遗传算法优化的BP神经网络预测水厂絮凝剂投加量

Prediction of Flocculant Dosage in WTP with BP Neural Network Optimized by Genetic Algorithm

文中选取了太湖地区某水厂原水水质及过程工艺水质与絮凝剂聚合氯化铝(PAC)投加量的360条原始数据样本,采用异常值剔除、空缺值的填补及偏态分布纠正,对该样本进行了预处理及数据清洗从而得到原始数据集。建模前使用了相关系数法分析评价了各水质指标对絮凝剂PAC投加量的影响程度,根据分析评价结果以及混凝基础理论,选取了对絮凝剂PAC投加量影响较大的部分水质指标作为建立投加量模型的建模数据集,其中80%作为训练集,20%作为验证集检验模型泛化能力。模型采用BP神经网络,使用遗传算法对网络的结构及关键参数进行了优化,最后使用最优模型对絮凝剂PAC投加量进行预测。模型在验证集72个样本上的平均绝对误差(MAE)为3.78 mg/L;训练集288个样本平均绝对误差为2.75 mg/L,结果表明模型能有效拟合絮凝剂投加量变化趋势,具有一定的参考价值。

This paper selected 360 original samples of raw water quality and process water quality and polyaluminum chloride(PAC)dosages of a water treatment plant(WTP)in the Taihu Lake area.The data of those samples was preprocessed using outlier elimination,vacancy value filling,and skew distribution correction methods.Before modeling,the correlation coefficient method was used to analyze and evaluate the influence of each water quality index on the PAC dosage.According to the analysis and evaluation results and the theory of coagulation,some water quality indices that have a greater impact on the PAC dosage were selected to establish model.In the modeling data set,80%was used as the training set and 20%was used as the validation set to test the generalization ability of the model.The model uses BP neural network,using genetic algorithm to optimize the structure and key parameters of the network.Then the optimal model was used to predict the PAC dosages.The results showed that the average absolute error(MAE)of the model on 72 samples in the validation set was 3.78 mg/L;the mean absolute error of 288 samples in the training set was 2.75 mg/L,indicating that the model can effectively fit the trend of alum consumption change,and exhibits certain reference value for water treatment.