This thesis serves to finalise the bachelor graduation project on the topic of self-supervised federated learning, specifically the on-chip implementation of the algorithms. The goal of the project is to implement a self-supervised learning setup in a decentralised approach using Field-Programmable Gate Arrays (FPGAs) for the processing of data. In this thesis, we endeavour to illustrate the possibility of employing FPGAs to move the fairly compute-intensive self-supervised learning algorithms to the edge. We have developed a number of modules that can accelerate key algorithmic blocks that underlie the major bottlenecks of the classical application of the algorithms and showcase prospective results, which are extensively discussed afterwards to pave a clear path towards truly autonomous and efficient edge-intelligence.

In today’s society, the need for alternate high-speed data transmission solutions has increased due to the radio frequency (RF) spectrum’s growing congestion. Visible Light Communication (VLC)-based LiFi has emerged as a possible alternative to current wireless technologies like WiFi. The difficulty, though, is that the photodiode performance as a receiver for VLC under outdoor conditions rapidly degrades. To solve this problem, the photodiode receiver was replaced by a PV cell receiver with the primary advantage of being optimized for outdoor conditions.
While earlier research has concentrated on light and modulation optimization to
improve the performance of VLC systems, our research adopts a novel strategy by
analyzing the effect of different light wavelengths on seven different PV cell technologies in hopes to design an optical filter and realize noise-free VLC for PV cells. We specifically are interested in identifying the LED wavelength that has the greatest bandwidth in order to increase the possible data transmission speeds in PV-VLC systems. In order to do this, a thorough characterization of several colored LEDs with different wavelengths was carried out across seven PV technologies, including PERC, AL-BSF (5INCH), AL-BSF (6INCH), SHJ, IBC, Busbar-free Al BSF, and TOPCon. Each LED wavelength was tested under three different intensities of light (100, 300, and 500 W/m2).
In terms of PV technology, TOPCon demonstrated superior performance at low bias voltages, while Busbar-Free Al BSF(EEPV) outperformed the other PV technologies at higher bias voltages, especially at the maximum power point. Furthermore, the analysis of light intensity revealed that the bandwidth does not only depend on capacitance but also on the internal resistance of the PV laminate. For the c-Si solar laminates tested the considerably larger resistance at lower light intensities in the bias voltage interval from 150mV to 450mV resulted in lower bandwidths at lower light intensities. The measurement results under different LED colors, concluded that when operating near maximum power point, the variation in bandwidth between different colored LEDs could significantly affect data rates, particularly when considering the higher SNR results at lower bias voltages that contribute to achieving faster data rates in the PV-VLC system.

With an increase in higher power (ultra) fast EV charging stations and a new EU law for mandatory deployments of EV charging stations, an optimal MVAC grid connection for these charging stations is desired. Considering this connection, the use of a solid-state transformer (SST) performs better in terms of weight, size and efficiency with a similar cost, compared to the conventional low-frequency transformer (LFT) for MVAC to LVDC conversions. This study compares three MMC-based SST topologies where a DAB is connected to each submodule: Double-Star, Single-Star and Single-Delta. The cost of an MMC mainly depends upon the required energy storage and the switch stress. The topologies are therefore mainly compared based on these parameters, but other parameters are considered as well. With the goal of minimising the required energy storage, an 80% MMC energy storage reduction compared to the theoretical minimum requirement has been achieved by control.

A novel mathematical model has been proposed for each SST; this uses an MMC model based on the switching function and a DAB model based on the single phase-shift control. The model has been implemented in simulation and proven to accurately determine the average behaviour of the SST while significantly reducing the simulation complexity and time. This model helps to understand the SST's behaviour and furthermore allows for control system tuning both manually and automatically using the Genetic Algorithm.

