Provided a method to normalize data values across multiple parameters Provided a process for creating network nodes as the basis of the routing demonstrator Depicted a method for categorizing parameters based on qualitative aspects rather than quantitative aspects. 9.

(16) 3 Routing Demonstrator The process of creating the routing demonstrator consists of five main parts: 1. Allocation of the area for consideration in the study 2. Creation of the network including relevant links and nodes 3. Assignment of weights to relevant links for parameters 4. Creation of routing algorithm 5. Initiation of algorithm and plotting routing outcomes The first and second aspects of this process consisted of identifying the relevant area for study and then building a network of nodes and subsequent links for utilization in a routing algorithm. The area in Norrköping in consideration can be seen below in Figure 1, as a network of 141 nodes with Nygatan as the southern edge, Norra Promenaden as the northern edge, Olai Kyrkogata as the eastern edge, and Kungsgatan and Drottninggatan crossing through the center. The approximate locations for the two air pollution sensors from Project: Testbed Kungsgatan funded by Norrköpings fond för forskning och utveckling, can be seen below in Figure 1 as two red dots labeled as sensor 1 and sensor 2. Sensor 1 is located near MyWay on Kungsgatan, and sensor 2 is located near the Visualization Center on Kungsgatan.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 1: Norrköping Network in Consideration. The routing demonstrator created incorporates both static parameters (pavement quality, elevation climb, and travel time) and dynamic parameters (PM2.5 and PM10 air pollution). To 10.

(17) implement these parameters into the routing demonstrator, relevant links connecting the various nodes in the network were established based on the available paths to traverse for a micromobility user, an average pedestrian, and a decreased speed pedestrian such a senior citizen, parent with a pram, or a disabled person. For each parameter, a corresponding matrix of link costs was established for the following: •. Length of the link (distance). •. Elevation Climb Categorized (static parameter). •. Pavement Quality Categorized (static parameter). •. Travel Time (static parameter). •. PM2.5 and PM10 Air Pollution categorized (dynamic parameter). As part of the process for establishing costs in a matrix for each of these parameters, various maps were created to show how relevant links in the network were affected by these characteristics. Assigning classifications for each link for the parameters regarding pavement quality and elevation climb was done with the assistance of Google Street View. Distances between nodes was measured using the “Measure distance” tool in Google Maps. Additionally, a map and subsequent matrix of costs for high congestion links was created for the purpose of configuring travel times in the network area. For each node in the network, a set of coordinates was determined for purposes of route plotting. Prior to discussing in detail the creation of the various parameters utilized for routing, it is useful to go through the components of the routing demonstrator generally and step-by-step for an overview of the general workflow. The demonstrator was created generally according to the workflow seen below in Figure 2.1.. Figure 2.1: General Overview of the Routing Demonstrator. In essence, this demonstrator collects origin, destination, and mobility mode as either a micromobility, average speed pedestrian, or decreased speed pedestrian from a user. In addition it also collects what their prioritization rating would be (1-10) for each of the parameters elevation climb, pavement quality, air pollution avoidance, and travel time. With this input, a series of cost matrices are then established according to data and knowledge of the network for these four parameters, and then normalized using max-min normalization in order to compare the parameters evenly. With these normalized cost matrices for each of the parameters, an overall cost matrix for the network is established also utilizing the user prioritizations for each of the parameters as defined 11.

