基于SSA-MGF的偏最小二乘回归神经网络的预报模型

Neural Network with Partial Least Square Prediction Model Based on SSA-MGF

本文采用奇异谱分析(S ingu lar Spectrum A na lys is,SSA)方法对原始降水序列重构,并用均生函数(M eanG enerating Function,M GF)方法对重构系列构造延拓矩阵,以此作为自变量,原始降水序列作为因变量,再利用偏最小二乘法提取对因变量影响强的成分作为神经网络的输入因子,原始序列作为输出因子,建立神经网络预测模型。通过对广西全区6月份降水量进行实际建模并与其它方法进行对比预测试验,结果表明,基于SSA-M GF的偏最小二乘回归神经网络预测模型较好,是一种具有较高应用价值的预测方法。

The primitive rainfall series are reconstructed as independent variables by singular spectrum analysis and restructured as dependent variables by mean generating function. The factor affecting is withdrew by means of partial least squares method to extract the most important components so that it can be input as the neural network, and a forecast model of the Neural Network is established with least squares regression based on singular spectrum analysis and mean generating function. The model of precipitation in June in Guangxi is worked out. Results show that the model is superior in predictions compared to the other models, and it is a useful model for the actual operational forecasting.