Three different control systems have been designed. Control system A, where the MMC current is controlled by the average submodule voltage and the DAB controls the LVDC bus voltage. Control system B, where the DAB controls each individual submodule voltage and the MMC current is controlled by the LVDC bus voltage. Control system C, which is similar to control system B, but uses a stable DC voltage source on the LVDC bus and a set MMC current. An improved DAB control has been designed for control systems B and C, resulting in control systems B* and C*. This control is based on the use of a current feedforward and a PI controller combined with the inverted equations of the DAB model. The three topologies have been successfully simulated with this improved DAB control system using an 80% MMC energy storage reduction compared to the theoretical minimum requirement. This improves the general behaviour and significantly reduces the cost of the SST. The improved DAB control system has been tested using a hardware DAB with a current step input, which has shown that the new control system performs better in voltage peak reduction compared to PI control.

Out of the three topologies, the Single-Star topology performs the best. It has the lowest energy storage requirement and switch stress. Furthermore, its behaviour lies the closest to the ideal model. The best result for the Single-Star SST is achieved with the proposed improved DAB control, which leads to the lowest energy storage requirement, switch stress, THD, submodule voltage deviations and settling time. Out of these two control systems, C* performs the best.

The compensation network in wireless power transfer (WPT) system has many functions, including increasing efficiency, providing constant voltage/current output, reducing volt-ampere rating, etc. With the popularity of wireless charging for electric vehicles, more and more compensation topologies with distinctive characteristics are proposed. Therefore, a comprehensive comparison among various com-pensation topologies under rated operating condition is necessary for the selection of compensation topology for WPT. In this thesis, eight compensation topologies are selected for benchmark and com-pared in terms of efficiency, component voltage/current stress, design freedom, misalignment be-haviour, etc. under the rated operating condition set based on SAE J2954 [1] standard. Given the analytical comparison results, the S-S compensation and LCC-S compensation are selected for further analysis and experimental verification.
In the practical design process of the compensation, the voltage/current stress on each compo-nent and the implementation of zero voltage switching (ZVS) on switches are two critical issues to be considered. For the calculation of components’ peak voltage/current, the fundamental-frequency analysis is the most widely used calculation method, which is inaccurate in some inductors’ voltage peak calculation. Therefore, a new resonant inductor voltage peak calculation method is proposed in this thesis, which is proved to be more accurate in both simulation and experiments in S-S and LCC-S compensations. For the implementation of ZVS, few studies have given the calculation method of switching current due to its complexity. In this thesis, a new switching current calculation method for LCC-S compensation is proposed and compared with the existing calculation method. A tuning method for ZVS implementation based on this calculation method is also proposed. Results from simulation and experiments under various operating conditions are provided to verify the accuracy of the newly proposed switching current calculation method. In the experiments, the LCC-S compensation and S-S compensation are compared under different power and input voltages. Experimental results show that the efficiency of LCC-S compensation is higher at low power, because of the lower conduction loss on MOSFETs. However, the efficiency of S-S compensation is higher at high power due to less losses on compensation components.

This thesis addresses a novel configuration of hot-swappable Multi-Active-Bridge (MAB) DC/DC converter in renewable-energy-based systems. As an extension of the Dual-Active-Bridge (DAB) converter, the MAB converter provides a multiport DC/DC converter solution with bidirectional power flow, simple phase shift control and galvanic isolation, but leaves problems in coupling operation, low availability and control complexity. In this thesis, the MAB converter with hot-swapping capability is proposed and studied. With a dedicated transformer design and circuit re-configuration, the converter allows a single port rigidly determines the voltage on the magnetizing inductance of the transformer, while the power flow of all other ports are naturally decoupled. The decoupled ports of this proposed MAB converter shows high independence in control and operation, thus can be hot-swappable and decentralized controlled. The hot-swapping interface circuit consisting of Negative-Temperature-Coefficient (NTC) and Metal-Oxide-Varistor (MOV) is proposed, and the benchmark of components selection is elaborated. The control of hot-swappable ports is implemented with a decentralized voltage loop and a SYNC module for phase shift control synchronization. A 2.2kW, 200kHz, two 350V, two 48V, four-port GaN-based MAB converter is simulated and tested for evaluation of decoupled operation, hot-swapping and decentralized control. Finally the hardware is designed and to be implemented for further and extended evaluation of hot-swappable MAB converter.

