基于PSO-BP神经网络的水厂智能消毒预测模型  

Research on Intelligent Disinfection Prediction Model of Waterworks Based on PSO-BP Neural Network

作　　者：杨存满 鞠佳伟 袁芳 李晓尚 兰华春 YANG Cun-man;JU Jia-wei;YUAN Fang;LI Xiao-shang;LAN Hua-chun(Research Institute for Environmental InnovationTsinghua,Suzhou 215000.China;General Water of China Co.Ltd.,Beijing 100022,China;Xiangtan Zhonghuan Water Affairs Co.Ltd.,Xiangtan 411100.China;School of Environment,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华苏州环境创新研究院,江苏苏州215000 [2]中环保水务投资有限公司,北京100022 [3]湘潭中环水务有限公司,湖南湘潭411100 [4]清华大学环境学院,北京100084

出　　处：《中国给水排水》2022年第3期57-61,共5页China Water & Wastewater

基　　金：中国节能环保集团有限公司重大科技创新项目(cecep-zdkj-2020-005)。

摘　　要：为解决水厂运行过程中粗犷式、经验式投加氯消毒剂的问题,建立基于PSO-BP神经网络的水厂智能消毒预测模型。选取流量、矾耗、水质参数作为预测模型的输入参数,利用粒子群算法优化神经网络权值和阈值,模型评价指标平均绝对百分比误差(MAPE)、均方根误差(RMSE)都低于传统BP神经网络模型,其中RMSE值下降207 kg,MAPE值下降1.80%,相对标准偏差(RSD)下降了2.4%,有效提高了模型预测的准确性和稳定性,并在实际应用过程中有助于降低水厂氯消毒剂药耗,平均可节约生产成本约1756元/d。可见,PSO-BP智能消毒预测模型是合理、可行的,为城市自来水厂加氯量预测提供了一种简单可行的思路和方法。In order to solve the problem of rough and empirical dosing of chlorination disinfectant in the operation of waterworks, an intelligent disinfection prediction model of waterworks based on PSO-BP neural network was established. The flow, alum consumption and water quality parameters were selected as the input parameters of the prediction model, and the neural network weight and threshold were optimized by particle swarm optimization algorithm. The average absolute percentage error(MAPE) and root mean square error(RMSE) of the evaluation indexes of the model were lower than those of the traditional BP neural network model, particularly RMSE decreased by 207 kg, MAPE decreased by1.80%, and relative standard deviation(RSD) decreased by 2.4%, which could effectively improve the accuracy and stability of the model prediction. In the process of practical application, it was helpful to reduce the consumption of chlorination disinfectant and the production cost could be saved by 1 756 yuan/d, averagely. It can be seen that the PSO-BP intelligent disinfection prediction model is reasonable and feasible, which provides a simple and feasible idea and method for the prediction of chlorine addition in the disinfection system of urban waterworks.

关 键 词：自来水厂 加氯消毒 PSO-BP神经网络 BP神经网络 预测 

分 类 号：TU991[建筑科学—市政工程]