改进GA-RBF神经网络的水厂混凝投药预测

Improved GA-RBF Neural Network for Predicting Coagulant Dosing in Waterworks

为了提高水厂混凝剂投加量预测准确性,针对投药系统易受多种水质因素影响,且投药后净水过程存在高度非线性的特点,通过改进遗传算法(GA)优化径向基函数神经网络(也称为RBF神经网络)的权值ω_i和高斯基函数中心宽度向量σ_i,构建GA-RBF神经网络净水厂投药量预测模型。Matlab仿真结果表明,GA-RBF神经网络预测模型可通过实现全局逼近来回避极值陷阱,提高了稳定性和全局寻优能力,相较于单一RBF神经网络预测模型,GA-RBF神经网络预测模型的拟合优度提高5.474%,平均绝对误差降低了4.14%,根均方误差降低3.392%,迭代速度和预测精度都有所提高,数据拟合能力更强。

Aiming to improve the accuracy of prediction of coagulant dosage in waterworks,with the dosing process had highly nonlinear and influence about a variety of water quality factors.The weightωi and the central width vectorσi of the basis function of the Radial Basis Function(RBF)neural network are optimized by improving Genetic Algorithm(GA),construction of GA-RBF neural network prediction model for chemical dosage of waterworks.The simulation results by Matlab showed that the GA-RBF neural network prediction model could avoid the extreme value trap by implementing global approximation,and improved the stability and global optimization ability.Compared with the single RBF neural network prediction model,the R2 of the GA-RBF neural network prediction model increased by 5.474%,the MAE decreased by 4.14%,and the RMSE decreased by 3.392%.The iteration speed and prediction accuracy were improved,and the data fitting ability was stronger.