This thesis work investigates and provides an analysis of the potential benefits of an electricity - gas - heat integrated energy system, putting extra focus on the waste heat potential from fuel cells and electrolysers. The main focus is given to the low-voltage distribution grid level, and a case study is presented for the Drechtsteden subnetwork operated by Stedin, which, together with the IEPG department at TU Delft, is the promoter of this study. The novelty items of this work consist in the analysis of the waste heat potential of hydrogen conversion assets connected to a district heating network along with the inclusion of electricity and hydrogen markets, all in the contest of an energy system optimization.

The integration between these three energy sectors is assumed to be chiefly driven by the operation of electrolysers and fuel cells. Two other main assumptions set the base for this work: first, the gas sector is assumed to be entirely repurposed to operate with hydrogen and, secondly, the waste heat coming from fuel cells and electrolysers is assumed to be the main thermal energy input of a district heating network. An electricity and a hydrogen market have also been modelled to simulate the interaction of this region with the external grid.
The analysis is carried out by means of a linear optimization algorithm coded using oemof, an open source python package for multi-energy system modelling and optimization. Most of the input data (i.e. energy demand and generation profiles) comes from the Integrale Ifrastructuurverkenning 2030-2050 study from TenneT, Gasunie and the Dutch DSOs, which has been regionalized for the Drechsteden region is order to optimize the investments on the main energy assets (transformers, hydrogen substations, electrolysers, fuel cells, batteries) needed to run the future energy system.

An optimal system configuration is also calculated for a "reinforcement" scenario, in which energy sectors remain independent, and a comparison with the system integration scenario is presented. The overall system costs appear to be only about 5% lower for the system integration scenario. The costs imputable to assets is higher with system integration due to the addition of expensive hydrogen conversion assets, but, in this scenario, the operation on the energy market (driven chiefly by hydrogen exports) is more advantageous.
Finally, a sensitivity analysis on the share of heat demand to be satisfied by district heating and on the price of batteries is carried out, along with an investigation on the effect of adding thermal storage to the system.

