Solar photovoltaic (PV) has seen the most rapid growth among the renewable energy sources in the last decade. The market share of bifacial PV modules is expected to rise up to 70% in 2033. Such technology enables different farm configurations; however, it introduces complexities during the modelling phase, particularly concerning the estimation of incident irradiance on the rear side of the modules. The aim of this thesis is to investigate the potential of E/W vertical bifacial PV farms, in terms of market revenues, with respect to N/S tilted counterparts. A bifacial PV model is developed to estimate the power generated by a large-scale farm. This is based on the view factor concept and a 2D assumption is adopted to enable fast simulations. Such model takes into account the non-uniformity of the incident irradiance as well as the spectral impact. A multi-dimensional matrix approach is implemented to minimize the computational time while performing the calculations for individual cells and wavelength values. The model is validated in collaboration with the company KIPP&ZONEN, focusing on the broadband rear irradiance and its non-uniformity. Overall, the model shows sufficient agreement with the measured data, namely a mean bias deviation of −1.29W /m2 (−2.22%) and a RMSE of 12.65W /m2 (21.69%) are obtained. The validation highlights larger errors in case of higher tilt values, especially during the clear days and at the edge of the modules. Specifically, the error is proportional to the amount of unshaded ground seen by a cell, in alignment with the limitations of the view factor concept. The profitability of vertical modules is studied in relation with various variables, including design parameters, market conditions, curtailment strategies and hybrid vertical/tilted configurations. A global scale is achieved by extending the simulation to 102 locations worldwide. To calculate the revenues of the PV farms, electricity price curves are modelled considering lower noon prices, hence different market conditions are identified by the minimum price and the ratio between morning/evening and noon prices. Whether vertical or tilted configuration is favourable in terms of market revenues is not dependent on the curtailment strategy unless the maximum power is limited to 70% of the nominal value. Combining vertical and tilted modules within a PV farm is found to be advantageous only in case heavy curtailment is applied. Among the design parameters, the row-to-row distance has a higher impact on the market revenues with respect to the modules’ elevation. Specifically, larger distances are favourable for both configurations even though such benefit is more evident for vertical modules whereas optimal height values can be identified. A minimum ratio between morning/evening and noon prices is calculated, which represents the lower limit for the higher profitability of E/W vertical modules with respect to the N/S tilted case. Such value is dependent on the specific location and the design parameters. The locations characterized by a low diffuse fraction are recommended to implement vertical PV farms.

Bifacial modules are expected to have a 60% share in the PV market, which require spectral irradiance data incident on both sides to optimize the design. Sensors that provide spectrally resolved data are expensive and thus a cost-effective spectrally resolved albedometer was designed in two previous PVMD Master Theses. This albedometer measures irradiance in three wavelength bands. The focus of this research is on developing a model that reconstructs the spectral irradiance incident on the top part of the albedometer. The spectral simulation program SMARTS was converted to MATLAB, after which a Particle Swarm Opimization Algorithm was added to determine values for unknown parameters. The model was first tested against spectra generated by SMARTS. After which both the original model and an model with additional modification coefficients was tested with clear-sky measurements from the albedometer and validated against an EKO spectroradiometer. The original model proved to fail in the first wavelength band, which caused the other wavelength bands to be reconstructed poorly as well. The adjusted model performed close to the performance of the EKO for clear-sky irradiances above 400 𝑊/𝑚2 in the total albedometer region, 320-1100 nm.

In existing standards for photovoltaic (PV) plants, little guidance is provided in the spatial allocation of pyranometers. No universal and easily scalable algorithms exist that help designers of monitoring infrastructure in solar parks choose the most representative positions for irradiance sensors. To fill this gap, a software tool is created for determining the most representative sensor locations requiring only the PV plant's main characteristics, layout, and future data usage as input. If necessary, advice on the number of sensors can be given. The sky view factor (SVF) was determined for all relevant planes in the PV plant. This was done by adding all SVF contributions of Na altitude bands and NA azimuth slices. The horizon obstruction at each location was determined using 30m spatial resolution digital surface model (DSM) data from Sentinel Hub imagery service. Two SVF modelling variables were optimised by varying them in SVF calculations at 3800+ existing PV plant locations in Europe. NA,optimal=1080 and rmax,optimal=2000m were found. The Perez model stood out from a literature review of sky diffuse model comparison studies. Moreover, six decomposition models were compared using data from twelve European Baseline Surface Radiation Network (BSRN) weather stations. A seasonal bias was found and compensated in an attempt to improve the already best-performing BRL model. The average normalised root mean squared error (nRMSE) improved from 31.8% to 30.5%. The average absolute relative mean bias error (rMBE) improved from 11.2% to 10.6%. The software tool determines SVF maps for different orientations and tilts. Ground albedo time series is extracted from the NASAPOWER database, and historical global horizontal irradiance is imported from PV-GIS. Plane-of-array irradiance maps are constructed through transposition modelling. Subsequently, error maps are created, from which the location with minimal measurement deviations can be extracted. The software tool was tested for a case study in Eisleben (Germany) and Kolindros (Greece). The relative prevented measurement deviation (rPMD) was up to 1.2% in the Kolindros case, with an average of 0.8% compared to 0.3% in the Eisleben case. Instantaneous measurement deviations up to seven times the rPMD were seen. Furthermore, a simplified allocation algorithm only based on the SVF maps was found, only valid under the current assumptions.

