The growing global energy demand increases the need for renewable energy sources. This increase requires land to be occupied, competing with other activities such as agriculture and residency. In such a situation, renewable energy sources expand to other environments like the ocean. However, this new scene poses some challenges, such as the effect of waves on photovoltaic (PV) performance. Consequently, this study aims to evaluate the power output of an Offshore Floating PV (OFPV) system located in the North Sea considering the effect of the waves. A 3D mechanical movement model, which has been validated with data from a real system, is developed for this purpose. A sensitivity analysis is conducted to determine how the size of fluctuations depends on the dimensions of the floater. The main outcome is that a heavy and wide floater aligned with the most common wind direction reduces angle variations. Results from DC power simulations show that sea fluctuations have a negative yet small influence on PV power production. Over the course of the year, these losses amount to just 0.1% of the annual energy yield. However, a hypothetical optimally-tilted PV system placed on water would still generate 14.6% more DC power output than the floating one. On the AC side, laboratory experiments show that these oscillations negatively affect the inverter efficiency during rough sea conditions by a decrease of over 2 percentage points compared to a still system.

Due to the inherent uncertainty in photovoltaic (PV) energy generation, an accurate power forecasting is essential to ensure a reliable operation of PV systems and a safe electric grid. Machine learning (ML) techniques have gained popularity on the development of this task due to its increased accuracy. Most literature, however, focuses only on less than 5 PV systems during training process, which does not ensure generalization to unseen systems. When in presence of a large feet, regional forecasts are the norm. Nevertheless, none of these approaches are usable when it comes to monitoring residential PV systems. In this work, we propose a single ML model that is able to predict the individual power of a large fleet of 1102 PV systems. XGBoost algorithm was selected as the most suitable algorithm for the task of PV yield nowcasting due to its performance and ease of use. This algorithm obtains Mean Absolute Error (MAE) of 0.877 kWh (considering an average system size of 4.44 kWp) and Mean Absolute Percentage Error (MAPE) of 23% for hourly data aggregated to daily values. XGBoost predictions for individual PV systems are on average two times better than currently used commercial software. We discuss the lack of a suitable loss function that can combine absolute and relative errors for residential PV yield forecasting. We also point out the lack of an adequate metric to compute the error made on the predictions and provide hints on developing a suitable one.

Machine learning is arising as a major solution for the photovoltaic (PV) power prediction. Despite the abundant literature, the effect of climate on yield predictions using machine learning is unknown. This work aims to find climatic trends by predicting the power of 48 PV systems around the world, equally divided into four climates. An extensive data gathering process is performed and open-data sources are prioritized. A website www.tudelft.nl/open-source-pv-power-databases has been created with all found open data sources for future research. Five machine learning algorithms and a baseline one have been trained for each PV system. Results show that the performance ranking of the algorithms is independent of climate. Systems in dry climates depict on average the lowest Normalized Root Mean Squared Error (NRMSE) of 47.6 %, while those in tropical present the highest of 60.2 %. In mild and continental climates the NRMSE is 51.6 % and 54.5 %, respectively. When using a model trained in one climate to predict the power of a system located in another climate, on average systems located in cold climates show a lower generalization error, with an additional NRMSE as low as 5.6 % depending on the climate of the test set. Robustness evaluations were also conducted that increase the validity of the results.

Monitoring residential scale photovoltaic (PV) systems is important for maximizing the energy yield and detecting malfunctions. Analytical-based approaches are not reliable in these systems because of the lack of on-site measurements and detailed PV system specifications. In this paper, a collaborative approach is proposed which does not depend on weather data but on similar PV systems. Based on the so-called performance-to-peer approach, the aim of this work is to improve this baseline model by adding PV systems characteristics and by optimizing with machine learning techniques. The methodology has been tested in a fleet of more than 12,000 PV systems located in the Netherlands with up to 7 years of data per system. The proposed model achieves an average (Formula presented.) of 94.1% and a NRMSE of 0.05, outperforming in terms of (Formula presented.) the baseline model by 1.4 points, and the analytical approach by 3.8. The data requirements of this model are not high: With 1,700 years of PV system data with daily resolution, the maximum performance can be achieved as long as a minimum of 6 months of data per system and 100 PV systems are considered. The application of this model for fault detection and categorization has also been shown. The proposed approach has shown its strengths with respect to other methods through its ability of distinguishing between system mismatch and actual fault and of adapting to new situations via retraining.

We introduce a novel simulation tool capable of calculating the energy yield of a PV system based on its fundamental material properties and using self-consistent models. Thus, our simulation model can operate without measurements of a PV device. It combines wave and ray optics and a dedicated semiconductor simulation to model the optoelectronic PV device properties resulting in the IV-curve. The system surroundings are described via spectrally resolved ray tracing resulting in a cell resolved irradiance distribution, and via the fluid dynamics-based thermal model, in the individual cell temperatures. A lumped-element model is used to calculate the IV-curves of each solar cell for every hour of the year. These are combined factoring in the interconnection to obtain the PV module IV-curves, which connect to the inverter for calculating the AC energy yield. In our case study, we compare two types of 2 terminal perovskite/silicon tandem modules with STC PV module efficiencies of 27.7% and 28.6% with a reference c-Si module with STC PV module efficiency of 20.9%. In four different climates, we show that tandem PV modules operate at 1–1.9 °C lower yearly irradiance weighted average temperatures compared to c-Si. We find that the effect of current mismatch is significantly overestimated in pure optical studies, as they do not account for fill factor gains. The specific yields in kWh/kWp of the tandem PV systems are between −2.7% and +0.4% compared to the reference c-Si system in all four simulated climates. Thus, we find that the lab performance of the simulated tandem PV system translates from the laboratory to outdoors comparable to c-Si systems.