Bidding Strategies for
Flexible Loads in
Air-Conditioned Buildings
CHRISTIAN UTAMA
Master of Science Thesis
KTH School of Industrial Engineering and Management Energy Technology: TRITA-ITM-EX 2020:567
Master of Science Thesis TRITA-ITM-EX 2020:567
Demand Side Flexibility and Bidding Strategies for Flexible Loads in Air-Conditioned Buildings
Christian Utama
Approved Examiner Supervisor
21 October 2020 Prof. Viktoria Martin Jagruti R. Thakur, PhD Sebastian Troitzsch
Commissioner Contact Person
Abstract
Demand-side flexibility (DSF) has been touted as a possible solution to the challenges in power system operation arising as a result of the increasing inter-mittent renewables penetration and the emergence of electric vehicles (EVs). In Singapore, where around 24 to 60% of electricity demand in buildings could be attributed to heating, ventilation, and air conditioning (HVAC) purposes, air-conditioned building present a potentially major flexibility resource which could be used to provide DSF and help accommodate these challenges.
This study aims to investigate the DSF potential of Singapore’s building stock and to explore how this potential could be realized through demand-side bidding. To this end, a building energy modeling tool with explicit model-ing of the relationship between occupant comfort and HVAC load, CoBMo, is central to the analysis. CoBMo allows optimal load scheduling to minimize overall electricity cost while maintaining occupant comfort through linear pro-gramming, which is used for the analysis of both topics. A simple price-based market clearing model is developed to evaluate demand-side bidding imple-mentation, for which a case study on a district in Singapore named Downtown Core is developed. Several scenarios with possible future utility-scale PV pen-etration in Singapore’s electricity system are explored, as well as a sensitivity analysis and a comparison between demand-side bidding with price-quantity pairs, centralized dispatch, and demand-side bidding with linear curves.
Sammanfattning
Efterfrågesidans flexibilitet (DSF) har framställts som en möjlig lösning på de utmaningar som drivs av kraftsystemet som uppstår till följd av den ökande intermittenta penetrationen av förnybara energikällor och framväxten av elfor-don. I Singapore, där cirka 24 till 60 % av elbehovet i byggnader kan hänföras till uppvärmning, ventilation och luftkonditionering (HVAC), utgör luftkondi-tionerad byggnad en potentiellt stor flexibilitetsresurs som kan användas för att tillhandahålla DSF och hjälp tillgodose dessa utmaningar.
Denna studie syftar till att undersöka DSF-potentialen i Singapores bygg-nadsbestånd och att undersöka hur denna potential kan realiseras genom bud-givning på efterfrågesidan. För detta ändamål är ett byggnadsenergimodelle-ringsverktyg med tydlig modellering av sambandet mellan passagerarnas kom-fort och HVAC-belastning, CoBMo, centralt i analysen. CoBMo möjliggör op-timal lastplanering för att minimera den totala elkostnaden samtidigt som pas-sagerarnas komfort bibehålls genom linjär programmering, som används för analys av båda ämnena. En enkel prisbaserad marknadsclearingsmodell har ut-vecklats för att utvärdera genomförandet av budgivning på efterfrågesidan, för vilken en fallstudie om ett distrikt i Singapore med namnet Downtown Core utvecklas. Flera scenarier med möjlig framtida PV-penetrering i nyckelskala i Singapores elsystem utforskas, liksom en känslighetsanalys och en jämfö-relse mellan budgivning på efterfrågesidan med prismängdspar, centraliserad leverans och budgivning på efterfrågesidan med linjära kurvor.
Acknowledgements
Starting off, I would like to express my gratitude to Jagruti Thakur and Sebas-tian Troitzsch, my two supervisors, for their unending support during the cre-ation of this thesis. From the beginning, Jagruti had been helpful on all fronts and I am especially thankful for her critical comments that helped shape the thesis into a logical, coherent, and complete work. Sebastian, as the creator of the main tools used, was extremely patient in walking me through the tools and helping me develop new features for my analysis. He also provided a lot of insights and angles to the analysis which I wouldn’t have thought of, all the while pushing the work to be one of valuable scientific contributions. Jagruti and Sebastian are as close as it gets to the ideal supervisors for me, provid-ing just the right amount of liberty to let me impose my vision on the thesis and steering me in the right direction. Both of them had a massive influence in helping me stay motivated despite the less-than-ideal situation of having to work from home for practically the entire duration of the thesis. I’ll always be indebted to them.
I would also like to thank my wife, Anne Maria, for always being there for me. Anne was always by my side through the entire process, restlessly cheering me up and on. She never lost faith in me even when I almost did. For that alone, I am eternally grateful. To my parents, brothers, sisters-in-law, nephews, and nieces, thank you. Also, thank you to all the professors and friends that I’ve met along the way during my studies at Instituto Superior Técnico and KTH Royal Institute of Technology. I’ll keep the memories that I’ve made with all of you for the rest of my life.
1 Introduction 1
1.1 Literature Review . . . 4
1.2 Objective and Research Question . . . 6
2 Background 8 2.1 Building Energy Modeling for DSF . . . 8
2.1.1 White-box models . . . 9
2.1.2 Black-box models . . . 9
2.1.3 Gray-box models . . . 10
2.2 Electricity Markets . . . 10
2.2.1 Market clearing process in the spot market . . . 11
2.2.2 Types of wholesale electricity markets . . . 13
2.2.3 Singapore’s electricity market . . . 16
2.3 Demand Side Flexibility . . . 20
2.3.1 Benefits of demand side flexibility . . . 21
2.3.2 Types of demand side flexibility . . . 23
2.3.3 Demand side flexibility in Singapore’s electricity market 25 2.4 Bidding Strategies . . . 26
2.4.1 Single GenCo optimization . . . 27
2.4.2 Game theory . . . 30
2.4.3 Agent-based . . . 31
3 Methodology 32 3.1 Building data acquisition . . . 33
3.2 Building use classification . . . 33
3.3 Building parameter sets derivation . . . 35
3.3.1 Approximation of building shapes and sizes . . . 36
3.3.2 Occupancy and appliances schedules . . . 37
3.3.3 Indoor comfort constraints . . . 38
3.3.4 Internal gain factors . . . 39
3.3.5 Construction material properties . . . 39
3.4 Load shifting potential quantification . . . 41
3.4.1 Building energy modeling with CoBMo . . . 41
3.4.2 Calculation of load shifting potential . . . 43
3.5 Bidding strategy formulation and evaluation . . . 44
3.5.1 Electricity price forecasting . . . 46
3.5.2 Bidding strategy formulation procedure . . . 50
3.5.3 Market clearing simulation . . . 52
4 Results and Discussion 61 4.1 Load shifting potential . . . 61
4.2 Bidding strategy evaluation . . . 63
4.2.1 Sensitivity analysis . . . 67
4.2.2 Centralized dispatch comparison . . . 69
4.2.3 Linear bid curves comparison . . . 71
4.3 Discussion on sustainability . . . 72
5 Conclusion 73
Bibliography 76
2.1 Example of electricity supply and demand offers . . . 13
2.2 Market clearing in centralized electricity markets [63]. MCP: market-clearing price, MCV: market-clearing volume . . . 14
2.3 Market clearing in decentralized electricity markets [63]. MCP: market-clearing price, MCV: market-clearing volume . . . 16
