基于GA-FCM-ANFIS的不稳定水质条件下混凝投药预测方法  

Prediction of Coagulation Dosing Under Unstable Water Quality Conditions Based on GA-FCM-ANFIS

作　　者：郭喜峰[1] 孟铭 宁一 肖乐 GUO Xi-feng;MENG Ming;NING Yi;XIAO Le(School of Electrical and Control Engineering,Shenyang Jianzhu University,Shenyang 110168,Liaoning Province,China;School of Control Science and Engineering,Dalian University of Technology,Dalian 116024,Liaoning Province,China)

机构地区：[1]沈阳建筑大学电气与控制工程学院,辽宁沈阳110168 [2]大连理工大学控制科学与工程学院,辽宁大连116024

出　　处：《中国农村水利水电》2024年第12期60-66,共7页China Rural Water and Hydropower

基　　金：中国高校产学研创新基金项目(2022IT039);辽宁省教育厅基本科研项目(LJKQZ20222276)。

摘　　要：净水厂实际工艺过程中水质复杂多变且存在水质检测仪器故障导致水质参数缺失问题,确定混凝剂的投加量面临挑战。烧杯搅拌实验或人工控制过量投药的传统方法容易带来金属离子超标等水质风险,危害用户健康并降低经济效益。为解决上述问题,将遗传算法优化的模糊C均值聚类(GA-FCM)与自适应模糊推理系统(ANFIS)相结合构建预测模型。该模型依靠模糊规则带来的推理能力和神经网络的自适应能力实现在不稳定水质情况下混凝剂投药量的精准预测。通过Pearson相关系数法筛选输入水质指标,提高模型运行效率和泛化能力;使用遗传算法确定FCM聚类的最佳参数,提升聚类效果并生成更精准的模糊规则;通过反向传播和岭回归混合的方法调整网络参数,处理非线性部分的同时优化后件网络的权重。将中国东北部某净水厂作为案例进行研究,实验结果表明对比传统线性回归和机器学习模型,GA-FCM-ANFIS在不稳定水质情况下拥有更好的预测性能。对比其他耦合模型,在参数缺失情况下GA-FCM-AN⁃FIS的预测精度明显更高。水质突变情况下预测评估指标R2、RMSE、MAE和MAPE分别为0.93、2.79、2.33和3.71%;参数缺失情况下R2、RMSE、MAE和MAPE分别为0.89、3.07、2.57和4.06%。所提模型能够显著提高预测精度,作为确定混凝剂投加量的可靠方法。In the practical processes at water purification plants,the water quality is complex and variable,and there are instances of miss⁃ing water quality parameters due to malfunctioning detection instruments,thus bringing challenges to determining the dosage of coagulants.Traditional methods,such as beaker stirring tests or manual control of excessive dosage,can easily bring water quality risks such as exces⁃sive metal ions,posing health risks to users and reducing economic efficiency.To address these issues,a predictive model combining Genetic Algorithm optimized Fuzzy C-Means clustering(GA-FCM)and Adaptive Neuro-fuzzy Inference System(ANFIS)was developed.This model leverages the inference capability of fuzzy rules and the adaptive capability of neural networks to accurately predict coagulant dosage under unstable water quality conditions.Pearson correlation coefficient method was used to filter input water quality indicators,enhancing model efficiency and generalization capability.The genetic algorithm was employed to determine the optimal parameters for FCM clustering,improving clustering outcomes and generating more accurate fuzzy rules.A hybrid method combining backpropagation and ridge regression was used to adjust network parameters,addressing nonlinear components while optimizing the weights of the consequent network.A water pu⁃rification plant in Northeast China was used as a case study.Experimental results indicated that compared to traditional linear regression and machine learning models,GA-FCM-ANFIS exhibited superior predictive performance under unstable water quality conditions.Further⁃more,in the absence of certain parameters,GA-FCM-ANFIS achieved significantly higher predictive accuracy than other coupled models.The prediction evaluation metrics R2,RMSE,MAE,and MAPE under sudden water quality changes were 0.93,2.79,2.33,and 3.71%,respectively;under missing parameters,the metrics were 0.89,3.07,2.57,and 4.06%,respectively.The proposed model significantly improves prediction accuracy and serves

关 键 词：混凝投药 水质突变 参数缺失 模糊系统 神经网络 

分 类 号：TU991.22[建筑科学—市政工程]