The transition from a centralized to a decentralized energy infrastructure is one of the most discussed features of the future energy system. With the fast growth of renewable energy technologies, which can be integrated in the built environment and in contexts like small production centers, the development of distributed energy generation and storage systems closer to consumers is expected to play a significant role in driving the change. Within this context, the role of Energy Communities is emerging and is at the center of numerous studies. The Green Village in Delft is developing the 24/7 Energy Lab project, focusing on providing reliable, affordable and clean energy to a small-scale energy community by means of a system composed of solar panels for energy generation, batteries for electrical energy storage, and an hydrogen storage system consisting of electrolyzers, fuel cells and hydrogen tanks for seasonal energy storage. Previous research has highlighted how an off-grid configuration would result in inconveniently high costs for the community's users, if compared to the average cost of energy in The Netherlands. The aim of this thesis is to study the system in a grid-connected configuration, and in particular to find the optimal sizes of the components in order to achieve the best trade off between three conflicting objectives : minimizing total costs, maximizing self- sufficiency and maximizing reliability. After modeling the system's components and their mutual interactions, the optimization was carried out on MATLAB using a variant of the NSGA-II algorithm, which provides a Pareto Set of equally optimal solutions for the problem. The solutions were then ranked with a Technique for Order Preference based on Similarity to the Ideal Solution (TOPSIS), to assist the decision-making process. The simulations have determined that an installed capacity of 85.41 kWp (composed of 234 panels of 365 Wp each) results in the most effective choice for the solar energy generation, irrespective of the external conditions imposed. The optimal storage capacity, however, results significantly more influenced by factors such as grid imports limitations and price uncertainties. Under the conditions of limited imports from the grid, an optimal capacity of 75 kWh in the form of batteries was found. In general, the study confirms that the adoption of an hydrogen storage system is far from being convenient on a small scale residential level, regardless of the pricing conditions. The research has also posed an accent on the incremented costs incurred to reach full reliability of the system with low values of dependence from the grid, due to the high costs of the necessary storage equipment. Additionally, despite the best solutions found represent the optimal compromises balancing the conflicting objectives, reasonable solutions in terms of costs faced by the Community's users are usually not among the first choices of the ranking algorithm, mainly because they necessitate of at least 50% of the load to be supplied through grid imports.

The growing global energy demand and correlated rise in carbon emissions are increasing the need of renewable energy sources. This spread requires land to be occupied, competing with other activities such as agriculture and residency. This project can help the spread of photovoltaic (PV) technologies in an environment still little explored: the water. Offshore floating photovoltaic (OFPV) gains increasing attention in research due to a substantial reduction in land occupancy and a lower operating temperature. Therefore, this study aims to evaluate the power output of a OFPV system located in the North Sea considering the effect of the waves on tilt and azimuth of PV modules. First, the JONSWAP spectrum theory was employed to simulate the sea surface. Then, the interaction between the waves and the floater was modelled to obtain the orientation of the PV modules. The system energy yield was simulated through the PVMD Toolbox, a physics-based tool developed by the PhotoVoltaic Materials Devices Group (PVMD Group) at Delft University of Technology. Finally, experimental analysis was conducted on a commercial string inverter emulating the DC power output from the OFPV modelled plant. The waves generally cause lower irradiance hitting PV modules. However, fluctuations do not always have a negative influence. For example, the research found that with a calm sea, a system under the effect of waves produces 1% more than an offshore stationary 0° tilt plant. Nevertheless, the variable PV orientation scenario shows substantial losses for higher sea agitation states compared to the 0° tilt situation. Over a year, an OFPV under waves effect loses 0.84% and 17.97% of the production compared to stationary 0° and optimal installation tilt, respectively. From the laboratory activity, it appeared that with the same irradiance, the oscillations have a negative impact on the efficiency, especially that of the maximum power point tracking (MPPT) block, which reaches -3.2% with rough sea. In the end, we can conclude that the power output losses are not as dramatic as expected and that the development of OFPV technology will probably depend on future costs for offshore installations and the competition for inland areas.