2.4 Schematic diagram of the NEMS [11] . . . 17
2.5 Example of an energy offer in Singapore’s wholesale market . 20 2.6 Markets and products accessible for DSF [61] . . . 21
2.7 Comparison of power system flexibility strategies. Adapted from [17], author’s analysis . . . 22
2.8 Bidding curves proposed in [8] . . . 30
3.1 Methodology workflow diagram . . . 32
3.2 Dataset joining process . . . 35
3.3 Illustration of building shape . . . 37
3.4 Occupancy rate in office buildings . . . 38
3.5 Appliances use rate in office buildings . . . 38
3.6 Map of the Downtown Core district with the buildings overlaid 45 3.7 Distribution of building types in Downtown Core . . . 45
3.8 Electricity price time series preprocessing steps . . . 46
3.9 Price forecast for the upcoming 48 time steps with price points marked . . . 49
3.10 Price forecast with rolling horizon update . . . 50
3.11 Derived supply curve in Singapore’s wholesale market com-pared to actual clearing prices . . . 53
3.12 10 GWp PV production profile for a typical day in Singapore . 56 3.13 10 GWp PV production profile for a typical day in Singapore with disturbances . . . 57
3.14 10 GWp PV production profile for a typical day in Singapore
with disturbances (per 3 time steps) . . . 58
3.15 10 GWp PV production profile for a typical day in Singapore with disturbances (per 4 time steps) . . . 58
4.1 Load shifting potential for all building types . . . 62
4.2 Downtown Core dispatch profile - Scenario 1 . . . 64
4.3 Downtown Core dispatch profile - Scenario 2 . . . 64
4.4 Downtown Core dispatch profile - Scenario 3 . . . 65
4.5 Sensitivity analysis on lower temperature limit . . . 67
4.6 Sensitivity analysis on frequency of price spikes . . . 68
4.7 Comparison of dispatch profiles - Demand-side bidding and centralized . . . 70
4.8 Comparison of clearing price profiles - Demand-side bidding and centralized . . . 70
4.9 Comparison of dispatch profiles - Block bids and linear curves 71 A.1 Occupancy rate in hospital buildings . . . 82
A.2 Appliances use rate in hospital buildings . . . 83
A.3 Occupancy rate in hotel buildings . . . 83
A.4 Appliances use rate in hotel buildings . . . 84
A.5 Occupancy rate in university buildings . . . 84
A.6 Appliances use rate in university buildings . . . 85
A.7 Occupancy rate in school buildings . . . 85
A.8 Appliances use rate in school buildings . . . 86
A.9 Occupancy rate in retail buildings . . . 86
A.10 Appliances use rate in retail buildings . . . 87
A.11 Occupancy rate in single residential buildings . . . 87
A.12 Appliances use rate in single residential buildings . . . 88
A.13 Occupancy rate in multi-residential buildings . . . 88
3.1 Datasets for building use classification . . . 33 3.2 Indoor comfort constraints for all building categories . . . 39 3.3 Internal gain factors for all building categories . . . 40 3.4 Construction material properties for residential buildings . . . 40 3.5 Construction material properties for non-residential buildings . 41 3.6 Summary of clearing prices and derived residual demand (1
January 2020) . . . 54 3.7 PV production estimation input parameters . . . 56
4.1 Distribution of cost reduction in buildings . . . 66
Introduction
In light of the global energy transition, increasing penetration of renewables and the emergence of electric vehicles (EVs) have threatened the security of electricity grid operation in ways never seen before. In contrast with the tra-ditional concept of electricity flowing in one way from large generators to consumers in the grid, the advent of distributed energy resources (DERs) has caused bi-directional power flows to emerge in modern grids, particularly at the distribution level. Rooftop photovoltaic (PV) installations are a common-place nowadays, with consumers producing at least part of the electricity they consume and therefore giving rise to the term prosumer. This has caused a paradigm shift in how the electricity grid should be operated, given the in-creasingly complex nature of the system. For example, electricity produced by wind and solar energy is inherently volatile in nature, which complicates the supply and demand balancing process. Furthermore, injection of electric-ity by prosumers at the distribution grid level might induce voltage issues if left uncontrolled. One concept which has been touted as a possible solution to alleviate these problems is demand side flexibility (DSF).
DSF strategies provide electricity producers, consumers, and system oper-ators altogether the opportunity to benefit from flexibility in energy consump-tion. In the case of the former, DSF provides an opportunity to reduce the degree of curtailment in intermittent renewable generation by shifting flexible loads to periods with high renewable production. Meanwhile, in the case of the latter, consumers are compensated for their flexibility through two mea-sures: price-based and incentive-based mechanisms [7] [20]. Price-based mechanism relies on time-dependent price signals to encourage consumers to shift their energy consumption behaviors. On the other hand, incentive-based mechanism relates to the provision of monetary incentives by external
ties (e.g. utilities or electricity retailers) in exchange for load reductions from the consumer side. Direct load control is an example of incentive-based DSF implementation, which involves consumers ceding control of some of their appliances to a third party so that the latter could disconnect the appliances during critical periods [7]. At the same time, distribution system operators (DSOs) could also benefit from DSF by avoiding large supply/demand peaks which might induce instabilities in the power system. Despite all this, efforts are still needed to quantify the true potential of DSF strategies and to optimize how they are deployed.
In tropical countries such as Singapore, air-conditioned buildings present an interesting opportunity for demand side flexibility, seeing as heating, ven-tilation, and air conditioning (HVAC) systems represent a significant fraction of electricity consumption in buildings. According to a survey conducted by Singapore’s National Environment Agency (NEA), 24% of an average house-hold’s electricity consumption could be attributed to air-conditioning [39]. When it comes to non-residential buildings, this figure climbs up to 60% [24]. A study conducted by Sekhar [47] found substantial evidence that overcooling in tropical buildings is prevalent and it is suggested that this is due to non-optimal design and operation of the HVAC system. All of this shows that there is definitely potential in applying DSF strategies in air-conditioned buildings to better plan their electricity consumption – and in turn reduce the strain on the distribution grid - without sacrificing the occupants’ thermal comfort at the same time.
network costs, market support services fee, and power system operation and market administration fees [14]. According to EMA, 40% of households have decided to switch to their retailer of choice as of 31 August 2019 and have enjoyed savings of 20 to 30% compared to the regulated tariff [13]. A study conducted by Loi and Jindal [35] found that, taking into account the influ-ence of oil prices and other volatile components, wholesale electricity prices decreased by 9.11% from 2014 to 2017, which could partly be attributed to recent retail liberalization efforts.