With the rise of various renewable energy sources, comes the possibility for combining the different type of sources together to balance their shortcomings. The goal is to find a renewable energy system that can be reliable year-round and be accessible for everyone. This research tries to model such a system. A model of a grid-tied PV-battery-electrolyser-fuel cell power system, which is based on a continuation of a series of master thesis projects, was expanded to include a neighbourhood with a fully electrical load or a combination of electrical and hydrogen loads. This model was developed to answer the following question. What is the techno-economic feasibility of a grid-tied PV-battery-electrolyser-fuel cell power system for a household area in the Netherlands which is either fully electrical or hydrogen integrated? This hybrid system is simulated by using the graphical interface program TRNSYS. The system size of the PV, batteries, electrolyser, fuel cell and hydrogen gas storage tank are optimised by the GenOpt, an add-on for TRNSYS. The optimisation algorithm will try to find the lowest levelised cost of energy(LCOE) while keeping the system self-sufficiency ratio(SSR) around 1 [%]. This will mean that only 1 [%] of the load is allowed to be extracted from the grid. The simulation is based on a neighbourhood that consists of 630 houses located in Pijnacker Netherlands. All houses will be equipped with a roof mounted solar PV system with centralised batteries, electrolyser, fuel cell and a hydrogen storage tank. If needed the model can be extended to include a small solar park next to the neighbourhood. The model will simulate two scenarios for a simulation time of one year, the first being that the neighbourhood is fully electrical and the second for a neighbourhood with integrated hydrogen gas in its consumption. The first one is the base, with only the electrical load demand of houses. Then the load profile will be extended by adding vehicle to the neighbourhood, including the heat demand of the house. These additional load profiles will either be electrical energy based for the fully electrical scenario or hydrogen gas based for the integrated hydrogen scenario. To estimate the economic development of this hybrid system, a price projection of PV, battery, electrolyser, fuel cell, hydrogen heating, heat pumps and inverters components were determined for the years 2020, 2030, 2040 and 2050. a, the cases will all be simulated for these years. The economic analysis will be over the systems lifetime, which is 25 years. Before the cases were simulated the model undertook a sensitivity analysis. From this resulted that the simulation start time can be moved from the 1st of January to the 2nd of March to relief the storage tank of getting depleted at the start of the simulation. A battery discharge constraint was lifted and this led the batteries to provide more energy. A forecasting method was applied to the system that effectively reduced the electrolyser on/off cycles by 60 [%], which increased the lifetime of the electrolyser component. From a technical feasibility analysis of the cases, it resulted that the integrated hydrogen scenario was not technical feasible with the PV system (roof mounted with the PV park) of this model. All the integrated hydrogen scenario cases resulted in a depleted hydrogen storage tank, which forced the system to buy the hydrogen demand externally. The system will rely on an external source more than the allowed 1 [%] (hydrogen gas SSR >> 1 [%]) of the load demand. From the fully electrical scenario the 2020 C-E-(V+H) case resulted not be technical feasible with a SSR value of 2.1 [%]. All the other cases were technical feasible. From an economic and cost perspective, the cases resulted that the LCOE reduced with the years. The lowest LCOE value found was for the C-E-(Base) case, which reduced from 0.44 [€/KWh] in 2020 to 0.21[€/KWh] in 2050. The cost breakdown of the cases resulted in the PV system and the storage tank to be the most expensive components of this system. Due to the fact that the C-H2-(H) case had to buy a significant amount of hydrogen from an external source, this became a significant expensive cost of the system. Comparing the two scenarios resulted that the integrated hydrogen scenario system sizes were smaller, but this is an effect of the system being more eager to buy hydrogen gas then to expand the hydrogen production components. As both scenarios had different SSR values of their respected energy demands, a conclusion of which scenario is more beneficial will be inadequate.

PV systems in urban environments frequently get shaded by nearby objects, which greatly reduces their yield. A potential solution for such situations is the use of reconfigurable PV modules. They are a type of modules which can reconfigure themselves under varying illumination conditions to ensure optimum yield at all times. The individual PV cells of the module are grouped into reconfigurable units/ cell blocks which can be interconnected in various ways to form a variety of unique configurations. The modules have an inbuilt algorithm which is responsible for controlling the reconfiguration process. There are various kinds of reconfiguration algorithms found in literature, each working with varying input parameters and operating principles. One such algorithm is the short circuit current sensing (SCCS) algorithm developed by the PVMD research group at TU Delft. The algorithm is meant for series-parallel
connected reconfigurable modules and uses each cell block’s short circuit current as input. It is a synchronous reconfiguration algorithm, which means it runs at regular time intervals and not when there is actually a change in the module’s irradiance conditions. Therefore, the aim of this thesis was to develop a robust reconfiguration algorithm with a shade detection mechanism. Consequently, numerous versions of a reconfiguration algorithm with shade detection mechanism were defined and tested using a simulation framework for modelling reconfigurable PV modules. It was observed that the reconfiguration algorithm that performed the best in terms of yield, reconfiguration count and accuracy was the one referred to as the ”CC4AP+IR algorithm”. The algorithm uses the module’s and a single cell block’s operating current and voltage as inputs for shade detection. It was observed that the algorithm reconfigured a fraction of the time that the SCCS algorithm reconfigured and had a DC yield comparable to the SCCS algorithm. The new algorithm was also tested using real-life data to validate its performance. This experiment indicated that if the module were to be run using the new reconfiguration algorithm instead of the SCCS algorithm, it would have run only 6% of the time that the SCCS algorithm ran, but have a comparable DC yield. The significantly lower reconfiguration count is relevant because in practice, the PV modules are connected to power converters which will not be able to tolerate the high fluctuations in output owing to frequent reconfigurations.