Future energy systems are expected to rely increasingly more on distributed energy resources (DER). Prosumers that own energy resources such as photovoltaic (PV) generation and energy storage systems, will play a crucial role in the realization of DER’s in future power systems. Limited grid connections due to local electricity congestion make it more difficult for prospective industrially-sized prosumers to find suitable locations for their business operation. Energy resources such as batteries and PV, can be used to fulfil the desired electricity demands while adhering to the congestion related limits. By combining multiple energy resources and controlling them together the operational cost is drastically reduced. This thesis develops a stochastic optimization model that is able to find the optimal dispatch strategy for energy resources while adhering to constraints set by the DSO in a congested area. Also, a deterministic model is developed to be used as a comparison to the stochastic model. A deterministic model is simple and fast but it does not accurately represent the stochastic nature of PV generation, electricity consumption, and market prices. The stochastic model uses a set of different input scenarios for each time step and finds the optimal strategy considering all scenarios. In this thesis two models that consider stochasticity are developed: a stochastic and a robust model. For the robust model, the imposed grid constraints from the Distribution System Operator (DSO) have to be respected at each time step for each scenario and are classified as hard constraints. The model is able to find a optimal strategy for a week in each season when considering 11 scenarios. All predetermined constraints are respected for each simulation and while this makes the strategy reliable it also makes it conservative. For the second stochastic model, the DSO grid supply constraint is designed to be flexible and allow for occasional overshoot. In this case, flexibility is obtained by constraining the statistical distribution of grid power rather than the value at each time step and scenario. The calculated control strategy results in a simulated revenue increase between 3.67% and 4.78% depending on the season. For each season the objective cost reduces relatively more than the occurrence of grid overshoot. The term grid overshoot is used to indicate a situation where the DSO imposed grid limit is exceeded for a time step. The overshoot remains within the predetermined value threshold of 4% for all simulations of the stochastic optimization model.

Changes in social, cultural, and economic paradigms have resulted in a collective shift from rural to urban areas over the last few decades. Cities are expanding to accommodate the ever-increasing population influx in urban areas. New buildings, roads, airports, and harbors are only a small part of urban development. It also includes the use of new agricultural lands and changes to water resources surrounding urban areas. These new developments invariably introduce complex geometry onto surfaces as well as a plethora of new materials into a field, which leads to the change in local albedo. Albedo indicates how well a surface reflects solar energy, and it is defined as the ratio of total reflected irradiation from a surface to total incident irradiation on the surface. Albedo has a wide range of applications such as estimating the total heat content of an area and assessing the global warming potential. In this project, we used our Geometric Spectral Albedo model developed in the PVMD research group and open source LiDAR data (AHN1 to AHN4 collected over roughly 25 years) to investigate how urban development affects the local albedo.

The urgency of taking actions against climate crisis is unprecedented. The human-produced CO2 is the largest contributor to global warming, which drives the climate change. Many countries around the world have committed to net-zero greenhouse gas emissions in the next coming decades. Power-to-X (PtX) technologies in combination with carbon capture technologies might contribute to a carbon-neutral future, since they can convert electricity and captured carbon to synthetic gases (e.g. hydrogen, methane), chemicals (e.g. propylene, ethylene) and liquids (e.g. methanol). This thesis aims to research the techno-economic potential of a system that is able to produce methanol at industrial scale via a PtX scheme, which is coupled with carbon capture technologies. The system of this thesis consists mainly of PV panels, an alkaline water electrolyzer, a polymer electrolyte membrane (PEM) CO2 electrolyzer and a single-stage Lurgi quasi-isothermal methanol reactor. The feedstocks for this system are water and carbon dioxide (captured from the flue gases of a cement plant). The whole production process of this system has been designed and modelled in such a way, that both electrolyzers can follow the intermittent power supply from the PV panels. The electrolyzer's dynamic operation is controlled by a deterministic control logic, which takes into account the variable energy efficiencies of the electrolyzers and the intermittent power output of the PVs. Moreover, it has been decided that the methanol synthesis is based on the CO2 hydrogenation process, which converts syngas (i.e. a mixture of CO, CO2 and H2) into methanol with the use of the methanol reactor. Hydrogen is produced by the water electrolyzer and the captured CO2 is reduced to CO by the CO2 electrolyzer. As far as the sizing and production results are concerned, the system is able to produce 3.999 kT of methanol per year, consuming 2,147 tons of CO2 per year and requiring an energy input of 30.3 GWh/year. The installed peak PV power is equal to 18 MWp and 99.60% of their energy yield is exploited by the system (the rest is dumped). Due to the implemented control logic, the operating energy efficiency range for the CO2 electrolyzer is 45.07-55.32% and for the H2O electrolyzer is 75.83-81.71%. In terms of economic analysis, the proposed system requires a total capital investment (TCI) of 195.7 M€ and its operational expenditures are equal to almost 2 M€ per year. The performed cash flow analysis showed that the system has an annual gross profit of € 855,535 per year. Despite the yearly profit, it was found that the system's net present value (NPV) is negative and equal to -179.5 M€ at the end of its lifetime (i.e. 20 years). Also, the levelized cost of methanol (LCOM) was found to be equal to 4.44 €/kg, which is almost 10 times higher than the current market price of methanol. Therefore, such an investment would be economically unfeasible, despite its environmental benefit, because it would result in a net loss of capital.

