基于GA-LMBP神经网络模型的大坝渗流压力预报分析

Dam Seepage Pressure Forecasting and Analysis Based on GA-LMBP Neural Network Model

为克服传统BP神经网络在渗流压力预测过程中收敛慢、计算量大和易陷入局部极小等缺陷,依据渗流压力的影响因素,研究了模型的结构和输入输出因子,建立了基于遗传算法和LM算法相结合的GA-LMBP神经网络的大坝渗流压力预测模型,即通过遗传算法(GA)的选择、交叉和变异操作得到BP网络的一组全局最优近似解(即网络的初始权值和阈值),再以该近似解为初值,利用LM算法对BP网络进行优化训练,将训练好的网络用于渗流压力的预测。实例应用结果表明,在相同精度的要求下,GA-LMBP神经网络模型收敛速度快、预测精度高,对大坝渗流压力的预测效果更佳,是值得采用的一种模型。

In order to overcome the slow convergence speed,large computing capacity and easy converge on local minimum point of traditional BP neural network in the process of seepage pressure prediction,according to the influencing factors of seepage pressure,a GA-LMBP neural network prediction model of dam seepage pressure based genetic algorithm and LM algorithm was established.Its structure,input factor and output factor were studied.A set of global optimal approximate solution（initial weights and thresholds of BP network）was got through the operations of selection,crossover and mutation of genetic algorithm.Then,BP network was optimized by LM algorithm with the approximate solution as initial values,and the trained BP network was used for the prediction of seepage pressure.The application results show that under the same accuracy requirement,GA-LMBP model is characterized by rapid convergence speed and high prediction accuracy,which is better and worth being adopted in dam seepage pressure prediction.