Given the rising research of thin film solar cells in recent years, flexible technology has been proven to be more light weight and cost effective. As photovoltaics is increasingly becoming the front runner in sustainable energy production, concerns over the associated impacts of solar modules throughout their life cycle are also increasing. This study quantifies the environmental impacts through LCA analysis for Roll-to-Roll (R2R) production process of thin film flexible silicon-based solar modules manufactured by a Dutch company HyET in The Netherlands.

This study considers three product lines over a life cycle demarcated into its manufacturing, encapsulation and installation. The three product lines (cases) are as follows.
1. Single junction a-Si with 7% efficiency.
2. a-Si/nc-Si tandem cell with 10% efficiency
3. a-Si/nc-Si tandem cell with 12% efficiency
The installation phase considers a rooftop setup of capacity of 2.1 kWp with Balance of System (BoS) components.

LCA analysis is carried out on Simapro 9.1.0.11 following the guidelines and the framework of International Organization for Standardization ISO14044. The outcome of LCA analysis is measured in terms of Global Warming Potential (GWP), Primary Energy Demand (PED) and Energy Pay Back Time (EPBT). Ecoinvent 3.5 is used as the primary database for these analyses to select the inventory. GWP is assessed using the CML-IA baseline method while PED is assessed using Cumulative energy demand v1.11 method. Sensitivity analysis is done by changing the location of production and up-scaling capacity.

The installation stage is observed to contribute the highest GWP and also has the highest PED on account of BoS components. The LCA analysis has demonstrated similar trends of GWP, PED for all three product lines. EPBT on the other hand, is longer for 10% tandem cell on account of relatively larger module area as compared to 12% tandem cell and also the relatively higher energy consumption as compared to the single junction cell. The choice of substrate material is seen to impact the assessment indicators significantly. Flexible glass is observed to be the optimal choice for large-scale production. The choice of encapsulant material also affects the indicators demonstrably. Sensitivity analysis shows a positive impact on the indicators through up-scaling, while the location is not established as a significant factor sufficiently under considered assumptions.

In urban environments, more and more building added and building integrated photovoltaic (PV) systems are found. These systems use conventional solar modules which have a poor performance under non-uniform illumination conditions. When the module is partially shaded, either at least one
subgroup is bypassed or the module current is limited by the current of the worst performing cell. This leads to significant and disproportionate power losses. One way to address this issue is to implement a solar module made of low reverse breakdown voltage (BDV) solar cells.
A thermo-electric simulation framework was developed in MATLAB to replicate two types of commercially available low BDV solar cells: Sunpower Maxeon gen2 and gen 3 cells. Using the above model the potential and performance of solar modules in urban landscapes has been evaluated. A comparative study of solar module models built with these cells and placed on three different locations on a single rooftop having completely different irradiance profiles has been assessed. A solar module with 3 bypass diodes made of gen 3 solar cells has a better performance than module made of gen 2 cells. This accounts not only for better forward parameters of gen 3 cells but also its improved mismatch losses owing to lower reverse breakdown voltage and better temperature coefficients.
The usage of low reverse breakdown solar cells is beneficial in conditions where partial shading is predominant. Other advantages of using these solar module over other technologies is there is that there is no need of any additional electronic components.