(18) previously. This overall cost matrix is then used to compute a lowest-cost route according to user prioritization of preferences, and this is then mapped and compared against that of a Dijkstra’s shortest path algorithm. A step-by-step description of this algorithm can be seen below in Figure 2.2, with red components corresponding to the main script file RouteCalc.m, the blue component representing the function databasecall.m, the purple component representing the function airPollution.m, the green component representing the function LoadData.m, and the yellow components representing the functions dijskstra.m, listdijkstra.m, exchangenode.m, and setupgraph.m as created by Dimas Aryo for MATLAB File Exchange (Aryo, 2012). Each component in Figure 2.2 is labeled with a number 1 through 14. When describing matrices, vectors, and variables in this figure, these are referring to the MATLAB matrices, vectors, and variables in the corresponding script files. These components within the diagram are described in more detail here: 1. This component of the main script file RouteCalc.m defines the user’s origin and destination node, the type of mode (i.e. micromobility, average pedestrian, or decreased speed pedestrian) and the user’s prioritization preferences as number between 1 and 10 for the parameters air pollution avoidance, elevation climb, pavement quality, and travel time. 2. In the second component, a built function LoadData.m is called utilizing a singular input of the variable mode. 3. The third component known as the function LoadData.m, first loads the distances on links, coordinates of nodes, and the various cost matrices for link distance and the parameters elevation categorized and pavement quality for each of the user groups, micromobility, average pedestrian, and decreased speed pedestrian. The air pollution cost matrix is also loaded as a skeleton of values of 1 for future modification. In addition, the travel time is also calculated here depending on which user mode is being utilized and saved as a single cost matrix for links in the network. The process for creating these cost matrices is described in greater detail in the subsequent sections 3.1 through 3.4. This function returns these matrices back to the main script file as the vectors dist, pave, elev, ped, WC, and travelTime. 4. In component 4, a built function databasecall.m is called to obtain the PM2.5 and PM10 data collected from the sensors on Kungsgatan. 5. In component 5, known as the function databasecall.m, firstly opens a connection to the database in which the PM2.5 and PM10 data is stored. The SQL query contained in this function requests the minute by minute concentrations of PM2.5 and PM10 data from both sensors from the present moment through to 60 minutes prior as can be seen below. query = ['SELECT device_name, timestamp, pm_10, pm_25 FROM tnk116.pm WHERE device_name = ? AND timestamp > now() - Interval '60 minutes''];. 12.

(19) PM2.5 and PM10 concentrations were split into vectors for each sensor location: PM2.5 at the MyWay sensor location, PM2.5 at the Visualization Center sensor location, PM10 at the MyWay sensor location, and PM10 at the Visualization Center sensor location. These four vectors are then returned to the main script file as myWay10, myWay25, VC10, VC25.. Figure 2.2: Workflow Process Algorithm for the Routing Demonstrator. 6. In component 6, the function airPollution.m is called with the four vectors myWay10, myWay25, VC10, and VC25 as input, to obtain a single cost matrix for the parameter air pollution avoidance. 7. Component 7, known as the function airPollution.m first averages the two PM2.5 and two PM10 vectors to four singular concentrations, one for each PM type at each location. These concentrations are then assigned as cost values for each of the relevant links in the vicinity of the sensors and categorized according to air quality indexes established by the European Environment Agency. The higher categorization between PM2.5 and PM10 on a given link is assigned as the overall cost for that link in the air pollution cost 13.

