摘要
洪水预报对防洪减灾,拦洪储存水资源等具有重要意义。对用遗传算法(GA)来改进神经网络的算法进行了分析,建立了用GA动态寻求权重的前馈网络模型(I)和用GA优化初始权重的前馈网络模型(简称模型Ⅱ)。借助于Matlab中的神经网络包和遗传算法包,编制了改进模型的计算机程序,对湖北省黄龙滩水库14场洪峰和洪水总量进行了模拟,用另外5场洪水检验了改进的模型,并与传统的前馈神经网络模型进行了比较。结果表明:用模型I对洪峰流量的检验效果明显优于模型Ⅱ和传统的前馈神经网络模型;模型Ⅱ对中低流量及洪水总量的检验效果相对较好。因此,本次改进的模型可以用于洪水预报。
The paper analyzes all kinds of methods of neural network improved by genetic Algorithm, and proposes two ANN models, among which model Ⅰ has the best weights dynamically searched by GA and model Ⅱ uses GA to optimize the initial weights. All models are achieved by using madab language. Moreover, it takes the Huanglongtan reservoir as an example to verify the result of the traditional ANN model as well as the other two proposed. The result shows that Model Ⅰ works best in forecasting the peak-flood amount, while Model Ⅱ is best in forecasting middle, low and the total flood amount. In the end, it is suggested that the above two models be coupled together to achieve a high precision in flood forecasting by making use of the relative advantages of each other.
出处
《水力发电》
北大核心
2005年第9期12-15,共4页
Water Power
关键词
人工神经网络
遗传算法
权重
洪水预报
artificial neural network
genetic algorithm
weight
flood forecasting