The electrical output of a solar panel can be optimised in several ways, through the inclusion of bypass diodes (BPDs) or to have a maximum power point (MPP) tracker or DC optimiser to ensure that the panel is always operating at its global MPP. However most of these systems have to be implemented externally to the panel. A new system, the active switching circuit was proposed as a means to actively sense shading and switch between three wiring configurations drawing inspiration from matrix converter theory. The active switching circuit prototype would operate for 16 solar cells arranged in a 4 x 4-cell solar panel and be able to change the wiring configuration of these sixteen cells using switches. The base unit of the solar panel is a substring composed of four solar cells in series and the circuit would switch between three wiring configurations. The shading would be measured through four current sense resistors and amplifiers, with one of these pairs in each substring and a differential operational amplifier system to measure the voltage across the strings. The current and voltage measurement helps determine the power output of the panel. A Simulink model of the system was first created and a panel with 4 x 4 solar cell arrangement ordered and then measurements made with shading to verify if the shading model was accurate. From these results a Simulink model for the active switching circuit was designed to test the theory of its functionality and if such a circuit could be controlled to switch between the three wiring configurations. The circuit was successfully built and was able to operate in its intended way.

Solar photovoltaic (PV) technologies offer a renewable alternative to power generation that is based on fossil fuels; however, to meet the rising demand in electricity a substantial amount of surface area is required. This will inevitably lead to the installment of these systems on agricultural land, subsequently intensifying the land-use conflict and threatening food security. To alleviate this, the use of agrophotovoltaic (APV) systems is investigated, which enable the simultaneous production of food and renewable energy. More specifically, the aim of this thesis is to determine the optimal topology for a medium-scale and stationary bifacial APV array, which is simulated under the climate of Boston, USA (42.37˚N, 71.01˚W). Irradiance modelling is performed with Radiance’s raytracing algorithm in combination with the daylight coefficient approach of Daysim and the Perez All Weather sky model, which is then coupled to the crop and electrical yield models to determine the overall land productivity. The modelling approach used is robust, while offering flexible manipulation of the array’s deployment configuration, which is crucial for multivariable optimization. The integration of bifacial PV offers various synergistic effects mainly due to the amplified ground irradiance necessary for crop growth. Owing to the decreased PV density and high elevation from ground, rear irradiance homogeneity and bifacial gain are enhanced. Widening of the row spacing resulted in a logarithmic increase of the incident ground irradiation, while the overall energy yield portrayed a negative exponential trend, which is attributed to the use of bifacial modules. East-west (E-W) facing and vertically installed topologies are better suited for shade intolerant species, or permanent crops since they permit additional light penetration, especially during the winter months. In contrast, south-north (S-N) facing and latitude inclined arrays lead to intense and non-homogeneous shading that is unfavorable for growth during winter or for crops that cannot acclimate to shade; nonetheless, energy yield and land equivalent ratio (LER) are significantly enhanced. Unlike previous studies where only conventional modules were examined, here the potential of a customized one is inspected to assess whether blueberries can grow effectively under shade. By integrating such a module in an E-W hinged PV topology, which is associated with the most optimal shading patterns and schedule, the agrivoltaic performance is optimized. In comparison to the reference case, the land’s productivity is increased by 59% with a reduction in electrical yield by a third. Through this holistic approach that incorporates a multi-scale sensitivity analysis, it is possible to achieve a spherical understanding of the limitations and potential synergies associated with the dual use of land, ultimately encouraging the sustainability of the agricultural sector.