(20) matrix. This process is described in greater detail in Section 3.4. The function then returns a final cost matrix for air pollution avoidance back to the main script file. 8. In component 8, the matrices for distance and the parameters for pavement quality categorized, elevation climb categorized, air pollution avoidance, and travel time are normalized using max-min normalization in the main script, as exemplified by the research discussed in section 2.1 (Đerek & Sikora, 2019). The output of this component results in a series of matrices of normalized values between zero and one. 9. In component 9, these normalized matrices from component 8, along with user prioritization preferences from component 1, are utilized to formulate a single modified cost matrix. This matrix mat represents the costs on each link in the network, with varying costs depending on the user’s preferences for each of the parameters discussed: pavement quality, elevation climb, air pollution avoidance, and travel time. This process is described in greater detail in Section 3.5. 10. In component 10, the user modified matrix mat representing user prioritization of routing features, the matrix dist from component 3 representing the distances as costs on each link, and the origin and destination from component 1 are used as input into the function dijsktra.m provided by Dimas Aryo from MATLAB File Exchange (Aryo, 2012). The function is called from the main script twice, the first time returning a vector of nodes to traverse based on the lowest-cost route according to the distance of the route (with inputs dist, origin, and dest), and the second time returning a vector of nodes to traverse based on the lowest-cost route according to user preference of the various parameters (with inputs mat, origin, and dest). 11. Component 11 is known as the function dijkstra.m provided by Dimas Aryo from MATLAB File Exchange (Aryo, 2012). This function calls the functions listdijkstra.m, exchangenode.m, and setupgraph.m recognized in component 12, which are also provided by Dimas Aryo from MATLAB File Exchange (Aryo, 2012). Together, these functions provide an output as a vector of nodes to traverse with a given cost matrix, an origin node, and a destination node provided as inputs. These vectors of nodes to traverse for the route according to shortest distance and the route according to user preference of the various parameters are returned back to the main script as the vectors route and routeMod, respectively. 12. Component 12 recognizes the functions listdijkstra.m, exchangenode.m, and setupgraph.m as by Dimas Aryo from MATLAB File Exchange (Aryo, 2012). These are called by the function dijkstra.m in component 11. 13. In component 13, the vectors route and routeMod from component 11, along with the matrix travelTime and vectors coordLat and coordLon from component 3, are utilized to extract the coordinates of the nodes traversed and sum the total travel time along the route. Here total travel time for each route is calculated, and the vectors of latitude and longitude for each route are provided as RouteLat & RouteLon, and ModLat, & MonLon. 14.

(21) for the route according to shortest distance and the route according to user preference of the various parameters, respectively. 14. In component 14, both of the routes calculated are mapped according to their corresponding latitude and longitude of nodes traversed, as given by the parameters RouteLat & RouteLon, and ModLat, & MonLon from component 13. The various script files and functions described in this algorithm are provided in GitHub with a link in the Appendix for further reference. Additionally, the original cost matrices, travel time coefficients, and node coordinates loaded from Microsoft Excel are provided in GitHub with the link in the Appendix for reference.. 3.1. Link Distance and Node Coordinates. Formulated as the matrix for length of each link, the distance in meters between each node was entered. A small sample of the costs assigned to each link for the distance parameter can be seen below in Figure 3. This represents the distances in meters between two nodes for nodes 1 through 10. In the demonstrator for this project, a cost matrix consists of all 141 nodes seen in Figure 1 and their subsequent link costs for distance and each of the different parameters. Throughout all of the matrices created, the higher the cost assignment, the greater the penalty for that specific link.. Figure 3: Cost Map Example for Dijkstra’s Shortest Path Optimization. For purposes of plotting the routes created in the routing demonstrator, coordinates for each node were obtained using Google Maps, and placed inside a sheet in Microsoft Excel. This is loaded into the main script RoutCalc.m in Component 3 of Figure 2, in a similar fashion as the original cost matrices for distance and the various parameters.. 15.

(22) 3.2. Pavement Quality and Elevation Climb Parameter. With the distances for each link and the coordinates of each node recorded, the pavement quality of each link for micromobility users was then mapped as can be seen below in Figure 4.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 4: Micromobility Pavement Quality Categorized within the Network. As can be seen in the key in Figure 4, there are five different categorizations of links within network: bike lane (green), gravel path (yellow), cobblestone (orange), neutral road path (black), and departure from micromobility required (red). These categorizations are mostly selfexplanatory; however, “departure from micromobility required” refers to links where it is either illegal to ride a micromobility mode and a user must walk, or where paths require stair climbs to traverse. From this map developed, each of the link categorizations green, black, yellow, orange, and red were assigned cost values of 1, 2, 3, 4, and 5, respectively and entered into the matrix for pavement quality categorized. This same process for micromobility pavement quality categorized was then repeated for pavement quality categorizations for average pedestrians and decreased speed pedestrians. The pavement quality categorization can be seen below for average pedestrians and decreased speed pedestrians in Figures 5 and 6, respectively.. 16.

