基于RBF神经网络的余氯浓度预测模型研究

摘要

余氯浓度是衡量供水管网水质的一个重要指标,采用混沌理论、模型校正等传统方法不能准确反映余氯浓度变化规律.根据RBF神经网络快速收敛和全局优化的特点,基于时间序列法,建立RBF神经网络余氯浓度预测模型.采用Matlab中的Newrbe函数进行函数逼近,结合某管网水质模拟控制系统提供的样本数据进行仿真计算,最终获得的余氯浓度预测值和实测值十分吻合.结果表明:RBF神经网络预测模型具有一定的工程实用价值.%The concentration of residual chlorine is an important parameter indicating the quality of water in water distribution network, but the traditional methods, such as chaos theory and model calibration, cannot reflect the variation rule of residual chlorine thoroughly. A prediction model of residual chlorine is established, based on the fast convergence and global optimization of RBF neural network and time series. The Newrbe function in Matlab is adopted to realize function approximation, combining with the sample data provided by the water quality control system of water distribution network to run simulation. The difference between predicted residual chlorine fit actual ones,which means this model has practical value.