基于机器学习算法构建新疆积雪覆盖率预测模型  

Construction of a Snow Cover Prediction Model in Xinjiang Based on Machine Learning Algorithm

作　　者：邓文彬[1] 侯雪晴 DENG Wenbin;HOU Xueqing(College of Architecture and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学建筑工程学院,新疆乌鲁木齐830046

出　　处：《应用基础与工程科学学报》2024年第6期1664-1677,共14页Journal of Basic Science and Engineering

基　　金：新疆维吾尔自治区自然科学基金项目(2022D01C55)。

摘　　要：积雪作为宝贵的淡水资源,其覆盖率的变动对农牧业经济的发展具有深远影响.当前对积雪覆盖率的预测研究较少,为提升积雪覆盖率预测的准确性,基于机器学习算法,构建支持向量回归(SVR)、粒子群(PSO)优化SVR、随机森林(RF)、XGBoost及优化后的XGBoost预测模型对新疆积雪覆盖率进行预测研究,并对模型预测精度进行对比分析.研究结果表明:RF和优化后的XGBoost模型的R^(2)均大于0.9;传统SVR模型的R2均小于0.8,而PSO算法优化后的SVR模型的R^(2)均大于0.8,部分大于0.9;XGBoost模型的R^(2)均低于0.4.说明RF、优化后的XGBoost及PSO-SVR模型在积雪覆盖率预测研究中呈现出较高精度,XGBoost模型的预测结果最差,且利用不同算法对传统模型进行优化在研究中十分必要.Snow cover is a kind of valuable freshwater resources,and the change of snow coverage rate has a profound impact on the development of agriculture and animal husbandry.There has been little research on the prediction of such coverage rates so far.In order to improve the accuracy of such prediction,this study uses machine learning algorithms to construct Support Vector Regression(SVR),PSO-optimized SVR,Random Forest(RF),XGBoost and optimized XGBoost prediction models,which are utilized to predict snow cover in Xinjiang,while comparing and analyzing the prediction accuracy of the models.The results show that the R^(2) values of both RF and optimized XGBoost models are greater than 0.9;the R^(2) values of all traditional SVR models is less than 0.8;the R^(2) values of all PSO-optimized SVR models are greater than 0.8,with some greater than 0.9;and the R2 values of all XGBoost models are lower than 0.4.These data indicate that the RF,optimized XGBoost and PSO-SVR models can deliver high prediction accuracy for snow cove;the XGBoost models have the poorest prediction results;and it is necessary to optimize traditional models with different algorithms.

关 键 词：积雪覆盖率 SVR 粒子群优化算法 RF XGBoost 参数寻优 

分 类 号：P468.025[天文地球—大气科学及气象学]