基于多组群教学优化的随机森林预测模型及应用

为有效提高水文预测预报精度,提出了一种基于多组群教学优化(MGTLO)的随机森林(RF)预测方法,利用MGTLO算法对RF两个关键参数进行优化,构建MGTLO-RF预测模型,并与基于MGTLO算法优化的支持向量机(SVM)、BP神经网络两种常规预测模型作对比分析。以云南省龙潭站月径流和年径流预测为例进行实例研究,利用前44 a和后10 a资料对MGTLO-RF等3种模型进行训练和预测。结果表明:所提出的MGTLO-RF模型具有更好的预测精度和泛化能力,可作为水文预测预报和相关预测研究的一种有效工具。

Random forest forecasting model and its application in hydrology based on optimization of multi-group teaching-learning

In order to effectively improve the accuracy of hydrological forecasting, a multi-group teaching-learning optimization (MGTLO) based random forest (RF) prediction method is proposed. The MGTLO algorithm is used to optimize two key parameters of RF to construct the MGTLO-RF prediction model, which is compared with the results of support vector machine (SVM) and BP neural network optimized by the MGTLO algorithm. The monthly runoff and annual runoff prediction of Longtan Station in Yunnan Province are taken as an example for case study. The first 44 years and the last 10 years are respectively used for model training and forecasting of three models of MGTLO-RF, MGTLO-SVA, MGTLO-BP. The results show that the proposed MGTLO-RF model has better prediction accuracy and generalization ability, and can be used as an effective tool for hydrological prediction and related prediction research.