Singapore’s wholesale market is a power pool consisting of half-hourly auction periods for energy, regulation, and reserve. For each of these periods, generators submit their supply offers and transactions are settled by the Mar-ket Clearing Engine (MCE) based on the overall least-cost dispatch schedule subject to forecasted demand and system and security constraints, resulting in the market clearing price (MCP) and the capacities allocated to the genera-tors [11]. In 2016, EMA introduced a demand response (DR) program with the objectives of enhancing competition in the wholesale market and improv-ing overall system efficiency, which could in turn result in downward pressure on electricity prices. A contestable consumer who could offer a load reduc-tion of at least 0.1 MW is eligible to participate in the DR program, either directly as a licensed load provider or through electricity retailers or licensed load providers. Consumers bid for curtailments in a similar way as to how generators bid for supplies in the electricity market, i.e. bids are submitted as price-quantity pairs. Incentives to participate in the program are calculated as one-third of the additional consumer surplus generated from the load reduc-tions and are paid out to consumers who are selected to be curtailed during certain market clearing periods [12].
under consideration.
This work serves as a continuation of the previous works done at TUM-CREATE, making use of existing frameworks developed for building model-ing based on model predictive control (MPC) and distribution grid with flexi-ble resources modeling. A case study based on an actual district in Singapore is developed to demonstrate how Singapore’s buildings could participate in the electricity market through demand-side bidding and to what extent they could influence the market prices.
1.1
Literature Review
One of the enablers of DSF has been the advent of smart grid. There is no uni-versally accepted definition of the term "smart grid", although it is commonly agreed that the general purpose of a smart grid is to improve the efficiency, reliability, quality, and safety of the power system operation. Modernizing the power grid becomes especially important in the era of ever-increasing re-newables penetration, as it increases the likelihood of potential imbalances between generation and consumption. Farhangi [19] presented a comparison of the features of the smart grid as opposed to those of the existing grid, as well as the basic ingredients of the smart grid, of which communication and data management hold a crucial role. Thakur and Chakraborty [53] discussed the benefits, risks, and challenges involved in smart grid and its importance for the generation of sustainable and reliable energy. Hence, smart meters which allow two-way communication between utilities and consumers are seen as one of the key components of implementing demand DSF in the smart grid. The rollout of smart meters, however, brings about the potential issue of bad practices in digitization and consumer data management, which is discussed by Ray and Pinson [45]. Schappert and Hauff [46] related smart grids with the notion of sustainable consumption and found that smart grids are highly linked to energy consumption in buildings.
[52] looked into the implementation of direct load control in centralized air-conditioning systems of commercial buildings to respond to urgent requests from the grid. Their proposed strategy was demonstrated to reduce peak de-mand by about 23% while maintaining the building temperature within the ac-ceptable limits. Meanwhile, Zhang et al. [65] modeled the dynamic behavior of a large population of air-conditioning loads and developed a control strategy to enable them to provide demand response services. The results of the study show that through the proposed aggregated control strategy, air-conditioning loads are able to perform peak load shaving and provide frequency regulation service.
As DSF is considered as a key component of future power systems opera-tion, studies on the integration of DSF into electricity markets are becoming increasingly important. Thakur and Chakraborty [54] analyzed the potential of DSF strategies for residential consumers and the obtained results show that DSF shows potential for load balancing as well as financial savings. An in-vestigation of the current DSF program as implemented in Singapore’s whole-sale electricity market was conducted by Zhou et al. [66]. The authors built a simplified market clearing model which demonstrates how transactions in the wholesale market are settled with the incorporation of the DSF program. They argued that in its current form, the DSF program limits the amount of flexibility that consumers could offer and as such, the full benefits of DSF might not be realized. A more recent study by Bai, Thoung, and Alvina [2] explored several alternative strategies to the current implementation of the DSF program: time of use pricing (ToU), real-time pricing (RTP), and direct load control. Nu-merical analysis were used to show the potential consumer benefits stemming from these strategies, although the results gave no indication or recommenda-tion as to which would be preferred. Considering the current restricted form of Singapore’s DSF program, it would be interesting to further investigate the other extreme, i.e. full demand-side bidding with all consumers being exposed to wholesale prices.
Shi, and Qu [32]. The authors classified optimal bidding strategy formula-tion models for generators into three categories: single generaformula-tion company (GenCo) optimization, game theory, and agent-based models. In the litera-ture, there are plenty of studies which explore optimal demand-side bidding strategy formulation. Fleten and Pettersen [21] developed a stochastic lin-ear programming model to formulate the optimal bidding strategy for a price-taking retailer in Nord Pool, the Nordic power exchange. The objective of the model is to minimize the expected cost of purchasing electricity from the day-ahead and real-time balancing markets. The model proposed by Herranz et al. [27] shares the same objective for a retailer in the Spanish electricity market, although genetic algorithm was employed to determine the optimal parameters which correspond to the best purchasing strategy. Meanwhile, Ge et al. [22] proposed a scenario-based optimal bidding strategy model based on linear programming for a retailer in a market with demand response program in place. Therefore, the retailer aims to minimize the electricity procurement cost while also considering the provision of demand response service to the power system operator. As per the author’s knowledge, there are only a few studies which explicitly model the relationship between consumer comfort and electricity consumption.
Based on the identified need of integrating DSF into electricity markets op-eration, this thesis aims to investigate how air-conditioning loads in buildings could provide DSF through demand-side bidding in the context of Singapore. To this end, social welfare of the electricity market will be a subject of analysis.
1.2
Objective and Research Question
The objective of this thesis is to investigate the DSF potential of Singapore’s building stock and to formulate bidding strategies for flexible building loads in a liberalized energy market. Specifically, it attempts to address the following research questions:
• How do we quantify the DSF potential of buildings in Singapore?
• How do buildings with flexible loads form their bids in the electricity market?
• Would DSF in the form of demand-side bidding affect social welfare in the electricity market?
Background
In this chapter, background information on topics relevant to this thesis is discussed. Therefore, the subsequent sections are dedicated to presenting a general overview of building energy modeling, electricity markets, demand side flexibility, and bidding strategies with a view to integrate these topics and model demand side flexibility in buildings with active participation in liberal-ized electricity markets.
2.1
Building Energy Modeling for DSF
Buildings around the world consume a considerable amount of energy, ac-counting for roughly 51% of the global electricity consumption in 2018 [58]. A significant part of the electricity consumption in buildings could be at-tributed to heating, ventilation, and air-conditioning (HVAC) purposes, rang-ing from 24 to 60 % in Srang-ingapore’s case [39] [24]. A number of studies have shown that much of what is consumed by HVAC systems is actually wasted due to equipment or operation problems, resulting in poor indoor comfort for the occupants [47] [57]. This highlights the importance of developing build-ing energy control and operation strategies which would help to increase the energy efficiency and indoor comfort in the buildings. To serve this purpose, building energy modeling tools have been developed and studied extensively in the literature. More recently, these tools have been integrated in model predictive control (MPC) applications for buildings which aim to develop pre-dictive control strategies for building energy management [40]. MPC allows the HVAC control problem to be expressed as a numerical optimization prob-lem with the aim of minimizing the energy cost while respecting the occupant comfort constraints, e.g. air temperature and indoor air quality limits. At the
same time, this also enables the explicit modeling of building comfort as a function of electric load - which is crucial for demand response. Depend-ing on the type of optimization problem desired (e.g. convex or non-convex), building model equations have to be put in an appropriate format before they could be used in MPC applications. Existing building energy modeling tools could be classified into white-box, black-box, and gray-box models. They are discussed briefly in the subsequent sections.
