摘要
针对目前短时降水量预测精度不高的问题,提出一种基于影响短时降水的物理量选取和RBF神经网络学习的降水量预测方法。首先对原始降水数据进行坐标格点化处理并对(NCEP/NCAR)数据文件中物理量参数进行读取和计算。在理论和数据表现上对降水量进行分析并使用多元线性回归进行预测,得到预测效果不佳。进而利用反向传播神经网络(BP)和径向基神经网络(RBF)进行预测。预测结果表明,径向基神经网络的预测效果更好。为选取最优模型建立自变量随机选取的21个RBF模型进行预测,仿真结果表明,选择六项物理量参数建立的RBF神经网络模型对未来24h降水量预测精度最高。拟合度、平均绝对误差,均方误差和均方根误差分别为0.998,5.04mm,56.34mm2和7.51mm,仿真时间为0.95s。
Aiming at the problem that the prediction accuracy of short-term precipitation is not high,a precipitation prediction method based on the selection of physical quantities affecting short-term precipitation and RBF neural network learning is proposed.First,the gridding of the original precipitation data is coordinated and the physical quantity parameters in the(NCEP/NCAR)data file is read and calculated.Precipitation in theory and data performance are analyzed and multiple linear regression is used to make predictions,and poor prediction results are obtained.Then back propagation neural network(BP)and radial basis function neural network(RBF)for prediction are used.The prediction results show that the prediction effect of RBF neural network is better in precipitation forecast for 24 hours.In order to select the best model,21 random RBF models with independent variables are selected for prediction.The simulation results show that the RBF model with six physical quantity parameters selected has the highest accuracy.The fitting degree,average absolute error,mean square error and root mean square error are 0.998,5.04mm,56.34mm and 7.51mm,respectively,and the simulation time is 0.95s.
作者
张启凡
王永忠
马俊逸
ZHANG Qifan;WANG Yongzhong;MA Junyi(China Commercial Flying Company Civil Aircraft Flight Test Center,Shanghai 201323;Civil Aviation Flight University of China,Guanghan 618300)
出处
《计算机与数字工程》
2022年第4期807-811,共5页
Computer & Digital Engineering
基金
国家级大学生创新创业训练计划项目(编号:201910624041)资助。
关键词
短时降水量
多元线性回归
BP神经网络
RBF神经网络
自变量随机选取
预测建模
short-term precipitation
multiple linear regression
BP neural network
RBF neural network
random selection of independent variables
predictive modeling