小波人工神经网络在径流预报中的应用

Wavelet Neural Network Model and Its Application in Runoff Forecasting

针对径流时间序列的非线性和多时间尺度特性,应用A Trous算法对盘石头水库的月径流序列进行了分析。在此基础上,将小波分析与人工神经网络相结合,建立了组合预测模型,并给出构造模型的一般步骤及关键算法。针对一般BP算法收敛速度慢、易陷入局部极小值的缺陷,提出了基于改进共轭梯度法的BP算法。实践表明:基于小波分析的人工神经网络模型在月径流模拟过程中具有很好的仿真能力,训练后的模型具有较高的精度。

According to the nonlinear and the multi-time scale character of the monthly runoff time series,the A Trous Algorithm was used to analyze the monthly runoff time series of Panshitou Reservoir.Based on this,the combination forecasting model was built by combining the wavelet and artificial neural network,and the general steps and key algorithm of the model were expounded.Dealing with the defects of the steepest descent and slowly converging,easy to immerging in partial minimum frequently,the improvement was made of conjugating gradient with BP algorithm to solve the problem.The results show that the model has better capability of simulation for the process of monthly runoff with higher accuracy.