凌汛开河历时预测模型研究

Study on Prediction Model for the Breaking up Date of Ice Flood

将小波函数作为BP神经网络隐含层节点的传递函数,提出了预测冰凌开河历时的小波神经网络模型。实例应用结果表明:该模型具有收敛速度快、预测精度高等特点,克服了单一BP神经网络在预测中极易陷入局部极小点等缺点,模型预测结果更接近实际值,其相对平均误差比单一BP神经网络和模糊神经网络分别降低了2.24%和4.83%。

A wavelet neural network model was established with the wavelet function as the transfer function between the panel points in the hidden layer of tire BP neural network. Proved by application, the model had high convergence rate and high prediction precision, and it also overcame the shortcomings of the stogie BP neural network such as it could easily fall into local minimum points. The prediction resnh of the model is more close to the actual value, the relative mean error decreases by 2.24% and 4.83% respectively comparing with the single BP neural network and fuzzy neural network.