(23) Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 5: Average Pedestrian Pavement Quality Categorized within the Network. For pavement quality for average pedestrians as seen in Figure 5 above, only two categorizations were assigned: pedestrian prioritized paths (green) and neutral road paths (black). The pedestrian prioritized categorization pertains to paths in which there is no motorized traffic and pedestrians have freedom of movement throughout the whole link, not limited to just sidewalks. The neutral road path categorization pertains to paths in which pedestrians are constrained to just sidewalks. From this map developed, the links categorized as green were given a cost of 1 and links categorized as black were given a cost of 2 for the matrix of average pedestrian pavement quality categorized.. 17.

(24) In the case of pavement quality categorizations for decreased speed pedestrians as seen in Figure 6 below, four categorizations were assigned for links within the network: neutral road path (black), gravel path (yellow), cobblestone path (orange), and stairs (red). For the links labeled as stairs in red, these links were removed from consideration for this user group. Utilizing this map developed, the links categorized were assigned costs of 1, 2, and 3, for neutral road path, gravel path, and cobblestone, respectively.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 6: Decreased Speed Pedestrian Pavement Quality Categorized within the Network. 18.

(25) Working with the elevation categorized parameter, a single map and matrix was created to be applied to each of the three user groups. First the network of links were categorized by elevation climb as seen below in Figure 7.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 7: Elevation Climb Categorized within the Network. Seen in the key in Figure 7, the links were labeled as either no elevation change (black), moderate elevation change (orange), or severe elevation change (red). Utilizing this map and knowledge of the elevation change from topographical maps and Google Street View, links were categorized directionally as the following: severe downhill, moderate downhill, no elevation change, moderate uphill, and severe uphill and assigned cost values of 1, 2, 3, 4, and 5, respectively. For example, link 117 to 119 would be assigned a severe downhill status while link 119 to 117 would be assigned a severe uphill status. To provide an example for moderate elevation change, link 65 to 64 would be assigned moderate uphill while link 64 to 65 would be assigned moderate downhill. In the case of no elevation change, the link in both directions were assigned no elevation change. These categorizations of elevation were done intuitively based on their appearance in Google Street View and not based off of a numerical method.. 19.

(26) 3.3. Travel Time Parameter. Formulation of travel times for each user group on each link was generated using the distance divided by a baseline speed for each user group multiplied by a series of coefficients corresponding to generalized congestion, pavement quality, and elevation climb as can be seen in equation 2 below. 𝑇𝑇𝑖𝑗𝑚 =. 𝑑𝑖𝑗 𝑠𝑚 ∗𝑐𝑖𝑗 ∗𝑘𝑖𝑗 ∗𝑝𝑖𝑗𝑚. (2). Where: i ≡ origin node ∈ {1 through 141} j ≡ destination node ∈ {1 through 141} m ≡ mode ∈ {micromobility, average pedestrian, decreased speed pedestrian} d ≡ distance s ≡ baseline speed c ≡ generalized congestion k ≡ elevation p ≡ pavement quality. First, a generalized speed was assumed for each of the user groups, as seen below in Table 2. Table 2: Baseline Speeds for Considered User Groups in Travel Time Equation User Group Micromobility Average Pedestrian Decreased Speed Pedestrian. Baseline Speed 5.5 m/s 1.5 m/s 1.2 m/s. These values for each of the user groups was assumed according to selected cyclist and pedestrian travel time research. In one research paper out of Montreal, average cyclist speeds ranged anywhere between 18.1 and 20.6 km/h (~ 5-5.72 m/s) (Strauss & Miranda-Moreno, 2017). In another research paper out of the UK, specific to designing and modeling pedestrian environments, researchers found that the average walking speed of a typical pedestrian to be 1.47 m/s, ranging from 1.55 m/s for people aged 16-25, and 1.47 m/s for people aged 26-50 (Willis, et. al, 2004). Additionally, an average speed of 1.16 m/s was found for pedestrians over the age of 64 (Willis, et. al, 2004). Through a study out of Germany on choices for wheelchair routing algorithms, researchers here compared the speed of two wheelchair routing algorithms known as OpenRouteService and Routino, and found that in the algorithms used for these routing services, speed for wheelchair routing was between 4 and 8 km/h (~1.1-2.2 m/s) (Tannert & Schöning, 2018). With these ranges noted, it was selected that micromobility users would have a baseline speed of 5.5 m/s; an average pedestrian would travel at 1.5 m/s; and a decreased speed pedestrian such as a disabled person, elderly person, or parents with prams and children would travel at 1.2 m/s.. 20.