Implementing renewable energy generation and more sustainable consumption behaviour in the future inland shipping industry is necessary. Electrification of the inland shipping fleet leads to an increase in electrical energy demand on the consumption side. The generation of decentralized photovoltaic (PV) energy systems on board of general cargo vessels can be one solution to these challenges. To estimate the potential of these decentralized energy systems, accurate simulation models are needed. The objective of this research is to determine the photovoltaic potential of the Dutch general cargo inland shipping fleet. A method is developed to simulate the energy yield of moving general cargo vessels. The estimated energy yield is based on hourly power calculations. A difference is made between container and bulk vessels. This simulation model consists of a skyline, a vessel, an irradiance, a PV module temperature, and an energy model. In the skyline model, skyline profiles for 3036 waterway points are generated using LIDAR AHN3 height data. The waterways skylines are corrected for every vessel individually. In the vessel model, AIS (automatic identification system) data is used to simulate the sailing behaviour of 2746 vessels. In the irradiance model, the diffused, direct and ground reflection irradiance received by the PV panel is simulated. The diffused irradiance is calculated according to the Perez model. The PV module temperature is estimated according to the fluid-dynamic model. The additional air caused by the forward movement of the vessel and the water temperature is taken into account. Finally, the PV energy yield is calculated by the received irradiance, the suitable PV surface and the corrected PV module efficiency. An experiment is performed to validate the developed model, where a PV panel is installed on the vessel Harmonie. Equipment is installed to monitor Harmonie during one docking week and two sailing weeks. A co-variance linear regression model describes the relationship between the measured and the estimated PV power. According to the co-variance linear regression model, the P-value for the simulated PV power is below 0.05 and, therefore statistically significant. The outcome of the linear regression model is overestimated by 4 % with a 95 % confidence interval between 0.87 and 1.05. 740,958 PV panels can be installed on the general cargo fleet. Together, these panels have an installed peak power of 267 [MW] and an annual estimated PV potential of 230 [GWh]. The annual PV energy per unit area of a container vessel is 171 [kWh/m2 ] and for a bulk vessel 168 [kWh/m2 ]. The annual PV energy per installed power for a container vessel is 864 [Wh/Wp] and for a bulk vessel 852 [Wh/Wp]. When the outliers are removed from the annual specific power [Wh/Wp] dataset, the complete fleet can be modelled by a Weibull distribution. A t locationscale distribution is suggested when the outliers are not removed from the dataset. The average annual energy demand of a container vessel is 1440 [MWh], 7.17 % of this demand can be supplied by the PV panels installed on the vessel. Bulk vessels have an energy demand of 1350 [MWh] on average, of which the installed PV panels can cover 5.82 %.

Due to the much needed energy transition as agreed under the Paris climate agreement, it is expected that large amounts of variable renewable energy supply will be connected to the Dutch Transmission Network. Future social, economic and technological developments will determine both the amount of electrification as the central or decentral nature of the future supply of electricity. By investigating different reports with scenarios for 2030 and 2050, key uncertainties for the Extra High Voltage (EHV) network can be identified. As of now, no open source model of the Dutch EHV network is available for research. In this thesis a synthetic model of the Dutch EHV network is created from only publicly available data in DIgSILENT PowerFactory. The created synthetic model can be used as a tool to investigate future steady-state power flows considering the sensitivities of the topology, component parameters and different operational scenarios. Besides this, a method is created for generation of future scenarios in Python. Using this model and method, simulations were done to gain key insights into the Steady-State security of the future Dutch EHV Transmission Network. For the case study, it is investigated what the effect of future additions of variable renewable energy sources (VRES) and selected locations of 10 GW installed capacity of electrolysers would be on the Dutch EHV transmission network for 2030.

Solar photovoltaic cells are destined to become an important contributor in the renewable energy sector, also for small electrical systems inside buildings. They can be used to mobilise the electrical infrastructure, by providing an independent energy source for medium-consumption products, like decorative lighting. This thesis shows all aspects of designing and building an interior, solar powered lamp. First, the ideal position for indoor light harvesting has been investigated extensively, by recreating two typical Dutch rooms with 3D modelling software Blender. The light simulation tool RADIANCE was then used to compute the irradiance at various spots in the room, considering three standard sky types. Weather data from the KNMI was used to classify the sky conditions for every hour during two weeks in November, and the results were compared with pyranometer measurements, showing an error of 20% on a daily basis, and 5% over a five day period. The same simulation method was applied to predict the PV energy yield of four common solar cell technologies for a full year, for multiple room orientations and positions on a wall. Based on these results, three concepts were designed, corresponding to three specific room positions with different indoor light characteristics. Ultimately, one concept was chosen to be build as a prototype, with tailor-made, foil-to-foil laminated PV module, consisting of laser cut SunPower interdigitated back contact (IBC) cell technology. At standard test conditions, the measured short-circuit current was 2.37 A, with an open-circuit voltage of 21.5 V and a maximum power point of 35.9 Wp, resulting in an efficiency of 20.3%. Furthermore, a charge controller with maximum power point tracking algorithm was used to charge a 12V polymer lithium-ion battery pack. The combination of pyroelectric infrared (PIR) motion sensor detector and a light sensor module assures a conservative use of a 2.4 W strip of light emitting diodes (LED).