2.1.1
White-box models
White-box models rely on physical equations to model building components and its sub-systems. Since they are based on well-defined laws of physics, their main advantage lies in their potential to accurately capture building dy-namics. However, white-box models are typically time-consuming to develop and solve [33]. This is mainly due to the need of having detailed and accurate building specification information as inputs, which might not be available for every case. Typical inputs for white-box models include weather data, build-ing structure, construction material properties, occupancy data, and properties of sub-systems. These inputs are plugged into a set of mathematical equations which calculate the estimated building energy consumption. Some of the most well-known white-box models for building energy modeling include Energy-Plus, TRNSYS, and ESP-r. However, none of these tools could readily be used for MPC applications relying on convex optimization, instead requiring non-convex optimization algorithms such as the exhaustive search method.
2.1.2
Black-box models
the model parameters. Machine learning algorithms such as linear regression, random forest, and artificial neural network have been used as the basis of black-box building energy models.
2.1.3
Gray-box models
The gray-box approach to building energy modeling combines the white-box and black-box approaches, which is why gray-box models are also called hy-brid models. As the name suggests, gray-box models aim to get the best of both worlds by using both physics equations and measurement data to get to the end result. In most cases, simplified physical models which only utilize several parameters are employed, and these are not prescribed but rather de-termined through data-driven methods in order to yield the best fit to the op-eration data. One of the shortcomings of the gray-box modeling approach is the computational effort required in the parameter fitting stage. Many exist-ing gray-box buildexist-ing-energy models for MPC are based on the resistance and capacitance (RC) network model, such as the ones found in [4] [25] [51].
2.2
Electricity Markets
In modern societies, electricity is considered as an essential service. Tradi-tionally, electricity is generated in large power plants located far from the con-sumption points, transmitted over long distances using transmission lines, and finally distributed to the end users - the electricity consumers. In the last few decades, there has been a small but significant change to this structure due to the rise of distributed electricity generation such as from rooftop photovoltaic (PV) installations, allowing consumers to generate electricity on their own and possibly injecting the excess of what they need into the grid. Nevertheless, it does not change the fact that there are typically three main activities in the provision of electricity:
• Generation: conversion of primary energy sources (e.g. coal, natural gas) or renewable energy sources (e.g. solar irradiation, wind) to elec-trical energy
• Transmission: transport of electricity from generating facilities to dis-tribution grids
While transmission and distribution refer to the same activity in principle, the two terms are used to differentiate the voltage level at which electricity is being transported, with that of the former being higher than the latter. In the past, all these activities are conducted by a single company, usually referred to as an electric utility company. Such companies are usually owned and operated by national governments but they could also be public companies [34]. Over the last two decades, steps towards restructuring the electricity sector have been taken and at the moment, most electricity markets in developed countries are liberalized or deregulated markets. The core idea behind the liberalization of electricity markets is to create a competitive market which should theoretically result in price reductions, as only the most efficient companies will remain in the market. Out of the three activities mentioned above however, only genera-tion is regarded as appropriate to be exposed to competigenera-tion. This is because transmission and distribution activities are considered to be natural monopo-lies. A natural monopoly occurs when it is more economical to have one single company supplying the whole market as opposed to multiple companies com-peting against each other, due to the marginal cost of production of the good in question being negligible compared to the investment costs [31]. Therefore, the liberalization of electricity markets brought about the competitive market for electricity generation: the wholesale electricity market. Most of the trans-actions in the wholesale electricity market happens in the day-ahead market, sometimes also called the spot market [1] and this is the main focus of the information presented subsequently in this report. In the upcoming sections, details on the wholesale electricity market procedures and classification are provided.
2.2.1
Market clearing process in the spot market
In the day-ahead or spot market, transactions are fixed at least 24 hours before their actual delivery. Regardless of the market type, the quantity and price of electricity to be traded are determined through some sort of market clearing process. In this process, generators (and sometimes large consumers) submit their bids (offers) to an external party, who will then determine the optimal dispatch schedule from this information. The optimization of the dispatch schedule has the objective of maximizing the market’s social welfare, which is described as the combined well-being of producers and consumers in a par-ticular market.
and total revenues, while the latter is usually measured by the difference be-tween the consumers’ willingness to pay and what they are actually paying. In practice, social welfare maximization in a market is achieved by setting the market-clearing quantity and price based on the intersection between the sup-ply and demand curves, called the equilibrium. The reason for this is that at the equilibrium, the consumers’ lowest willingness to pay is equal to the marginal cost of producing the last unit of product, which represents an efficient con-dition for the market [31]. To better illustrate these concepts, a simplified example is presented below.
Figure 2.1: Example of electricity supply and demand offers
2.2.2
Types of wholesale electricity markets
Depending on the level of coordination between the market participants and the system operator, electricity markets could be categorized as either a cen-tralized or decencen-tralized market. The main aspects of both types of electricity market and their respective advantages and disadvantages are presented in the subsequent sections.
Centralized markets
markets, also commonly referred to as power pools. Examples of power pool include the New England, PJM, Midwest, New York, and California markets in the US [1].
Figure 2.2: Market clearing in centralized electricity markets [63]. MCP: market-clearing price, MCV: market-clearing volume
Decentralized markets
In decentralized markets, producers are encouraged to do self-commitment
1
and eventually submit price-quantity pair bids to the market operator. In this way, producers submit single price bids, which presumably cover all their internal costs. This is in contrast with the bid structure encountered in central-ized markets presented previously, where detailed cost breakdown is needed. The market operator and the system operator are usually separate entities, with the latter being considered "only" as a provider of transmission services. Hence, the system operator is not directly involved in the market clearing pro-cess and is only required to intervene when security constraints in the system are violated. Participation in the market is usually voluntary in decentralized markets. In this way, producers are consumers are allowed to conduct trades through bilateral contracts, so long as they notify the system operator and pay the necessary charges for grid use. As opposed to nodal pricing, decentralized markets usually employ zonal pricing, according to which electricity prices are uniform inside each zone. This means inside a zone, participants can freely trade between each other in secondary markets such as the intra-day market. Most decentralized electricity markets employ some sort of two-sided auc-tion, with both producers and consumers required to submit bids. For the latter group, it is often the case that small consumers are not allowed to partic-ipate directly but are instead represented by retailers. In some cases, large con-sumers (e.g. industrial concon-sumers) are allowed to submit bids on the wholesale electricity market. The price formation mechanism in decentralized electricity markets, sometimes also called power exchanges, is presented in Figure 2.3. Modern European electricity markets are typical examples of power exchange, e.g. Nord Pool, European Energy Exchange (EEX) etc. [63]
1
Figure 2.3: Market clearing in decentralized electricity markets [63]. MCP: market-clearing price, MCV: market-clearing volume
One advantage the decentralized market design holds over its centralized counterpart could be attributed to its flexible nature, which suits new and emerging technologies such as energy storage and demand response [1]. As described above, decentralized markets allow participants to trade in secondary markets in a straightforward manner, making it easier for them to adjust their dispatch schedules based on additional price signals from these markets. On the other hand, drawbacks of the decentralized design include the increased burden on producers to determine their optimal dispatch and potentially higher transaction costs for the producers as a result of increased intra-day trading fre-quency [1]. Moreover, Ahlqvist, Holmberg, and Tangeras [1] also argued that decentralized markets tend to have overly large zones, leading to intra-zonal constraints being ignored and unnecessarily increasing the amount of real-time adjustments needed in the system.
