基于机器学习的马尾松毛虫发生面积预测模型

Prediction Models of Occurrence Area of Dendrolimus punctatus Based on Machine Learning

为了提高马尾松毛虫预测预报的准确性,分别采用随机森林、支持向量机和深度学习3种机器学习模型,应用气象因子预测马尾松毛虫发生面积,并将模型预测结果与实际发生面积进行比较。结果表明:3个机器学习模型的拟合效果均优于多元线性模型,其中深度学习对马尾松毛虫发生面积的预测最为稳健,拟合决定系数(R2)和预测准确率(PA)最高(R^2=0.9901,PA=99.27%),均方根误差(RSME)最低(RSME=0.1711);支持向量机模型预测效果也较好(R^2=0.9617,RSME=0.3275,PA=92.13%)。深度学习可用于马尾松毛虫发生面积与气象因子非线性模型的构建。

In order to improve the accuracy of predicting the occurrence area of Dendrolimus punctatus,we used three machine learning models (random forest,support vector machine,and deep leaning) and meteorological factors to predict the occurrence area of D.punctatus,and compared the predictive results and the actual occurrence area.The results showed that the fitting effect of these three machine learning models was all better than that of the multivariate linear model.The deep leaning model was the most robust to predict the occurrence area of D.punctatus,its fitting determinant coefficient (R^2) and prediction accuracy ( PA ) were the highest,being 0.9901 and 99.27%,respectively,and its root mean square error ( RSME ) was the lowest (0.1711).The prediction effects of the support vector machine model were also better ( R^2=0.9617,RSME =0.3275,PA =92.13%).This study suggested that the deep leaning model could be used for the prediction of occurrence area of Dendrolimus punctatus by meteorological factors.