An investigation of photovoltaic power forecasting in buildings considering shadow effects: Modeling approach and SHAP analysis

The power generation of distributed photovoltaic (PV) systems often suffers interference due to shadows cast by surrounding buildings. To improve the accuracy of PV power forecasts, this paper presents a PV power prediction method that takes shadow effects into consideration. Firstly, a convenient P...

Full description

Saved in:
Bibliographic Details
Published in:Renewable energy Vol. 245; p. 122821
Main Authors: Fu, Jiaqian, Sun, Yuying, Li, Yunhe, Wang, Wei, Wei, Wenzhe, Ren, Jinyang, Han, Shulun, Di, Haoran
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.06.2025
Subjects:
algorithms
ambient temperature
China
Convolutional deep neural networks
Photovoltaic power
power generation
Power prediction
prediction
relative humidity
Shadow effect
SHAP analysis
solar collectors
solar energy
solar radiation
China
Power prediction
SHAP analysis
Photovoltaic power
Shadow effect
Convolutional deep neural networks
ISSN:0960-1481
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The power generation of distributed photovoltaic (PV) systems often suffers interference due to shadows cast by surrounding buildings. To improve the accuracy of PV power forecasts, this paper presents a PV power prediction method that takes shadow effects into consideration. Firstly, a convenient PV shadow model was formulated for predicting the proportion of PV shaded (PPS), using theoretical derivation and a zoning shading judgment strategy. Subsequently, a PV power prediction method was proposed based on PV shadow forecasting and the convolutional deep neural network algorithm. Finally, this method was applied to a carport PV system in a building in Beijing, China, and SHAP analysis was utilized for the interpretation. The results show that the proposed method can automatically recognize shadow conditions, and significantly improve the predictive accuracy of PV power, reducing the MAE by 10.1 % and increasing the R2 value from 0.91 to 0.94. The ranking of feature importance to the PV power prediction model is as follows: solar radiation, hour, ambient temperature, PPS, and relative humidity. This study offers a feasible solution for predicting power generation of PV systems that are subject to shadow shading from buildings.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0960-1481
DOI:10.1016/j.renene.2025.122821