(27) In order to account for links in which high volumes of pedestrians and micromobility users are traversing and consequently slowing the speed in which an individual user is able to travel, a map of high congestion links was created identifying this phenomenon. This can be seen below in Figure 8.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 8: Links with Heavy Non-Motorized Congestion within the Network. In order to identify links with heavy non-motorized congestion within the network, Strava’s Global Heatmap was consulted to view the links with highest foot or cyclist traffic (Strava, n.d.). The following links seen in red in Figure 8 were assumed to be high congestion areas that would subsequently slow the speed in which these user groups are able to travel. To formulate this as the generalized congestion coefficient seen in equation 2, the links seen in red were assigned with the value 0.9, slightly decreasing the overall speed when multiplied by the baseline speed for each mode provided. Links seen in black were assigned a value of 1, having no effect on the overall speed when multiplied by the baseline speed.. 21.

(28) Next a series of coefficient values were assigned for pavement quality in accordance with the pavement quality categorizations described previously in Figures 3, 4, and 5, for the user groups micromobility, average pedestrian, and decreased speed pedestrian, respectively. These coefficient values can be seen below in Table 3. Table 3: Pavement Quality Coefficients for Travel Time Equation User Group. Micromobility. Average Pedestrian Decreased Speed Pedestrian. Pavement Quality Categorization Departure from Micromobility Required Cobblestone Gravel Neutral Road Path Bike Lane Neutral Road Path Cobblestone Gravel Neutral Road Path. Pavement Quality Coefficient 0.27 0.5 0.9 1 1.1 1 0.5 0.9 1. Where pavement quality coefficients are less than 1, this causes a decrease in the overall speed when these links are traversed. Where pavement quality coefficients are greater than 1, this causes an increase in the overall speed when these links are traversed. Coefficients equal to 1 have no effect on the overall speed. Based on research on cyclist speed and travel time from Strauss & Miranda-Moreno, it was identified that cyclists traveling on cyclist-specific infrastructure had increased speeds as compared to those traveling on links without (Strauss & Miranda-Moreno, 2017). Based on this assumption, bike lanes resulted in a coefficient of 1.1, increasing the overall speed and decreasing the travel time for those specific links. It is assumed this effect would also be seen in other micromobility users as well. It is important to note that the “Departure from Micromobility Required” classification results in a coefficient of 0.27. This value was chosen because when multiplied by the baseline speed for micromobility, it results in an overall speed equal to that of an average pedestrian. For example, (5.5 m/s)*0.27 ≈ 1.5 m/s. Additionally, the coefficient for pavement quality for average pedestrians were all given a value of 1, as it can be assumed that this user group is able to traverse at relatively the same speed regardless of whether it is typical sidewalk or some other pedestrian-prioritized path. The last coefficient to define in equation 2 is related to the elevation climb categorizations described previously in Figure 7. These coefficient values for elevation climb were assigned the same regardless of mode, and can be seen on the next page in Table 4.. 22.

(29) Table 4: Elevation Climb Coefficients for Travel Time Equation Elevation Climb Categorization. Elevation Climb Coefficient. Severe Downhill Moderate Downhill No Elevation Change Moderate Uphill Severe Uphill. 1.5 1.1 1 0.9 0.5. As was the case with the pavement quality coefficients previously, where elevation climb coefficients are less than 1, this causes the overall speed to increase when these links are traversed. Where elevation climb coefficients are greater than 1, this causes an increase in the overall speed. The coefficient for no elevation change is assigned to 1 and therefore has no effect on the overall speed. These determinations for elevation climb coefficient values were largely made intuitively, based on an expected increase or decrease in effort made by users whether they were traveling uphill or downhill. With each of the coefficients sorted in equation 2, the travel times for each link, dependent on which user mode is selected, is then calculated, and stored in a matrix for the travel time parameter.. 23.

