API

Another aspect of the model that was making it not very user-friendly is the fact anyone who wants to use the model for a site of his interest, has to firstly open the software for the meteorological data, download a .csv file from there, import the .csv into another Excel sheet, copy the data and paste them into the file of the Model and see the results. A way to make this process quicker was studied and an API was the right answer.

API stands for “Application Programming Interface” and it is a software intermediary that allows two applications to communicate with each other. With an API, the database of a software can be accessed directly from a python script, via internet. The major contribution of the API is that it allows the program to automatically access the meteorological data by just reading the latitude and longitude inputs.

In order to do so, a research online was made [83]. Firstly, the python libraries “json” and “requests” had to be downloaded and implemented on Jupiter. Then, a URL given by the software has to be found in its website. On this regard, the software SODA still does not offer an automatic access through API for Python nor for Excel, whereas PVGIS gives the possibility to utilize this service for Python free of charge. Next, through few lines of code, the script automatically access the radiation data, in TMY format, from the server with only the information of latitude and longitude, which will be the only input that the user will have to give, together with the other characteristics of the Power System. A visual explanation of how the process works is summed up in Figure 31

36 This is an important step forward for the model and for the department of Ericsson Site Solutions, not only for the easier approach to the costumers, but it also opens the possibility of building an App of the model that the users will be able to download on their smartphone.

Figure 31 - Visual working principle of the automatic access through API

3.3 Spectral Effect

From the research presented above, it appears clear that including spectral effect can make a significant difference when modelling large bandgap technologies like a-Si and Cd-Te, especially in tropical location. However, since amorphous silicon cell has high costs and a very low efficiency, around 10 % [84], and cadmium is harmful for the environment due to its toxicity [85], both a-Si and Cd-Te PV technologies are not considered by Ericsson for potential investments. On the other hand, CIGS technology is in the company interest and the research will focus on it. Figure 32 shows the flowchart of the methodology described in the following paragraph. The technologies of crystalline silicon and thin film CIGS will be compared on the base of two KPIs:

 The annual PV yield

 The annual self-sufficiency achieved by the system

37 Figure 32 - Flowchart of methodology

With reference to small bandgap technologies like CIGS and crystalline silicon, the annual variation of the spectral impact is usually lower than a-Si and Cd-Te. However, in mid and high latitude locations, there is a clear difference in the astronomic and weather conditions between the winter and summer seasons. In fact, one of the reasons why the yearly impact is lower, is because the gains during one season are offset by the losses during the other.

This difference happens to be beneficial for power applications, since it results in an increased output during the winter, a time when the production is very low, and in constant or lower output during the summer, when a PV

38 system is usually already overproducing. Moreover, if the average monthly spectral impact in December is 3.3%, it means that there will be days and hours in which the gains are higher and lower than that. In fact, considering that the spectral gains of small bandgap technologies are enhanced by larger values of AM, as emerged from the literature review, it follows that, during the morning and late afternoon hours, there will be higher gains than in the central hours and this would be valid for the whole year. This would be extremely helpful in making the output spread more evenly during the day, combating the tendency of solar production facilities to concentrate most of their output at midday, causing the price to drop and spike again in the evening, as explained by the duck curve mentioned in the introduction and making it challenging to store a large amount of energy in few hours and realising it during the whole night. Having a flatter output curve during the day would reduce the storage capacity requirements, which is one of the most significant cost, when designing a renewable energy system. This reasoning encouraged to dig deeper into the effect that different spectral distribution can have on CIGS technology-based hybrid power systems when a high time resolution study is performed, where the spectral effect is calculated for every hour of the year.

In the literature review, AM and cloudiness were identified as the parameters that affect the distribution of the radiation spectrum. While the AM is mentioned in several works, cloudiness appears in fewer research papers.

Moreover, quantitative data about the AM influence can be found, whereas the cloudiness effect is mostly described qualitatively, and its effect depends also on the thickness of the clouds. Besides, the AM is easy to be calculated and can be integrated into a model, since it is related to the angular position of the sun. On the other hand, the cloudiness is a parameter that is difficult to model and to forecast, in particular, it is very challenging to elaborate an algorithm that predicts the thickness of the clouds. In addition, from the Freiburg study, the monthly average values of the APE and the AM were reported and compared in a graph in Figure 33.

Figure 33 - Monthly average values of APE and AM in the city of Freiburg.

It can be seen how the two parameters, representing the AM, red graph, and the spectral distribution, blue graph, are related. When the AM increases, the APE decreases, resulting in a red-shifted spectrum. The two trends are almost specular and suggest that considering only the AM when studying the variation of the spectrum, neglecting the effect of the clouds, can be considered valid.