The growing demand and improvements in manufacturing capabilities, supported by government subsidies, has allowed the increase in the installed capacity of utility scale solar parks. Due to the remoteness in their location, the costs associated with dispatching personnel for maintenance is extremely high. A major contribution towards unscheduled downtime of these plants is due to the inverter faults. Currently, reactive and preventive maintenance are the most prevailing methods to identify and fix inverter faults. The presumption that the components will not under-performor fail until the scheduled visit, leads to a significant loss of production and revenue. To deal with the disadvantages of current maintenance methods, the solar industry is very keen on understanding the possibility of early detection of inverter faults by implementation of predictive maintenance. This research assessed the applicability of Machine Learning (ML) towards early signal detection of inverter faults in order to generate predictive maintenance alerts. The data for building the ML algorithms was acquired from a Shell owned 26.6 MWp utility scale solar park located in Moerdijk, The Netherlands. The early signal detection algorithmdeveloped, was based on the comparison between the actual and the predicted active power. The model built to predict the active power was based on two supervised learning methods;Elastic Net and Gradient BoostingMachine (GBM) with quantile regression. These models were capable of predicting the active power with a Mean Absolute Error (MAE) of 0.98kW & Root Mean Square Error (RMSE) of 1.8kW using Global Plane of Array irradiance (GPOA) and module temperaturemeasurements available from theMoerdijk data. The early signal detection relied on differentiating between prediction error and actual error. A window was created to encompass the maximum extent of prediction errors to avoid any false positive signals. This window for elastic net was found to be ¡¾ on the lower side and 2¾ on the upper side. Although when elastic net method was tested on 337 inverters- by looking at their residual variation 1-week prior to registered fault- it was found that the predictions suffered a periodic structural error. This was due to the erroneous predictions at times with extreme irradiance values. To mitigate, this the GBM with quantiles of 0.01 and 0.99 of GPOA was built to create a range of predictions giving rise to a wider range for normal operation. The results from both the algorithms indicated no early signals for inverter fault detection. This was partly due to data quality issues with fault tags in the Supervisory Control and Data Acquisition (SCADA) monitoring system; only 7 actual fault cases were identified. Additionally, the economic feasibility of implementing predictive maintenance was found to potentially reduce the current Operational Expenses (OPeX) by up to 10%. Despite the issues with data quality, an approach of using ML towards early fault detection for inverters in utility scale solar parks has been realised through this research.

By 2050, the global demand for food and energy is expected to grow by 70% and 50%, respectively, as a result of the increase of the world’s population. To keep up with the growing world population, significant changes have to take place in the agricultural sector regarding land and energy use. In this work, two developments in the agricultural sector are combined: the integration of photovoltaic (PV) modules in a horticultural system and the use of light emitting diodes (LEDs) as supplemental lighting source for crops. Four greenhouse systems having different configurations of these two technologies and a plant factory are designed. The performance of these LED-based agrivoltaic systems are analysed for the crops lettuce and tomato and for three different climates: Sweden, the Netherlands and the United Arab Emirates. This study shows that the integration of a PV system and LED technology in a greenhouse system improves the crop production per unit of energy compared to conventional greenhouse systems. Furthermore, it has been found that the production of crop per piece of land can be significantly increased by the use of a plant factory. In extreme climates, the plant factory is the optimal system for the cultivation of crops.

With emphasis on finding storage solutions for renewable based power plants, hydrogen has emerged as one of the prominent options. Hydrogen required in fertilizer, oil and gas industries etc., is produced using fossil fuels which emits carbon dioxide 10 times the produced hydrogen. It is important to produce the hydrogen from green energy sources for this industrial use or for the energy storage use. Solar energy harvested from Photo-Voltaic (PV) technology can be used to produce hydrogen in an alkaline electrolyzer. Directly coupling the PV and electrolyzer systems will have least components contributing to inefficiencies, complexities etc. In this study, a tool was made using MATLAB-Simulink, to optimize the PV-Electrolyzer directly coupled system. In literature many authors have done the same but none of them have included variation in space irradiance over PV farm, variation in PV module parameters due to manufacturing defects or defects arising over the period of use. These variations affect the IV curve of PV modules and in turn affect the performance of the whole system. It was observed that after optimizing, the coupling efficiency in the range of 90-95% could be achieved in directly coupled systems with shorter string lengths of PV. This was even after including the variation up to 20% in parameters like irradiance and PV module parameters. If the best configuration is not available in the market, then the set of next best configurations is available as output of simulation. Even with variation of 10% in input parameters, the maximum difference between the global maximum and other local maximums in the results is 3%. This 3% compromise results in energy loss of 870 kWh and 148 euros loss per year for a 50 kWp system. Which is 2.96 Euros/kWp loss, at 1 GW scale it may result into a loss of 2.96 million Euros, if we ignore the variations. In comparison to the DC-AC-DC configuration that comprises of inverter, rectifier and transformer, the directly coupled system performed better. The efficiency was almost 5-10% more for directly coupled system in comparison to DC-AC-DC system. The weighted efficiency for both the configurations were calculated, where the weights were based on the occurrences and the energy contribution of an irradiance bracket. The weighted efficiency for directly coupled system was 95.7% and for DC-AC-DC was 90.63% for Amsterdam. Even after considering all the seasonal, weather parameter and module parameter variations, directly coupled PV-Electrolyzer systems with shorter string length and bigger electrolyzer cells gave the best efficiency amongst all, which was above 90%.

