以前期500hPa高度场、海温场为预报因子,采用径向基函数(RBF)神经网络与主成分分析相结合的方法,建立了广西中部5月平均降水预报模型。在5年独立样本的预测检验中,预测的平均相对误差、均方误差及平均绝对误差分别为18.12%、50.52和34.23。对比分析RBF神经网络与BP(Back Propagation)神经网络的预测结果,表明RBF神经网络预测结果更准确、精度更高。
Based on previous 500 hPa geopotential height and sea surface temperatures, a prediction model of the monthly mean rainfall in May for the central part of Guangxi is established with RBF neural network technology and principal component analysis (PCA) method. The results of the forecast experiment with 5-year samples indicate that the mean relative error is 18.12%, the root mean square error is 50.52, and the mean absolute error is 34.23. The prediction results of RBF neural network are proved to be more accurate compared with BP neural network model.