In order to apply this finding in a practical model, the instantaneous values of the spectral impact and their relation to the AM are needed. The paper in Freiburg measured the first but not explicitly the second. Figure 34, shows two scattered graphs about the MM for CIGS and high-eff c-Si calculated every 5-min throughout 3 years and a half. This information may be used to define the highest and lowest instantaneous values of spectral impact that can be typically reached. With reference to CIGS, the highest and lowest 75^th percentile MM values are 1.08 and 0.96. These two values are taken as upper and lower boundaries instead of the absolute maximum and minimum points in order to filter out any mistakes coming from errors in the measurements. Following the logic, the minimum MM will occur for the minimum AM value(AM=1), MM equal to 1 refers to the standard condition (AM=1.5) and the maximum MM value will be reached for an average high enough AM value in Freiburg, that

-1

39 means, the typical highest value reached at the first and the last hours of the winter days. Therefore, the AM values in Freiburg are needed.

Figure 34 - Spectral mismatch factor for CIGS (left) and high-eff c-Si (right) vs broadband irradance measured with a pyranometer.

Boxes indicate 25th, 50th and 75th percentile of all values. Dara are 5-min averages from the period 01.06.2010 to 31.12.2013.

[61]

In the Excel model, the AM for every hour in Freiburg can be calculated with the following equation, developed by Kusten and Young [86], as a function of only the zenith angle 𝜃 :

𝐴𝑀 = 1

𝑐𝑜𝑠𝜃 + 0,050572 ∗ (6.07995 + 90 − 𝜃 ) ^.

Table 8 shows the values of the hourly AM for three different days in the year, calculated with the Excel Tool, according to the equation . Since the maximum value of spectral impact should occur mostly during the earliest hour in winter, AM=10 was picked to be the third point needed to build the relation between MM and AM with a linear interpolation.

Table 8 - Hourly AM values for three days in Freiburg, calculated with the equation above.

AM

21-jan 21-jun 21-dec

05:00 - 7,04 -

06:00 - 3,36 -

07:00 - 2,19 -

08:00 13,68 1,66 28,36

09:00 5,10 1,37 6,76

10:00 3,40 1,21 4,11

11:00 2,80 1,12 3,29

12:00 2,64 1,10 3,08

13:00 2,80 1,12 3,29

14:00 3,40 1,21 4,11

15:00 5,10 1,37 6,76

16:00 13,68 1,66 28,36

17:00 - 2,19 -

18:00 - 3,36 -

19:00 - 7,04 -

40 However, the simple linear approximation did not return similar monthly average values to the ones in Table 1, column CIGS, therefore, a polynomial trend was built as a series of linear trends with different slope, to ease the computation, which produced monthly values with an acceptable deviation from the result in the case in Freiburg.

As illustrated in Figure 35, the spectral impact grows fist from -4 to 0 for AM values lower than the standard 1.5, then it keeps growing from 0 to 8 and stays constant for AM values higher than 10.

Figure 35 - Extrapolated relation between the spectral impact and AM for hourly calculation of SF

Then, this relation was added to the Model in Excel for the city of Freiburg and the spectral impact for every hour of the year was calculated as a function of the AM. In Table 9, the results of the hourly spectral impacts for the same three days as Table 8 are shown, together with the daily average. As expected, the maximum spectral gains are registered during the first and last hours of the winter days, whereas, the central hours of a summer days present spectral losses.

Table 9 - Hourly spectral impact (%) in Freiburg for the same three days as Table 1. The daily average values are calculated at the bottom and are in line with the findings of [42] as illustrated in Table 1

Hourly Spectral Impact (%)

41 Next, the average monthly spectral impact values obtained with the tool were compared to the values in the literature, see Table 1. In

Table 10, it can be seen that the average values follows well the paper findings during the winter season, with differences smaller than 10%, but overestimate the losses during the summer months, with deviations up to 60 %.

Besides, September and October show the highest differences, around 100%, highlighted in red in the “difference”

column. It appears clear that more insights about the instantaneous relation between AM and spectral impact are necessary to formulate a better relation between them all year around.

Table 10 - Comparison between the average monthly spectral impact (%) calculated from the Tool, following the relation in Figure 35, and the values from the literature [61], for CIGS, in Freiburg. September and October present the highest difference

Calculated Source difference (%)

2,65 2,6 2

The same method was applied to the case of high-eff c-Si. From Figure 34, graph on the right hand side, 0,98 and 1,10 were picked as the highest and lowest MM values and a linear relation between MM and AM was built and shown in Figure 36. An Am equal to 20 was taken as the highest point for the spectral gain because, in Figure 34, graph on the right hand side, the MM reaches values close to 1,10 only for very low irradiance power, below 100 W/m2

Figure 36 - Extrapolated relation between the spectral factor (%) and AM for high-efficiency c-Si

Then, the relation was applied in the Excel model and the average monthly values obtained were compared to the one in the paper. Table 11 illustrates how in this case, the linear approximation respond well for this type of technology. The monthly averages are very similar from October to February, deviating by just 20-30% from the value in the literature, while only June and July present a larger difference, highlighted in red in the “difference”

-2

42 column, but not as big as in the CIGS case. Besides, in this case, the output is underestimated, meaning that if this method was applied, it would not result in a better performance than the reality.