Global average temperatures are setting new records compared to the previous decade, primarily due to the accumulation of greenhouse gases such as CO2. The growth potential in emerging economies worsens the current global situation. These CO2 emissions are needed to be curtained and efforts are underway. The emissions target set by the International Maritime Organization seems far-fetched given the forecast for the growth of the shipping sector. Although electrification has proven to reduce emissions, the energy efficiency metrics (EEDI, EEXI) are incapable of providing the quantitative benefits that can be realized by switching to a bipolar DC (BiDC) grid when compared with the AC grid. This thesis aims to evaluate the feasibility of BiDC grids over the complete operating profile of the ship by comparing ten arrangements, five for AC and five for DC, with the Key Performance Indicators (KPIs). The arrangements are two diesel generators (AC1 & DC1), two diesel generators and battery (AC2 & DC2), two diesel generators, battery, and shore power (AC3 & DC3), one diesel generator and battery (AC4 & DC4), one diesel generator, battery, and shore power (AC5 & DC5). The comparison is made for key performance indicators like CO2 emissions, fuel consumption, electrical power requirement, propulsion power cable losses, capital costs, operating costs, propulsion system weight, and load carrying capacity of the ship. In this study, a ferry is chosen to compare the two grid architectures. To establish the required background knowledge for the reader of this thesis, the literature on the current emission regulations, the different propulsion systems, and the differences in AC and DC grids are presented. The ship’s propulsion and auxiliary power demand are not freely available and are in the possession of the ship operator. Therefore, first, a preliminary methodology was developed for estimating the ship’s mechanical power requirements. Second, the ship’s electrical power requirements are estimated from the understanding of the literature. The cable power losses are also calculated for the AC and BiDC grid topology. Third, the diesel generator model was developed from the data points of a marine diesel generator for the AC and BiDC grid operation, i.e., the fixed speed and variable speed operation. Finally, an optimization problem was formulated for all the ten arrangements for minimising the CO2 emissions over the complete operational profile of the ship. Overall, it was observed that the minimum emissions achieved by BiDC grids, when compared with AC grids for all the arrangements, the emissions from the former were lower. Furthermore, these results depend on the operating profile of the ship, the shore infrastructure, the diesel generator and battery configuration and their technical parameters. However, these findings demonstrate the possible potential of BiDC grids in emission reduction from the shipping sector.

Although the metro is a very energy efficient option for passenger transportation compared to internal combustion vehicles, it is still a very prominent energy consumer. Further, due to the global pressure to reduce CO2 emissions, there is a worldwide demand for efficiency and sustainability improving solutions and approaches for railway traction systems. Also, in Amsterdam, the municipality sets great ambitions regarding installed PV power in the city and the emissions related to public transit. The municipality of Amsterdam has the goal to have an installed PV power capacity of 250 MWp in 2022 and to reduce CO2 emissions with 55% by 2030 compared to 1990. Also, the metro system has to become 35% more energy efficient in 2030 compared to 2013. Moreover, the municipality wants to offer emission free public transport in 2025. To reach these ambitious goals, no piece of land suitable for PV can be left unused.  In the past, PV systems have already been installed on metro station roofs, connected to the AC-side of the metro traction network. However, no PV systems have yet been connected to the DC-side of the traction network, which would have enabled the connection of PV systems anywhere along the track. Also, there is still a lot of unused land along the metro tracks in Amsterdam, with a total available surface of 47991 m2. This land could potentially be used to install PV systems with a total installed power of 4511 kWp [1]. To increase the amount of installed PV power and to fully utilise the available land along the metro tracks, this report focuses on the connection of PV systems directly to the metro traction system, at the DC-side of the traction network, to accelerate the energy transition and to reach the sustainability goals in Amsterdam. This report includes a comprehensive study on the challenges related to the integration of PV systems at the DC-side of the traction network, as well as possible solutions to mitigate these challenges. To further increase the energy efficiency of the traction network with PV systems, the integration of additional solutions like energy recuperating inverters in power substations and energy storage systems are discussed as well. To study the energy saving effects as well as challenges of PV systems connected to the DC-side of the traction network, a Simulink model of the traction network is created, using measurement data of an M5 metro, the metro timetables and traction network parameters as an input. The input data for a specific pilot location, located nearby station Amsterdam RAI, is used for the simulations. Also different PV systems, with sizes of 202 to 799 kWp are simulated with PVsyst, using meteorological data for the pilot location. The simulation results are used in the Simulink model, to simulate a PV system connected to the traction network. Also inverters and storage systems are simulated using the Simulink model.  The results show an energy saving enhancement effect, due to decreased ohmic losses in the network as a result of the injection of PV power, meaning that more energy is saved than supplied by the PV system. However, due to the intermittency of the load and regenerative braking, only 51.5-83.0% of the potential PV energy yield can be supplied to the traction network, depending on the PV system size, resulting in a waste of PV energy. A variable PV output control strategy is proposed, which can increase the energy output of the PV system with about 13-22%. When using an inverter in a substation, the amount of supplied PV energy can be increased to 83.4-97.6%, by exporting the surplus of energy to the power grid. Also, using the inverter, a significant amount of braking energy can be recuperated, which would otherwise have been wasted in the braking resistors of the metros, leading to even more energy savings. Also, an energy storage system is simulated. The results show that the storage system could theoretically prevent any waste of PV energy, while also saving braking energy and decreasing ohmic losses even more.  Using the results for an economical feasibility study on a PV system with and without inverter shows that a PV system operating alone would have payback period of 11 years, which can be reduced to 5 years by using an inverter. Although this research focuses on integrating PV systems on the metro network of Amsterdam, the methodology can be applied to DC traction systems worldwide.

