Photovoltaic power plants in mountainous area: Environmental impacts analysis based on random forest algorithm

The rapid growth of mountain photovoltaic (PV) plants has brought both environmental benefits and challenges. However, there is a lack of environmental impact prediction models for these plants, and existing models have limited predictive indicators, unable to differentiate between shaded and non-shaded areas of PV arrays. This study investigates the environmental impacts of a mountain PV plant in Hubei Province, China, and develops predictive models using 16 machine learning (ML) algorithms. Data collected over one year includes meteorological and soil parameters from shaded and non-shaded areas. The Random Forest (RF) model achieves the best performance, with R2values exceeding 0.90, particularly for soil temperature predictions (R2 > 0.99). Predictions show that the non-shaded area (IT) outperforms the shaded area (BL), as expected due to the absence of shading effects. Deep soil temperature shows the most stable prediction across both areas (R2 = 0.995). Soil electrical conductivity proves more challenging, but RF performs better in the IT site (R2 = 0.903) compared to the BL site (R2 = 0.880). This research improves existing environmental impact prediction models by incorporating more parameters and advanced ML techniques, supporting the sustainable development of solar energy systems and their integration with the environment.