2.2.3
Singapore’s electricity market
with the center area representing an overlap between the two markets. The arrows shown in the figure indicate financial flows for the supply of electricity and related services between the entities. Note that the figure might be slightly outdated with regard to the retail market as it was most recently updated in 2010.
Figure 2.4: Schematic diagram of the NEMS [11]
The key stakeholders in the NEMS are:
• The regulator: the Energy Market Authority (EMA) holds the responsi-bility to ensure that the NEMS meets Singapore’s needs.
• The market operator: the Energy Market Company (EMC) operates and administers the wholesale market.
• The power system operator (PSO): Singapore’s PSO is a division of the EMA and is responsible for ensuring the reliability of electricity supply and secure operation of the power system .
• The transmission system operator (TSO): SP PowerAssets owns and manages the operation and maintenance of Singapore’s electricity trans-mission system, both at high voltage (transtrans-mission) and low voltage (dis-tribution) levels.
• Market support services licensees (MSSLs): SP Services Ltd is the sole MSSL in Singapore’s electricity market at present and they provide market support services such as meter reading and meter data manage-ment. In addition to that, they also supply electricity for non-contestable consumers and assist contestable consumers and retailers to access the wholesale market.
• Retail electricity licensees: retailers sell electricity to contestable con-sumers and they could either be market participant retailers (MPRs) or non-market participant retailers (NMPRs). The former buy electricity directly from the wholesale market while the latter obtain their electric-ity from the MSSL.
• Consumers: previously, consumers could be classified as either con-testable or non-concon-testable 2. Contestable consumers could purchase their electricity from a retailer, directly from the wholesale market (sub-ject to licensing and registration by EMA and EMC), or indirectly from the wholesale market through the MSSL. Meanwhile, non-contestable consumers are supplied by the MSSL. Following the completion of the nationwide Open Electricity Market (OEM) rollout in May 2019, all consumers in Singapore’s electricity market now are contestable, al-though they still have the option to continue buying electricity from the MSSL [13].
The following section is dedicated to a detailed description of the whole-sale market.
Structure of the wholesale electricity market
In reality, the wholesale market consists of two markets:
1. The spot market for energy, regulation, and reserve
2. The procurement market for other ancillary services
The spot market is an electricity spot market in the traditional sense, where buyers and sellers trade energy, regulation, and reserve via the EMC. On the other hand, the procurement market is used by the EMC to procure ancillary
2
services outside reserve and regulation which are necessary to ensure the se-cure operation of the power system, including reliability must-run, black start, and reactive power.
Transactions in the spot market are settled through auctions. For a given day, there are 48 half-hourly bidding periods and for each of these periods, the market prices for energy, reserve, and regulation are determined alongside the respective quantities allocated to each generation facility. The spot market is a power pool with one-sided auction, where only generators are required to submit their supply offers. On the demand side, load forecasts are prepared by the EMC based on the information received from the PSO. These are subse-quently fed to the market clearing engine (MCE) to determine the overall least-cost dispatch schedule and market prices. The MCE employs nodal pricing, i.e. a single price is assigned to each node or bus in the transmission network, which reflects the transmission losses and the transmission network’s physical limitations. Generators are paid the nodal price of the node they are assigned to, while buyers pay the weighted-average of the nodal prices at all off-take nodes, referred to as the Uniform Singapore Energy Price (USEP).
Figure 2.5: Example of an energy offer in Singapore’s wholesale market
2.3
Demand Side Flexibility
Figure 2.6: Markets and products accessible for DSF [61]
As the above figure shows, DSF could participate in various aspects of the power system, such as relieving technical constraints, increasing system adequacy, optimizing the operations of balance responsible parties (BRPs), and providing ancillary services. Therefore, it could be seen that DSF offers an alternative to the traditional concept of maintaining grid stability through supply-side flexibility or "generation following demand" and shifting the op-erating mode to "demand following generation". This is particularly relevant for modern power systems with high levels of intermittent renewables penetra-tion, as a significant part of the generation mix is non-dispatchable, i.e. unable to ramp up/down at will. The following sections present the perceived benefits of DSF and how DSF strategies are typically classified.
2.3.1
Benefits of demand side flexibility
Figure 2.7: Comparison of power system flexibility strategies. Adapted from [17], author’s analysis
According to Shen et al. [48], there are three main sources of economic benefits stemming from the DSF deployment. First of all, DSF allows for re-duction in peak demands, which rarely occur but nevertheless have sizable economic impacts. A study by [48] found that in many systems, at least 10% of costs are incurred as a result of peak demands occurring less than 1% of the time. During peak demands, significant jumps in electricity prices might be observed as a result of peaking power plants, whose marginal costs are on the higher end of the price spectrum, being called into action. In these events, DSF acts as a direct substitute to peaking units by reducing the demand and preventing them from being dispatched, therefore allowing the investments in these peak units to be deferred. Secondly, DSF could also provide ancil-lary services, which are typically procured from the producer side. These services are often provided by generating units operating under sub-optimal efficiency. Replacing them with demand reductions would absolve the power system from resorting to such inefficient solutions and improve the overall sys-tem efficiency. Last but not least, transmission and distribution losses could be reduced through DSF. The electrical losses on a transmission or distribution line is proportional to the square of the current flowing through it, which is in turn linearly proportional to the amount of power being transported. Trans-mission losses usually amount to 5 to 10% and could get even higher during times of peak loads, which is significant by all standards. DSF makes it possi-ble to reduce the power being transported, therefore reducing the load on the lines and at the same time, the losses.
renewable generation periods, thus reducing renewable energy curtailment, and moving the corresponding load to off-peak period, during which electric-ity generation is usually less carbon-intensive. On top of that, by doing away with peaking power plants as mentioned previously, greenhouse gas emissions from these typically fossil-fuel-based plants could be avoided, resulting in a cleaner environment.
2.3.2
Types of demand side flexibility
Based on the driving force for participation, DSF strategies fall into one of the following two categories: implicit and explicit DSF. Both categories are briefly explained and compared with one another in the following sections.
Implicit demand side flexibility
Implicit demand side flexibility, also called price-based demand side flexibil-ity, relies on the consumers adjusting their electricity consumption in response to price changes. Since it relies on price signals, implicit DSF is only appli-cable to consumers who are exposed to variable electricity rates. The U.S. DOE [56] identified three options for the implementation of implicit DSF pro-grams: time of use (ToU) rates, real-time pricing (RTP), and critical peak pricing (CPP).
Under ToU rates, variable electricity rates are encountered depending on the time of day. There are usually several periods with distinct prices, e.g. peak, mid-peak, and off-peak hours and these prices are always known by the consumers beforehand. While this type of pricing scheme has been widely applied to commercial and industrial consumers worldwide, most residential consumers are still paying flat rates for their electricity, which limits the ap-plicability of implicit DSF to the latter group.
