Development and Implementation of a Shadow Loss Model to Improve the Accuracy of Hourly Simulated Photovoltaic Power Generation

In regions with a high share of decentralised Photovoltaic (PV) systems, shading-related generation losses can significantly impact power forecasts and grid planning. While centralised systems are less affected, a shadow loss model is essential for accurately estimating PV electricity output, particularly in distributed networks. This thesis presents the development and implementation of a shadow loss model aimed at improving the accuracy of hourly PV power simulations. The model was evaluated using 40 reference systems located across Sweden, primarily in the southern half. Each system included known parameters such as module specifications, tilt, azimuth, and measured generation data from 2017 to 2024. Simulations were first run without the shadow model and then repeated with it, incorporating surrounding LiDAR data to generate a shadow map and calculate system-specific shading effects. Model performance was assessed using Becquerel Sweden’s Skill score-method, which compares the coefficient of determination (R2), root-mean squared error (RMSE), and mean absolute error (MAE) between baseline and shadow-adjusted simulations. Results showed improvements of up to 9.6% in R2Skill score, 29.5% in RMSESkill score, and 24.4% in MAESkill score. A majority of the systems (36 out of 40) showed some improvement, with 18 exhibiting substantial gains. Only two systems experienced notable regressions, primarily in RMSE and MAE. These results confirm the model’s effectiveness in reducing overestimation and enhancing simulation accuracy, especially for systems subject to partial or heavy shading.