基于随机森林模型的太阳辐射中长期变化分析

ANALYSIS OF MEDIUM-AND LONG-TERM CHANGES IN SOLAR RADIATION BASED ON RANDOM FOREST MODEL

该文基于多源辐射观测资料,采用随机森林(RF)算法、季节差分自回归移动平均(SARIMA)模型及特征重要性等方法,对山东省济南市太阳辐射长期变化趋势和影响因素进行综合分析。结果显示:RF模型拟合月太阳辐射效果较好,决定系数和平均绝对百分比误差分别为0.92和9%,优于SARIMA模型;济南市及周边地区月太阳辐射1980—2020年经历“变暗”到“变亮”的过程,空间呈现西北高东南低的特点;最高温度和日照时数是影响太阳辐射月变化拟合准确度的主要因素,降雨量是导致月太阳辐射总量突变的重要原因,大气污染物中SO2和O3与太阳辐射的相关性最大。

This paper presents a comprehensive analysis of long-term solar radiation trends and influencing factors in Ji’nan City,Shandong Province,based on multi-source radiation observations,using the random forest(RF)algorithm,seasonal autoregressive integrated moving average(SARIMA)model and characteristic importance.The results show that the RF model fits the monthly solar radiation better,with the coefficient of determination and the mean absolute percentage error of 0.92 and 9%,respectively,which are better than the SARIMA model.The monthly solar radiation in Ji’nan City and the surrounding area experiences the process of"darkening"to"brightening"from 1980 to 2020.The maximum temperature and sunshine hours are the main factors affecting the accuracy of the monthly variation of solar radiation,rainfall is an important cause of sudden changes in total monthly solar radiation,and SO2 and O3 among atmospheric pollutants have the greatest correlation with solar radiation.This paper is important for guiding the development of the solar energy industry and environmental management in Ji’nan.