The residential and service sectors are responsible for the 39% of the total energy consumption. 80% of this energy is used for heating purposes. Achieving nearly zero energy buildings is a major goal for the near future. The design of an integrated electrical and thermal system can substantially help to reach such objective. The façade is the biggest area of a building and could be utilised to convert incident solar energy to electrical and thermal energy. In previous work (2019), the combination of building integrated PV modules with solar chimney on a façade was investigated to study its energetic performance. One of the main findings was that using the thermal energy from the photovoltaic modules on the air inside the cavity of the chimney is not possible due to its low quality of heat. It was concluded that the utilization of heat could be better by using auxiliary systems, like photovoltaic thermal systems for water heating. The aim of this work is to create a computational model to calculate the thermal and electrical performance of PV thermal (PV/T) modules combined with a solar chimney. Two new designs for better heat utilization where considered and compared. The first design is for the PV modules to be installed at the front of the solar chimney with fins (Fins design) and the second design is to couple the PV modules installed on the wall inside the solar chimney with water pipe system (Water pipe design). These designs are also compared with the same configurations but without the thermal systems (No fins & No water pipes designs). In addition, two new strategies are introduced to improve the performance of the model. The first strategy is the optimization of the solar chimney openings with respect of time for the highest power generation. The second strategy is the variation of the water mass flow with respect to time for highest water-thermal energy and optimum cooling of the PV cells. The model was developed in MATLAB and it is based on 2D finite difference transient implicit method. To reduce the computational time, the density of the nodes varies in respect of temperature gradient. In addition, secondary models were created to increase the accuracy and sensibility of the model such as water tank and water pump models. The models were studied for a building in Delft, Netherlands, and the scale of the chosen system is 10m2. It is concluded that Water pipes design provides the overall best energetic performance mainly because of the high water-thermal energy with yearly energy efficiency of 64.3%. However, the best electrical performance is observed from the Fins design and the best air-thermal performance from the No water pipes design.

Trolleybus systems may be a sustainable and cheap alternative to diesel and battery buses. However, to improve sustainablity, the trolleybus system must be powered by renewable energy sources. In this work, a case study on the Arnhem trolleybus grid is performed. First the bus power demands and bus traffic are constructed for the whole system, afterwards Photovoltaics and wind energy sources assited by ESS are sized to investigate the effects on renewable power utilization. It was found that for the Arnhem trolleybus system, a centralized approach with equal shares PV (51%, 4.3 MWp) and wind (49%, 1.75 MWp) and diurnal energy storage (5MWh - 10MWh) can provide a 72% to 78% of the total trolleybus load. This system will reduce the CO2 emissions by 80% over a 30-year lifespan.

The installation of photovoltaic systems in the urban environment is becoming increasingly common. These photovoltaic systems are affected by a large fraction of reflected irradiance as well as partial shading. In a similar way these issues also have an important effect on bifacial photovoltaic power plants. Irradiance models that are able to take these issues into account are therefore gaining interest. For an accurate economic assessment of a PV system and the calculation of delivered power it is becoming very important to be able to accurately determine the irradiance on a PV system. Therefore there is a rising demand for accurate irradiance models that are able to take into account the effects that arise in complex geometric scenarios. The research in this thesis report aims to develop such a model which is able calculate the irradiance incident on bifacial modules and PV arrays in complex landscapes by combining the concepts of ray tracing and 3D view factors. The implementation of the proposed model decouples the irradiance simulation into independent blocks that allow to efficiently calculate the spectrally-resolved irradiance incident on a PV module. This decoupling makes the proposed model suitable for simulation of tandem devices in the near future. The developed simulation has been experimentally validated using measurements from the PVMD monitoring station and also in comparison with other sophisticated irradiance simulation models. Experiments with a large fraction of specular reflected irradiance showed a good match between the measured and the simulated irradiance. Validation over a longer time period was also performed for 3 different sensors on the PVMD monitoring station roof. The results show an overall low mean bias error between the measured and simulated irradiance for the time period between mid-August 2020 to mid-October 2020. The proposed model is able to model the irradiance impinging on a PV system in a complex urban environment using a decoupled structure and its performance is comparable to sophisticated existing ray tracing models. The performed simulations show a good match with the measurements for different situations. Further improvements are the implementation of higher order reflections, computational optimization and more extensive validation.