Table 11 - Calculated average monthly values and their comparison with the values in the literature for high-eff c-Si. [55]

Calculated Source difference

Once the relation between the AM and the spectral factor for CIGS and high-eff c-Si was validated by comparing it with the monthly average values found in the literature, seven locations of growing latitude were taken as a case study, in order to investigate how the inclusion of this phenomenon would affect the performance of the PV technologies of interest and how the effect is related to the latitude, since it is expected to register more significant variation at high latitudes. These locations were the same as the case in chapter 3.1.5.1:

 Maracaibo, Venezuela, 10,7 latitude north.

 New Delhi, India, 28,6 latitude north.

 Los Angeles CA, USA, 34,1 latitude north.

 Milan, Italy, 45,46 latitude north.

 Frankfurt, Germany, 50,12 latitude north.

 Copenhagen, Denmark, 55,69 latitude north.

 Stockholm, Sweden, 59,34 latitude north.

Then, the spectral effect relation was implemented in the tool for each of the location and a comparison was made for the case with and without spectral effect, considering the same type of energy system, in terms of load, type and number of PV panels and batteries. The system was made of 36 monocrystalline 455 Wp panels, 5 Li-ion batteries with a total storage capacity of 48 kWh and no wind turbines, pure solar mode was chosen, for a 1.8 kW constant load. As before, for Maracaibo and New Delhi, the two locations at the lowest latitudes, a PV array of just half the number of the panels was chosen, because the 36 array was too large and already covered 100%

of the yearly load, making impossible to understand if any improvements would be gained, when using a different method.

3.4 Optimal Tilt Angle

It could be useful to integrate an optimum tilt angle suggestion in the model, that the customer might choose to follow when designing its site, in order to understand which type of structure to purchase.

A typical structure for the panels array is the roof above the enclosure shown on the left of Figure 37. A vertical deployment above the antenna itself could also be applied, saving space on the ground and other foundations work related costs, shown on the right. The latter configuration would work better for locations where the best tilt angle is very high anyway

43 Figure 37 - a) roof structure for panels above an outdoor enclosure. b) Vertical deployment above the mast of the site However, in a real case, maximizing the yearly PV production might not be the optimal choice for the overall system. In fact, in case of an overproduction, maximizing the PV yield could result mostly in an increase in the excess of production, which might be still be useful when it is possible to sell that unutilized power to the grid but not in the case of stand-alone system. Therefore, in this chapter, an attempt to evaluate the optimal tilt angle is made for two cases. In the first case, the objective is to maximize the solar production, in the second, the objective is to maximize the self-sufficiency by reducing the hours of backup needed.

An extensive literature review was already performed in this report to address the first case. On top of that, the optimal tilt angles suggested by the software PVGIS were consulted and reported here. When the user investigates the performance of a PV system in PVGIS, instead of setting the tilt angle beforehand, he can choose to let the software pick the one that maximize the PV yield, based on the radiation for the location and year chosen. 19 locations at various latitudes were examined and the suggested tilt angle plotted in a scattered graph. The “Linest”

function incorporated in Excel was used to draw a linear trend of the values and to obtain an equation that could describe it.

Figure 38 - Suggested Optimal Tilt Angle that maximize the yearly PV production for several locations, according to PVGIS [81]

Next, the optimal tilt angles, both in terms of production and self-sufficiency, of the Ericsson hybrid system were manually searched for several different cases and locations, in order to see if a specific pattern could be found.

y = 0.5405x + 10.519 y = 0.5405x + 10.519

0 10 20 30 40 50 60

0 10 20 30 40 50 60 70 80

Tilt Angle

Latitude

Optimal Tilt Angle in PVGIS

44

3.5 Dispatch strategy

Depending on which components are present in the hybrid system and whether the threat of power outages is real, the model manage the sources in different ways. A general hierarchy list that sums up the priority among the resources is:

1. RES -> Load 2. Grid -> Load

3. BESS (until 20% SOC) -> Load 4. DG -> Load

5. BESS (until 0% SOC) -> Load

Figure 39 illustrates the storage logic in the general case in which the system is composed by Solar PV – Grid – DG – BESS.

Figure 39 - Battery storage logic

In the next two subchapters, two sites will be analysed for different penetration level of renewables and storage in terms of number of panels and batteries.

The KPIs to assess the performance of each scenario will be:

 the percentage of generation from renewables achieved,

 the yearly amount of emission reductions,

 the payback period and the cumulative cash flow value after 25 years, discounted by an interest rate of 5% to consider the time value of money.

The first site is a grid-connected with no Diesel generator nor power outages, lcoated in Dittenheim, Germany.

The second, is lan offgrid site located in Mexico, that relies on a Diesel Generator.