(30) 3.4. PM2.5 and PM10 Air Pollution Avoidance Parameter To assign PM2.5 and PM10 air pollution categorizations to relevant links, two sensors from the Project: Testbed Kungsgatan funded by Norrköpings fond för forskning och utveckling were used for collecting PM2.5 and PM10 data that was then stored inside of the Testbed Kungsgatan backend system database. These approximate locations can be seen as two red dots labeled as sensor 1 and sensor 2 below in Figure 9. The links seen in orange are characterized as main links while the links seen in yellow are side links for application of air quality categorization. Links seen in black are excluded from utilization in this parameter. The relevance of this link classification is described later in the formulation of air quality categorization.. Background Base map and data from OpenStreetMap and OpenStreetMap Foundation (OpenStreetMap & OpenStreetMap Contributors, n.d.). Figure 9: Relevant Links for Air Pollution Modeling within the Network. To utilize the measured concentrations from these sensors, a function in MATLAB was created to generate SQL queries to fetch the data. This function databasecall.m can be seen in Figure 2.2 as component 5. To connect to the database in which the PM2.5 and PM10 concentration data is stored, a virtual connection was made utilizing PuTTY. Within this function, an SQL query obtains minute by minute concentrations for both PM2.5 and PM10 for both sensors, from the current moment all the way back to 60 minutes prior. This results in four vectors of concentrations (2 separate vectors for PM2.5 at each location, and 2 24.

(31) separate vectors for PM10 at each location) that are then sent back to the main script file RouteCalc.m. With these vectors now available, another separate function airPollution.m takes these vectors of recorded air pollution values and creates air quality classifications for each of the relevant links. First, the average of each of these vectors is calculated to produce two PM2.5 and two PM10 concentration values. The values for PM2.5 and PM10 taken from sensor 1 are applied directly as the cost to the orange links 23-35, 35-46, and 46-65. The values for PM2.5 and PM10 taken from sensor 2 are applied directly as the cost to the orange links 65-78, and 77-78. Looking at the side links, the PM2.5 and PM10 values from sensor 1 are applied as half of the concentration to links 34-35, 35-36, 45-46, 46-47, 65-79, and 65-98. The PM2.5 and PM10 values from sensor 2 are applied as half the concentration to link 78-96. Regarding the side links and main links, the assumption was made that Kungsgatan is a main link with relatively homogenous features and therefore could ascertain the PM2.5 and PM10 concentration values directly. The side links coming off of these features on the other hand have less traffic and ascertained half the concentration values collected from the sensor, assuming that this air pollution would dissipate moving away from the main connection of Kungsgatan. Whether or not a link received concentration values from either sensor 1 or 2 was dependent on its relative proximity to that sensor. Now that these values for PM2.5 and PM10 have been assigned to the relevant main and side links in the vicinity, these values were then compared against the European Air Quality index categorizations for PM2.5 and PM10 established by the European Environment Agency. These categorizations according to the European Environment Agency can be seen below in Table 5. Table 5: PM2.5 and PM10 Air Quality Categorizations According to the European Environment Agency Air Quality. Good. Fair. Moderate. Poor. Very Poor. Extremely Poor. PM2.5 (µg/m3). 0-10. 10-20. 20-25. 25-50. 50-75. 75-800. PM10 (µg/m3). 0-20. 20-40. 40-50. 50-100. 100-150. 150-1200. (European Environment Agency, n.d.). Depending on the PM2.5 concentration value, each link was assigned a cost as either 1, 2, 3, 4, 5, or 6 for air quality categorizations good, fair, moderate, poor, very poor, and extremely poor, respectively. This cost assignment was also performed for PM10 according to these categorizations. With each link having a cost assignment based on these air quality categorizations, the final air quality parameter cost assignment was taken as whatever the max cost value was between PM2.5 and PM10 categorizations for each link. For example if link 35-46 was assigned a cost of 1 for. 25.