CPP provides a middle ground between ToU and RTP. With the imple-mentation of CPP, electricity rates are defined in a similar way to ToU rates under normal conditions. During reliability-related events or when whole-sale electricity prices are exceptionally high however, the rates are replaced by higher prices. In order for CPP programs to be effective, consumers have to be informed of such events well in advance so that they could reschedule their consumption appropriately [36].
Explicit demand side flexibility
Explicit demand side flexibility is often referred to as incentive-based demand side flexibility and it encourages active participation in electricity markets from consumers. In this type of programs, consumers are given incentive pay-ments for their willingness to be curtailed at certain times, which are paid separately from their electricity bills. These incentives are usually calculated based on the electricity price reductions caused by load reductions, as in the case of Singapore’s DR program [12]. Usually, explicit DSF programs set a minimum amount of load reduction which can be offered (e.g. 100 kW), as lesser reductions are unlikely to significantly affect electricity prices. For this reason, small residential consumers typically are not able to participate in incentive-based programs directly, but rather through their utilities, retailers, or another type of intermediary known as DR aggregators. Explicit DSF is ar-guably more complex compared to its implicit counterpart, as it requires a set of supporting rules and regulations to ensure that the participants are not gam-ing the system by artificially increasgam-ing their baseline consumption. Accord-ing to U.S. DOE [56], most incentive-based programs fall into one of the fol-lowing five subcategories: direct load control (DLC), interruptible/curtailable service, emergency DR, capacity market, and ancillary service market.
DLC programs rely on the consumers ceasing the control of some of their appliances to their utilities or aggregators, the latter being able to turn the appliances on/off at will. Payments for these programs commonly consist of fixed monthly payments and additional payments when load reductions are triggered. DLC programs are mainly directed towards residential and small commercial consumers, partly due to its simplicity and low cost of implemen-tation [56]. For instance, residential water heaters could be a candidate for DLC participation, as explored by Jones et al. [30].
consump-tion and in return for the load reducconsump-tions provided, incentive payments in the form of discounts or electricity bill credits are given. Heavy penalties may be imposed for non-compliance to the load reduction signal [56].
Emergency DR programs rely on the concept of using DSF to get around fault events which jeopardize the power system’s reliability. Payments for emergency DR programs are based on the consumer’s outage costs or com-monly referred to as the value of lost load (VOLL). Participation in emer-gency DR programs are usually voluntary, with the participants only getting paid when they respond to load reduction signals and therefore, no penalties are in place in case of no response [56].
Capacity market programs encourage DR resources to participate in capac-ity markets in the place of conventional generators. In contrast with generators in the capacity markets which are remunerated based on their availability to produce at any given time, consumers are compensated for their availability to reduce load. Separate payments are also due in the case of actual load re-ductions. Similar to interruptible service programs, failure to comply to load reduction signals provided may impose significant penalties, due to the fact that participants are paid for their readiness to curtail their consumption [56]. Participation in ancillary services markets is a novel application for DSF. Ancillary services provided through DSF are equivalent to those supplied from the generation side, i.e. flexible generators with high ramp-up/down capa-bility. Depending on the type of ancillary service, technical requirements vary and may require precise set-point following capability from the loads for frequency-regulating services [56].
2.3.3
Demand side flexibility in Singapore’s electricity
market
The Demand Response (DR) program of the NEMS was established in 2016. According to EMA, the DR program was intended to bring about market wide benefits as follows:
• Improvement of the overall efficiency of the market and enhanced price discovery process by allowing contestable consumers to bid in the whole-sale electricity market through demand-side bidding.
• Improvement of system reliability during supply shortage.
Singapore’s DR program is neither purely explicit nor implicit, as it combines demand-side bidding and an incentive-based compensation scheme.
Demand-side bidding allows consumers to either bid their loads directly as a licensed load provider or bid through relevant third-parties (retailers or licensed load providers) to be scheduled in the energy market, similar to how generators bid their supply offers. Similar to how supply offers are structured, demand-side bids are required to have up to 10 price-quantity pairs and each tranche should offer a load curtailment of at least 0.1 MW. A price floor is put in place to prevent potential gaming behavior, i.e. licensed load providers increasing the baseline load to create artificial demand for load reductions. Only demand bids above the price floor will be accepted by the MCE and it is set at 1.5 times the current Balance Vesting Price (BVP), implying that load curtailments are only dispatched during half-hour periods of extremely high electricity prices called DR events.
The compensation scheme for participation in the DR program is based on the consumer surplus sharing mechanism. Consumers or licensed load providers who are selected to be curtailed will be paid one-third of the ad-ditional consumer surplus generated from the load reductions. In the case of a DR event, the MCE will be run twice - with and without the load curtailments. The former results in the actual market clearing price to be paid by participants on the demand side, while the latter generates the "counterfactual" price which would have applied if the DR program had not been in place. The difference between these two prices is then used to calculate the additional consumer surplus.
Since its implementation, participation in the DR program has been lim-ited. Up to 2019, only two DR bids have been dispatched in the span of three years. According to EMA, this could be attributed to a combination of the current low USEP and the relatively high price floor. In 2018, only roughly 2.35% of the total number half-hourly bidding periods qualified for DR bids, i.e. prices above the floor price were observed [15]. This indicates that there is arguably a need for a more lenient price floor level or alternative anti-gaming measures to increase participation in the DR program and realize its potential.
2.4
Bidding Strategies
prod-uct. In keeping with the perfect competition assumption, electricity generators should also bid at their marginal costs in a perfectly competitive power pool or power exchange - zero in the case of non-dispatchable renewables such as solar and wind energy. However, the assumption might not hold for real elec-tricity markets. Although the purpose of elecelec-tricity markets deregulation is to promote competition between market participants and incentivize operational efficiency, it might be the case that some parties have sufficient market power to influence the market prices and gain additional profit by bidding strategi-cally. From a supplier’s point of view, this is done by bidding at prices above the marginal costs in an effort to increase the market-clearing price and gen-erate more profit. Similarly, from the demand perspective it is also possible for retailers or large consumers to bid at prices below the levels that they are actually willing to pay for, provided that there is enough reason to do so.
However, the benefits of strategic bidding in electricity markets are not only limited to participants with sufficient market power. Even if both pro-ducers and consumers are assumed to be price-takers, they would benefit from organizing their bids in a way that maximizes their profits, subject to their expectations regarding price and capacity realization in the market. Supply-side bidding strategies have been studied extensively in the literature, whereas their demand-side counterparts are encountered less frequently, simply owing to the fact that the former have been around for longer than the latter. There exist different strategic bidding models for electricity suppliers and according to several authors [32] [43], they could be classified as follows:
1. Single generation company (GenCo) optimization models
2. Game theory models
3. Agent-based models
The first category is described and discussed in detail in the following section and short descriptions of the other two categories are provided afterwards.
