降雨量及叠加预测方法研究

Analysis of Prediction Method for Rainfall and Superposition Rainfall

为了得到精确度较高的降雨量预测值及其叠加预测精度,利用小波神经网络和NARX动态神经网络对降雨趋势和降雨量进行预测,并分析降雨量叠加预测值的误差。研究表明,小波神经网络分析的月降雨量多个变化周期以及总的变化趋势较为准确;NARX动态神经网络预测模型测试误差为0.21%,回归效果图的相关系数R为0.99993,回判和检验误差分别只有0.22%和0.40%;降雨量叠加预测和检验误差较小,均未超过2%,能够满足降雨量不断叠加预测的要求。该方法能为边坡动态稳定性预测提供精确度较高的降雨量预测值。

In order to obtain High-precision rainfall predictive value and Superposition Prediction,Wavelet Neural Network and NARX Dynamic Neural Network methods are used to provide prediction of rainfall trends,rainfall amounts,and also to analyze the error of the superposed predictive value of rainfall. The results shows that the Wavelet Neural Network provide accurately in analyzing the monthly rainfall multiple variation periods and the overall variation trend;as the error test value of NARX Dynamic Neural Network prediction model is 0. 21%,the correlation coefficient R of regression renderings is 0. 99993,and feedback and test error value are only 0. 22% and 0. 40% respectively;the errors of superposed prediction and test of rainfall are small,less than 2%,which can meet the requirement of continuous superposed prediction of rainfall. This method will provide high-precision predictive value of rainfall for the prediction of slope dynamic stability.