In this thesis, a new photovoltaic fault detection and classification method is proposed. It combines the generation of a synthetic photovoltaic training database and the use of a machine learning model to detect and classify faults in small-scale residential PV systems. The database was generated in Matlab, and the machine learning modeling was done with the scikit-learn library for Python. From the modeled PV systems, solely power yield is used as an indicator, combined with system age and meteorological conditions. Using these features, four types of machine learning models are used to detect malfunctioning PV systems and classify short-circuit faults and open-circuit faults. This thesis also shows the benefit of a synthetic PV training base as opposed to alternative methods, with increasing performance due to control of database balance. The result of this thesis is a method that can be used to construct a model for detection and classification of photovoltaic faults, specific to a single residential PV system. Malfunctioning systems can be detected with an accuracy of 80.4% using a random forest algorithm. For fault type classification, an F1-score of 0.759 was achieved, also using a random forest.

An increasing number of photovoltaic (PV) systems are being installed worldwide and residential sector is responsible for a large part of this growth. Small scale PV systems do not have complex measuring devices and their breakdowns are not spotted immediately by the system owners. This might lead to prolonged time without generating power and creating both financial loss and environmental damage. This thesis presents a method of PV yield nowcasting laying foundations for remote monitoring. Early detection of faults is the first step towards eliminating the described issues. In this project four machine learning models for predicting solar yield were developed: ElasticNet, Polynomial Regression, Random Forest, and Extreme Gradient Boosting (XGBoost). The models were created both for daily and hourly data sets, as some inverters can log daily yields only. In both cases, the utilized data set consisted of data for the time between July 1st, 2018 and June 30th, 2019 and corresponding to 1,102 PV systems which is five times more than the largest data set studied in the found literature. The average PV system size in the data set is 4.44 kWp. Utilized inputs next to weather data and previous yields included shading factor describing fraction of direct light unable to reach PV system due to surrounding obstacles. Calculation of shading facor was based on 360⁰ pictures taken at the site. XGBoost algorithm turned out to be the most suitable for the task of PV yield nowcasting obtaining RMSE of 1.48 kWh and MAE of 0.877 kWh for hourly data aggregated to daily values and evaluated on future time steps. Currently used commercial software of Solar Monkey has RMSE equal 2.237 kWh and MAE equal 1.5 kWh. XGBoost model trained on daily data obtained RMSE 1.185 kWh and MAE 0.698 kWh outperforming hourly model most likely due to utilization of Hidden Markov Model for data cleaning. Next to overall performance, per system metrics were calculated for the hourly XGBoost. Mean individual RMSE for previously seen systems is 1.656 kWh while for unseen systems it equals 1.666 kWh. This means the model scales well to previously unseen systems and implies that its parallelized version is not necessary. Also, the model's learning saturates after seeing data corresponding to one year and 278 PV systems. Precalculation of GPOA worsened performance with respect to the model utilizing GHI. Hourly XGBoost has hourly RMSE of 0.281 kWh under clear sky and 0.377 kWh under partly cloudy sky which indicates it is more mistaken for cloudy conditions. This could be caused by low quality of cloud coverage data. The model also has large relative errors for small irradiance values which occur mostly in January and December, as well as just after sunrise and just before sunset. This issue is caused by using squared error as loss function during model training. Despite these shortcomings, the conclusive results recommend industrial implementation of the developed model.