2.4.1
Single GenCo optimization
all the variables have fixed values, whereas the latter considers that at least some of the variables could vary with certain probabilities. Many mathemati-cal programming methods have been used to formulate optimal bidding strate-gies in single GenCo optimization models, including mixed-integer linear pro-gramming (MILP), multi-objective linear propro-gramming (MOLP), non-linear programming (NLP), dynamic programming (DP), newsvendor, and Markov decision process (MDP) [32]. As shown by Conejo and Prieto [9], many vari-ations of the single GenCo optimization problem could be modeled as large-scale MILP problems. An example of such application is given below.
Conejo, Nogales, and Arroyo [8] formulated the bidding problem of a price-taker power producer in a wholesale market. In the study, the hourly market-clearing prices are assumed to be highly uncertain but their probabil-ity densprobabil-ity functions could be estimated using a time-series forecasting proce-dure. It is assumed that the market-clearing price at hour t is a random variable denoted by λt, whose predicted value and estimated standard deviation are λestt
and σtest, respectively. It could be shown that λtapproximately follows a
log-normal distribution. Within this distribution, upper and lower bounds on the value of λtcould be defined according to
λt,upper = λestt + atσtest (2.1)
λt,lower = λestt − btσtest (2.2)
where atand btare calculated according to the desired degree of confidence,
e.g. 95% or 99%.
The self-scheduling problem of the price-taker generator could be formu-lated as the following profit maximization problem
max pt T X t=1 λestt pt− ct s.t. pt ∈ Π (2.3)
where ptrepresents the electricity sold at hour t and should be within the
feasi-ble operating region of the generating facility as dictated by its minimum and maximum power outputs, ramp-up and ramp-down rates, and minimum up and down times. Meanwhile, ctdenotes the operating cost at hour t. The optimal
solution to the problem, denoted by p∗t, represents the best possible production
It is assumed that the bidding curves of the generator need to have only up to two price-quantity pairs. For each hour t, the corresponding bidding curve is a function of the optimal production p∗t as follows:
a) If p∗t = 0, bid a single block of power ¯P at price λt,upper
b) If 0 < p∗t < ¯P , bid two blocks of power p∗t andP − p¯ ∗t at prices λt,lower
and λt,upper, respectively
c) If p∗t = ¯P , bid a single block of power ¯P at price λt,lower
Figure 2.8: Bidding curves proposed in [8]
2.4.2
Game theory
2.4.3
Agent-based
Methodology
In this chapter, the methodology used to derive the analysis presented in this thesis is described. To maintain the accessibility of the data used and the repro-ducibility of the analysis conducted, open access data and tools were utilized in the process whenever possible. All data processing was conducted in Python unless stated otherwise. Figure 3.1 illustrates the overall workflow and each item is described in detail in the upcoming sections. The first three sections of the methodology cover the test case development, which is subsequently used in the analysis of two main topics: load shifting potential quantification and demand-side bidding implementation.
Figure 3.1: Methodology workflow diagram
3.1
Building data acquisition
As the first step towards developing the test case, the geometries of all build-ings in Singapore were sought. These were obtained from OpenStreetMap, whose data is available under the Open Database License, with the help of the Overpass API [41]. The Python script containing the query sent to the API could be found in the author’s GitHub repository [62] and the output is a geospatial data stored in GeoJSON format. The buildings are represented by polygons in latitude and longitude axes in accordance with the EPSG:4326 coordinate reference system (CRS).
3.2
Building use classification
To classify the buildings based on their main usage, a number of additional datasets were acquired. A summary of these datasets is presented in Table 3.1.
Table 3.1: Datasets for building use classification
Name Provider Source
Master Plan 2014 Land Use Urban Redevelopment Author-ity (URA)
[59]
Voluntary Disclosed Building Energy Performance Data
Building and Construction Au-thority (BCA)
[6]
HDB Property Information Housing Development Board (HDB)
[28]
longitude, thereby it has to be projected into the EPSG:4326 CRS before it could be used in conjunction with the building data obtained in the previous section. This projection is carried out with GeoPandas’ internal functions. Ultimately, similar to the building data, the land parcels are represented by polygons in latitude and longitude axes, with a single use associated with each parcel.
As for the remaining datasets, additional data processing is necessary. The second dataset contains information regarding all commercial buildings in Sin-gapore, while the third addresses government-owned public housing. They are provided as CSV files and by default, they do not contain any geospatial infor-mation. Therefore, the first thing to do is to find the coordinates (i.e. latitude and longitude) of each building in these two datasets before they could be integrated into the initial building data. This was done with the help of the OneMap API, Singapore’s national map which is developed and maintained by the Singapore Land Authority [49]. Each building’s address forms the main part of each query sent to the API and the latter sends the latitude and longitude values corresponding to the address in return. Subsequently, each building in these additional datasets could be matched with the ones in the initial building data.
1. Single residential 2. Multi-residential 3. University 4. School 5. Retail 6. Office 7. Hotel 8. Hospital 9. Industrial 10. Other
Figure 3.2: Dataset joining process
3.3
Building parameter sets derivation
3.3.1
Approximation of building shapes and sizes
Buildings vary greatly in terms of size and shape. For the sake of simplifica-tion, in this thesis the buildings are assumed to be homogeneous shape-wise, with the floor plan taking a cuboid shape as depicted in Figure 3.3. The four sides of a certain cuboid represent the walls of the corresponding building and are denoted by s1 and s2, which are expressed in meter. For every building, these parameters could be calculated as a function of the building’s floor area A in m2
and perimeter P in m according to
s1 · s2 = A (3.1)
2(s1 + s2) = P (3.2)
The above equations represent a system of two linear equations with two un-knowns which could be solved in a straightforward manner. The principal motivation behind this approach is that by maintaining the buildings’ floor area and perimeter, the floor area to perimeter ratio (A/P ) of each building is kept constant and in turn, it enables more accurate approximation of the build-ings’ heat transfer characteristics. It is further assumed that s1 > s2 and walls which measure s1 are oriented towards north and south, while walls measur-ing s2 face east and west as indicated in the figure. The height of a particular building h is calculated based on the levels or f loorarea information obtained in the previous section, which describe either the number of floors present in the corresponding building (Nf) or its gross floor area (GF A) in m2. In the latter case, the number of floors present is calculated according to
Nf = Round(
GF A
A ) (3.3)
Finally, h is obtained by simply multiplying the number of floors with the typical floor height, taken to be 3.6 m [60]:
Figure 3.3: Illustration of building shape
3.3.2
Occupancy and appliances schedules
Figure 3.4: Occupancy rate in office buildings
Figure 3.5: Appliances use rate in office buildings
3.3.3
Indoor comfort constraints
Singapore’s building assessment method. Since these limits are imposed to ensure occupant comfort, they are closely related to the occupancy schedules presented in the previous section. It would be logical to suggest that only when there are enough occupants in the buildings will effort be made to accommo-date occupant comfort. Hence, here it is assumed that the limits are only active when the occupancy rate is above 0.3. Conversely, when the occupancy rate is below the minimum threshold, temperatures are allowed to vary between 20 and 35◦C whereas there is no minimum fresh air flow rate enforced.