(32) PM2.5 but a 3 for PM10, the overall cost assignment for air pollution for that link would be given a 3. Upon this final step, each of the relevant main and side links seen in Figure 9 now have cost assignments between 1 and 6 for the air pollution parameter.. 3.5. Routing Algorithm. With each of the links in the network now corresponding to various cost assignments for the elevation climb categorized, pavement quality categorized, travel time, and air pollution avoidance parameters in the form of matrices, the first step in the routing algorithm takes these matrices and then normalizes them according to min-max normalization. This was conducted according to equation 3 below. 𝑥𝑛𝑜𝑟𝑚 = 𝑥. 𝑥−𝑥𝑚𝑖𝑛 𝑚𝑎𝑥 −𝑥𝑚𝑖𝑛. (3) (Đerek & Sikora, 2019). Where xnorm is the normalized value of a given value x in a matrix, xmin is the minimum value in that matrix, and xmax is the maximum value in that matrix. Utilizing this formula, all values within each matrix are normalized to values between 0 and 1. This normalization makes it possible for the different matrices to be combined in a common formula for finding a shortest path with added weights for each of the parameters. With normalized costs now applied to each link for every parameter included, the routing algorithm then implements route optimization for both shortest path and individualized userpreference of parameters. The optimization model used in this demonstrator is a Dijkstra’s shortest-path optimization problem as developed by Dimas Aryo in MATLAB File Exchange (Aryo, 2012). Generally, the MATLAB function takes an input of a cost matrix of links in the network, along with an origin and destination node, and returns and output of a total cost and nodes utilized along the route. The function can be seen below in equation 4. [costOfRoute vectorOfNodes] = Dijkstra(costMatrix, origin, destination) (4) Where: costOfRoute ≡ returned route cost vectorOfNodes ≡ returned route vector costMatrix ≡ matrix of cost values for each link origin ≡ starting node for the route destination ≡ ending node for the route. (Aryo, 2012). This function in (4) then seeks out the lowest cost route with the desired origin node, desired destination node, and a modified cost matrix used as input. The function (4) then returns a vector of nodes to traverse across, along with a total unitless cost for the route. The cost matrix implemented into this optimization function (4) is calculated as seen on the next page in equation 5.. 26.

(33) 𝑚𝑎𝑡 = (𝑝𝑎𝑣𝑒𝑁𝑜𝑟𝑚 ∗ 𝑝𝑎𝑣𝑒𝑅𝑎𝑡𝑒 + 𝑒𝑙𝑒𝑣𝑁𝑜𝑟𝑚 ∗ 𝑒𝑙𝑒𝑣𝑅𝑎𝑡𝑒 + 𝑎𝑖𝑟𝑁𝑜𝑟𝑚 ∗ 𝑎𝑖𝑟𝑅𝑎𝑡𝑒).∗ 𝑑𝑖𝑠𝑡𝑁𝑜𝑟𝑚 + 𝑇𝑇𝑁𝑜𝑟𝑚 ∗ 𝑇𝑇𝑅𝑎𝑡𝑒. (5). Where: mat ≡ modified cost matrix. paveNorm ≡ normalized pavement quality cost matrix elevNorm ≡ normalized elevation climb categorized cost matrix airNorm ≡ normalized air pollution avoidance cost matrix distNorm ≡ normalized distance cost matrix TTNorm ≡ normalized travel time cost matrix paveRate ≡ user prioritization for pavement quality elevRate ≡ user prioritization for elevation climb airRate ≡ user prioritization for air pollution avoidance TTRate ≡ user prioritization for minimal travel time. To formulate the modified cost matrix seen in equation 5, user prioritizations for each of the parameters were included as the variables paveRate, elevRate, airRate, and TTRate. These prioritization variables are assigned values 0-10, where 0 corresponds to no priority, 1 corresponds to very low priority, values 2-9 represent intermediate prioritization, and value 10 represents highest priority, all according to a user’s preference for each of these route characteristics. Additionally, it is worth noting that any prioritization of one variable does not take away or add to another, each are assigned independent of the others, as users can have many prioritizations for these parameters at the same time. It is worth noting that the parameters for pavement quality, elevation climb, and air pollution avoidance are all multiplied by the distance of the link. This is based on a method observed within the Literature Review in section 2.4 where a thesis project out of KTH amplified the weights of links by the distance the link traverses, to obtain realistic route choices (Plynning, 2016). Travel time is excluded from being multiplied by the normalized distance like the other parameters, as distance was already considered when calculating the travel times for each link. This new cost matrix mat represents the modified user-prioritized cost values for each link in the network. This matrix created along with an origin and destination node are then used as input into the shortest-path algorithm developed by Dimas Aryo as discussed previously. Additionally, a shortest path route is formulated for comparison, utilizing just a matrix of distances as the costs on each link in equation 4. As output of this algorithm, a vector of network nodes is created as the most optimal path, along with a total cost for the route. Utilizing this vector of network nodes, the demonstrator then extracts the coordinates of each node within the route and plots them on an OpenStreetMap map viewer utilizing MATLAB Mapping Toolbox. In addition to this user-prioritized route selection, a shortest path optimized route is also plotted for comparison. Moreover, the travel times for each route are calculated by summing the travel time on each link that the routes traverse. This is simply taken from the cost matrix for travel times prior to normalization.. 27.

