摘要
采用遗传学习算法对神经网络BP模型的初始权重进行优化,即先用遗传学习算法进行全局训练,再用BP算法进行精确训练,使网络收敛速度加快和避免局部极小。将该方法运用于洪水预报问题,并利用山西省文峪河水库的历史资料条件建立一个网络,以洪水预报的各种控制因素相关资料作为样本,对网络进行训练并用训练好的网络进行预报。网络的训练速度及预报结果表明,该算法收敛速度较快,预测精度很高,为洪水预报提供了一种新思路和新方法。
The initial weights of BP model are optimized via genetic algorithm, which means a global train based on GA and a precise train based on BP to speed up convergence and avoid local minimum. This method was utilized on flood forecast, which was based on the historical hydraulic data of wenyuhe water reservoir to train the ANN. The training speed was fast, the prediction was precise. This method supplies a new way for flood forecast.
出处
《太原理工大学学报》
CAS
北大核心
2005年第5期585-588,共4页
Journal of Taiyuan University of Technology
关键词
神经网络
遗传算法
洪水预报
neural network
genetic algorithm
flood forecast
