结合均生函数的神经网络在中长期水文预报中的应用

针对中长期径流预报在水库中长期运行方案制定及调度决策形成中的作用,基于传统和智能预报方法各自的优势,利用均生函数模型记忆时间序列的内在规律,采用偏最小二乘方法对预报因子进行降维处理,建立了结合均生函数的神经网络预报模型,并利用神经网络模型修正预报结果。实例计算表明,该模型不仅可提取径流序列的特征,且预报精度也较单一的均生函数模型和神经网络模型有所提高。

Application of Neural Network Combined with Mean Generation Function to Mid Long Term Hydrological Forecasting

Mid long term runoff forecasting contributes to the establishment of reservoirs operation scheme of mid long term and the formulation of dispatching decision. Based on the advantages of traditional and intelligent methods, artificial neural network model combined with mean generation function is set up to forecast mid long term runoff. Inherent law of time series is memorized with mean generation function. And then the partial least squares method is used to reduce the dimension of forecasting factors. Finally, artificial neural network model is applied to correct predicted result. Application example shows that the model can extract the characteristics of runoff series, and the forecasting accuracy is higher than that of single mean generation function model and neural network model as well.