基于小波消噪的混沌神经网络径流预报模型

The chaotic neural network model of runoff forecast based on wavelet de-noising

水文要素时间序列中的噪声不仅影响水文混沌特性识别,更影响径流预测精度。本文基于小波消噪理论对水文序列中的噪声进行了处理,并利用混沌理论中的相空间重构技术计算出饱和嵌入维数作为混沌神经网络输入层节点数,将小波技术、混沌理论和神经网络方法结合起来对汛期日径流进行了预测。与消噪前相比,模型结构从7-10-1简化到6-8-1,预测合格率从77.56%提高到85.32%,平均绝对百分比误差从12.52%减少到10.86%,由此表明水文系列中的噪声会影响预测模型的参数和精度,本文所建立的模型是值得借鉴的。

Noise in the hydrological time series influences not only the hydrology chaos identification, but also the runoff prediction precision. The wavelet technology is used to eliminate noise in the daily runoff time series. The saturated embedding dimension as the input node number of chaotic ANN model is computed by the phase restructure of chaos theory. The above gained model is used to predict the daily runoff in flood season. Compared with the model gained by the original daily runoff time series, not only the model structure is simplified from 7 - 10 - 1 to 6 - 8 - 1, but also the prediction precision is increased from 77.56 % to 85.32 % and the mean absolute percentage error is decreased from 12.52% to 10.86%. It shows that noise can influence the parameter and the precision of the prediction model. The built model is worthy to be considered for other predictions.