基于机器学习的青海荒漠区光伏电站土壤温度预测OA

Photovoltaic array(PVA)layouts induce spatial differentiation in soil water and heat,which can affect soil nutrients and vegetation restoration in desert areas.This study examined soil temperature beneath fixed-axis and inclined single-axis PVAs at the Gonghe photovoltaic power station in Hainan Tibetan Autonomous Prefec-ture,Qinghai Province.We applied artificial neural networks(ANNs),random forests(RFs),support vector ma-chines,and AdaBoost models to predict soil temperatures at 10 cm,20 cm,and 40 cm depths,and compared pre-dictive performance across algorithms.Results showed that PVAs produced seasonal effects on soil temperature:a summer cooling effect of 4.15℃and a winter warming effect for fixed-axis arrays of 2.05℃.The key factor in-fluencing soil temperature under inclined single-axis PVAs and in the control area was the air temperature,while the saturated vapour pressure deficit was most influential beneath fixed-axis PVAs.Model performance varied by location and depth:the ANN model performed best in the control area at 10 cm depth(R2=0.983,RMSE=1.136℃),whereas the RF performed best beneath PVAs(R2=0.984,RMSE=1.040℃).Overall,prediction accura-cy declined with increasing soil depth,and accuracy under inclined single-axis arrays exceeded that under fixed-axis arrays.These results offer a practical reference for predicting soil temperature at power stations using PVAs in desert environments.