基于水厂大数据的混凝投药系统智能模型的构建

Modeling of Big Data-Based Inteligent Model for Coagulation Dosing System in WTP

为实现混凝投药的精准控制,以大型水厂粗粒度数据和细粒度数据为基础,采用箱线图和滑动均值异常数据识别技术的预处理方法,对长短期记忆网络、支持向量机、随机森林、XGBoost等算法建立混凝投药模型的效果进行了评估。研究发现:基于A、B水厂的大数据,采用XGBoost算法建立的混凝投药模型效果均最优,对聚合氯化铝(PAC)和氯化铁(FeCl_(3))的建模效果评估MAPE分别为3.42和3.72;采用箱线图结合移动平滑的技术对异常值进行处理对建模效果大幅提升;将一种药剂的投加量作为另一种药剂预测的特征值输入的方法,对双药投加模型的预测效果提升有限;原水时均水量、浑浊度、温度、历史加药数据对混凝剂投加量的准确预测有重要影响。

In order to realize the accurate control of coagulant dosing,based on coarse-grained data and fine-grained data of large-scale water treatment plant(WTP),the effects of long short-term memory network,support vector machine,random forest,XGBoost and other algorithms to establish coagulation dosing model were evaluated by using box plot and moving average abnormal data recognition technology.It was found that based on the big data of A and B WTP,the coagulation dosing model established by XGBoost was the best,and the evaluation MAPE value of modeling effect of polyaluminum chloride and ferric chloride were 3.42 and 3.72 respectively.Abnormal value was processed by box diagram combined with moving smooth technology,which greatly improves modeling effect,for the method of inputting the dosage of one agent as the agent of the prediction of another,the prediction effect of two drug dosing model was limited.Hourly average water volume of raw water,raw water turbidity,raw water temperature,and historical dosing data had important influence on the accurate prediction of coagulant dosage.