Table 3.2: Indoor comfort constraints for all building categories
Category Temperature (
◦
C) Min. fresh air flow rate (L/m2) Source Min. Max. Single residential 23 26 0 [10] [50] Multi-residential 23 26 0 [10] [50] University 23 25 0.6 [5] [50] School 23 25 2.8 [5] [50] Retail 23 25 1.1 [5] [50] Office 23 25 0.6 [5] [50] Hotel 23 25 0.6 [5] [50] Hospital 23 25 0.6 [5] [50] Industrial 23 25 0.3 [5] [50] Other 23 25 1.1 [5] [50]
3.3.4
Internal gain factors
Internal gain represents the heat emitted within an internal space by people, lights, electronic equipment, etc. As such, it is obvious that internal gain af-fects a building’s heating or cooling load. Internal gain factors in the buildings are typically expressed as a function of the gross floor area, i.e. in W/m2. The values used in this thesis are also obtained from CEA and in keeping with the categorization presented in occupancy and appliances schedules, separate val-ues for occupancy and appliances internal gain factors are also derived. Table 3.3 presents a summary of the obtained values.
3.3.5
Construction material properties
Table 3.3: Internal gain factors for all building categories Category Occupancy (W/m2) Appliances (W/m2) Single residential 1.2 7 Multi-residential 2 7 University 3.5 12 School 7.5 12 Retail 6.5 20 Office 4.5 12 Hotel 3.5 10 Hospital 3.5 18 Industrial 6 50 Other 6.5 20
properly approximated. For simplicity, it is assumed that all buildings belong-ing to the same use category are constructed with the same material. Further simplification was made to classify buildings into two categories: residen-tial and non-residenresiden-tial, the former consisting of single and multi-residenresiden-tial buildings and the latter encapsulating the rest. This was done in order to enable the derivation of construction material properties from Green Mark standards, from which most of the necessary properties were obtained. For properties which are not covered by Green Mark standards, additional data was sought from CEA. Table 3.4 and 3.5 summarize the construction material properties derived from both sources for the two building categories.
-Table 3.5: Construction material properties for non-residential buildings Com- po-nent Heat capacity (kJ/m2.K) Thermal resistance (m2.K/W) Ab- sorp-tivity Em- misiv-ity Window-to-wall ratio Wall 1360 1.43 0.85 0.94 0.4 Roof 1360 1.25 0.85 0.94 -Floor 1800 0.34 0.6 0.95 - Win-dow 0 0.36 0.4 0.89
-3.4
Load shifting potential quantification
To quantify the load shifting potential of the various building types, benchmark buildings were defined for every building type. They are assumed to be of the same size - their gross floor area being 10,000 m2 and spread over 20 floors. This means that the floor plan area is set to be 500 m2, with s1 and s2 fixed at 25 and 20 m, respectively and the height of the buildings h taken as 72 m. While these specifications make sense for some of the building types (multi-residential, university, retail, office, hotel), they might be considered as being too large for the other types (single residential, hospital, school, industrial). Therefore, the results obtained should be interpreted as the upper limit of the buildings’ load shifting potential. In practice, these buildings are represented in CoBMo and an optimization problem is formulated to estimate their load shifting potential. This is described in the upcoming sections.
3.4.1
Building energy modeling with CoBMo
Building models in CoBMo are represented as vectors in state space form which contain the differential equations for the zone temperature and surfaces in the buildings, namely:
x =[Tz]T, [Ts]T
T
(3.5)
u = [[ ˙Qgen,heatz ], [ ˙Qgen,coolz ], [ ˙Vzahu,heat], [ ˙Vzahu,cool],
[ ˙Vztu,heat], [ ˙Vztu,cool]]T (3.6)
v =Tamb, Tsky, [ ˙qirrd ]T, [ ˙qzocc]TT (3.7)
y =h[Tz]T, [ ˙Vzahu]
T, Phvac,eliT
where the vectors x, u, v, and y are the state, control, disturbance, and out-put vectors. By applying zero-order hold discretization to these vectors, the thermal building model could be transformed to its time-discrete form, given by:
xt+1 = Axt+ Buut+ Bvvt (3.9)
yt= Cxt+ Duut+ Dvvt (3.10)
A, C are the state and output matrices, while Bu, Du and Bv, Dv are the input and feed-through matrices on the control and disturbance vectors, respectively. More detailed information as to how this state space model is formulated could be found in CoBMo’s official documentation [55].
In the simplest case, CoBMo could be used to determine the optimal elec-tricity consumption schedule of a building given its state space model by for-mulating it as an optimization problem as follows:
min X t∈T Pthvac,elct∆t s.t. 3.9, 3.10 ∀t ∈ T, Tz,tmin ≤ Tz,t≤ Tz,tmax∀z ∈ Z, t ∈ T, ˙ Vz,tmin ≤ ˙Vz,tahu ∀z ∈ Z, t ∈ T, (3.11)
where T is the set of time steps in the planning horizon and Z is the set of all zones in the building. The objective function of the above problem is to mini-mize the total electricity cost consumed for HVAC purposes over the planning horizon T , with Pthvac,el and ctbeing the power consumed by HVAC
equip-ment and electricity price at time step t expressed in MW and $/MWh , re-spectively. ∆tdenotes the length of the time steps considered in the problem, expressed in hours (h). The first constraint refers to the state space model of the building as described above, while the last two constraints are indoor comfort constraints, namely temperature and minimum fresh air flow limits. At each time step, the temperature at every zone in the building Tz,tshould always fall between the lower (Tz,tmin) and upper limits (Tz,tmax) and the fresh air flow
sup-plied by the air handling unit to every zoneV˙z,tahushould be over the minimum
required (V˙z,tmin). The parameters of the state space model and comfort limits
are defined for various building types as described in Section 3.3. By varying the electricity prices ct, the optimal consumption levels according to the
cor-responding price level Pthvac,el(ct) could be determined, which represent the
3.4.2
Calculation of load shifting potential
Load shifting potential irrefutably varies with time, e.g. a building should offer less flexibility when it is fully occupied as opposed to when it is empty. To calculate the load shifting potential of the benchmark buildings at every half-hour interval over the course of a given day, the following setup is proposed:
1. Starting at the first half-hour interval (t = 1), the prices at all the other intervals are set to be an arbitrary value x
ct6=1 = x, (3.12) while at the interval in question the price is set to be the product of x and a factor of 1 + α
ct=1 = x(1 + α) (3.13) 2. The cost minimization problem is run with this configuration, resulting in the optimal consumption level for the corresponding time interval if the price were to increase by a factor of α, Pthvac,el(ct=1 = x(1 + α))
3. The optimization problem is re-run, but with the price at the interval being analyzed set to be
ct=1= x(1 − α), (3.14)
representing the case if the price were to decrease by a factor of α. Ac-cordingly, the result is the optimal consumption level at a lower price level, Pthvac,el(ct=1 = x(1 − α))
4. The process moves on to the next half-hour interval and the whole pro-cess is repeated until all intervals are exhausted
5. The load shifting potential at each interval is then taken to be
LSPt= Pthvac,el(ct = x(1 − α)) − Pthvac,el(ct= x(1 + α)), (3.15)
illustrating how much the load could possibly change if electricity prices vary by ±α.
