基于遗传-神经网络的电网流域面雨量预报方法研究

摘要

Taking the key hydropower plant and large and middle reservoirs as a major consideration,23 pow-er basin in Guangxi are divided based on the landform and physiognomy and the distribution of medium and small rivers,and a nonlinear neural network prediction method for area rainfall over power basin has been developed.For the six power basin in May and June,such as Long Tan and Long Jiang power basin,the genetic neural network prediction models for area rainfall over power basin are established,and the connection weight and structure of the BP neural network is optimized using the genetic algorithm.Results of independent samples show that the genetic neural network prediction model is better than the traditional stepwise regression method for area rainfall over power basin,and is superior to the predictions converted from Japan and Germany numerical prediction products,moreo-ver,the prediction capacity of the genetic neural network model is the same as that of the corresponding integrated area rainfall prediction products from meteorological department.Therefore,the genetic neural network model for area rainfall over power basin opens up a vast range of possibilities for operational weather prediction.%以重点水力发电厂和大中型水库为主要考量，并兼顾地形地貌和中小河流的分布特征，将广西划分为23个电网流域，研究了基于非线性的神经网络电网流域面雨量预报方法。以5-6月龙滩近库区、龙江流域等6个电网流域为例，利用遗传算法优化 BP 神经网络的连接权和网络结构，建立了各电网流域的遗传-神经网络电网流域面雨量预报模型。对独立样本的预报结果表明，基于遗传-神经网络的电网流域面雨量预报模型的预报能力要优于传统的逐步回归预报模型，也明显优于日本、德国数值模式预报产品所换算成的电网流域面雨量预报，并与气象部门同期制作的综合面雨量预报产品能力相当，因而，遗传-神经网络面雨量集合预报模型有较好的业务应用前景。