Identification of PV system shading using a LiDAR-based solar resource assessment model: An evaluation and cross-validation

Photovoltaic (PV) systems are subject to several different systematic de-rating factors, such as soiling, degradation, inverter mismatch and shading. With increasing penetration of PV in the local grid, Distribution Network Service Providers (DNSPs) are inclined to assess such losses, in order to accurately estimate the total regional power output of distributed PV. The most influential de-rating factor is shading, which can cause ramps on the generated power output, similar to clouds. In this study we evaluate and compare two fundamentally different methods for module orientation parametrisation and shading analysis of PV systems that have been developed in previous work. In the first method, LiDAR (Light Detection and Ranging) data are used to derive the PV module orientation and shading, referred to herein as the LiDAR model. The second method, referred to as the QCPV-Tuning model, is based on reported PV power generation, which is firstly quality controlled and parameterised in order to derive module orientation and a loss factor, LF, representing systematic de-rating factors. Secondly, variations in de-ratings throughout the day, mainly due to shading, are explored in a process referred to as Tuning. For both methods, binary time series are derived expressing the presence of shading, which are used to evaluate how the methods corroborate. We evaluate four cases; (case 1) evaluates the original versions of the LiDAR and QCPV-Tuning models, while in cases 2–4 improvements to the models are introduced. A new filter for extracting representative LiDAR data points for the shading analysis was introduced for the LiDAR model (case 2). For the QCPV-Tuning model significant inaccuracies in the parametrisation of the module orientation were identified due to strong shading in either morning or evening and were thus corrected to observed parameters (case 3). For (case 4) improvements on both models were introduced. The Pearson correlation coefficients of shading events for the methods were 0.28, 0.36, 0.42 and 0.50 for cases 1–4, respectively. A mismatch in the timing of shading events motivated the comparison of the mean hourly shading, with correlation coefficients of 0.34, 0.43, 0.49 and 0.57 for cases 1–4, respectively. The results of this study show that both methods can confidently be used for solar resource assessment, given the suggested improvements.