(34) 4 Results When testing the demonstrator for its outputs with different parameter prioritizations, it is important that the generated results reflect the cost assignment on the different links for each of the parameters included. That is, if a specific parameter is prioritized, the modified route provided should differ from the shortest path route, considering the characteristics of each corresponding parameter on the link it traverses. Additionally, there should be a level of intuition in regard to the modified route choice, in the sense that even if one parameter is prioritized, it does not force the user to travel in a way that seems overly burdensome in comparison to the shortest path. In essence, the user-prioritized route should seem within a reasonable travel time as the original shortest-path. The following Figures 10.1 through 19 below, illustrate the differences between routes for userprioritized parameters and that of only shortest path. With three different modes available, utilizing four different parameter prioritizations included, ranging from values of 1-10, there are a large number of possible outcomes that could be explored to see the differences between the routes created for shortest path and user prioritization of preferences. Rather than exploring all of them within this report, the maximum user assignment of each parameter prioritization alone has been highlighted for each of the three different modes, along with a few routes that combine these parameters for micromobility users. The scenarios can be described as the following below: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.. Micromobility: maximum prioritization of elevation climb alone Average Pedestrian: maximum prioritization of elevation climb alone Decreased Speed Pedestrian: maximum prioritization of elevation climb alone Micromobility: maximum prioritization of pavement quality alone Average Pedestrian: maximum prioritization of pavement quality alone Decreased Speed Pedestrian: maximum prioritization of pavement quality alone Micromobility: maximum prioritization of air pollution avoidance alone Average Pedestrian: maximum prioritization of air pollution avoidance alone Decreased Speed Pedestrian: maximum prioritization of air pollution avoidance alone Micromobility: maximum prioritization of travel time alone Average Pedestrian: maximum prioritization of travel time alone Decreased Speed Pedestrian: maximum prioritization of travel time alone Micromobility: maximum prioritization of both pavement quality and elevation climb Micromobility: maximum prioritization of both pavement quality and air pollution avoidance Micromobility: maximum prioritization of both pavement quality and travel time Micromobility: maximum prioritization of both elevation climb and air pollution avoidance Micromobility: maximum prioritization of both air pollution avoidance and travel time Micromobility: maximum prioritization of both elevation climb and travel time. For each of these scenarios, the prioritization level of the parameter for pavement quality, elevation climb, air pollution avoidance, and travel time (i.e. paveRate, elevRate, airRate, TTRate), as well as the origin, destination, and mode (M = Micromobility, P = Average Pedestrian, WC = Decreased Speed Pedestrian) have been listed on the next page in Table 6.. 28.

No results found