基于机器学习算法的蒸发量模型评估

目的]探讨合理的气候因子个数,建立蒸发量模型,提出基于特征选择算法筛选最优特征集。方法]以陕西榆林、泾河和汉中3个气象站16年(2005-01至2021-03)的逐小时观测资料为研究对象,利用特征选择函数和遍历循环方法对模型参数、特征变量个数进行优化。基于最佳参数结合随机森林模型和多元线性回归模型两种机器学习算法建立榆林、汉中和泾河地区蒸发量模型,采用平均绝对误差、均方根误差和平方相关系数三项指标评估模型的预测精度。结果]特征变量和随机森林模型中的决策树个数分别是8,61时,模型预测效果最佳。采用优化的随机森林模型、多元线性回归模型评估3个地区的平均绝对误差均为0,均方根误差除泾河地区相等外,榆林、汉中地区的均方根误差均小于优化的多元线性回归模型。优化的随机森林模型预测榆林、泾河和汉中地区蒸发量拟合效果分别为0.85,0.90,0.86,优化的多元线性回归模型的拟合效果分别为0.77,0.83,0.79。结论]整体而言,优化的两种模型都具有良好的预测效果且随机森林模型的预测效果优于多元线性回归模型。

Evaporation Model Evaluation Based on Machine Learning Algorithm

Objective]The optimal feature set was determined in order to explore the reasonable number of climatic factors to establish an evaporation model, based on filtering by feature selection algorithm. Methods] 16 years(2005-01 to 2021-03)of Yulin, Jinghe and Hanzhong weather stations of hour-by-hour observation data were taken. The model parameters and the number of characteristic variables were optimized by using feature selection and traversal loop methods. Based on the best parameters combined with two machine learning algorithms, the random forest model and the multiple linear regression model, evaporation model suitable for 3 regions was established, and the three indicators of average absolute error, root mean square error and square correlation coefficient were used to evaluate the model accuracy. Results] The number of feature variables and decision trees are 8 and 61, the prediction effect of the model is the best. The root mean square error was used to evaluate the prediction effects of the optimized random forest model and the multiple linear regression model, and the error size of the two models was 0. Except for the Jinghe area, the root mean square error of Yulin and Hanzhong was smaller than that of the optimized multiple linear regression model. Predict effects of evaporation in Yulin, Jinghe and Hanzhong fitted by the optimized random forest model are 0.85, 0.90 and 0.86, and the optimized multiple linear regression model fitting effects are 0.77, 0.83 and 0.79. The optimized random forest model of the average absolute error and the root mean square error are lower than optimized multiple linear regression model. Conclusion] On the whole, the optimized two models have good prediction effects and the prediction effect of the random forest model is better than the multiple linear regression model.