基于优化机器学习的炼化企业污水场均质池出水水质预测研究

Research on the prediction of effluent quality of the homogenization tank in the refinery sewage treatment plant based on optimized machine learning

炼化企业生产工艺流程复杂且装置繁多,炼化污水水质水量波动大,污染物组成复杂,导致下游污水处理系统频繁受到冲击,难以及时响应调控。以华南某炼化企业污水处理系统均质池出水真实水质数据为基础,对水质参数进行相关性分析和特征降维,分别构建了多参数水质预测模型和时间序列水质预测模型。研究结果表明:电导率(Cond)与COD存在一定相关性(PCC和SCC分别为0.493和0.513),水质参数与COD相关性排序为Cond>pH>NH_(3)-N>TP;SVR和BP-ANN多参数预测模型均未取得理想的预测效果,决定系数R^(2)均低于0.5;SVR和BP-ANN时间序列模型预测准确率较多参数模型大幅提高,决定系数R~2平均提升45%,均高于0.7,预测值与实测值拟合度高;模型现场验证结果表明,当上游污水水质发生波动时,模型对水质波动趋势预测较为准确,可以有效的指导现场对工艺参数进行调控。

The process flow of petrochemical industries is complex and there are many production devices.The water quality and quantity of the petrochemical wastewaters treatment system fluctuate greatly,making it difficult to respond to regulation in a timely manner,resulting in frequent impacts on downstream wastewater treatment systems.This study is based on real water quality data of homogeneous pool effluent,conduct correlation analysis and feature dimensionality reduction on water quality parameters,and constructs a multi parameter water quality prediction model and a time series water quality prediction model.The research results indicate that there is a certain correlation between conductivity(Cond)and COD(PCC and SCC are 0.493 and 0.513,respectively),and the correlation order between various water quality parameters and COD is Cond>pH>NH3-N>TP.SVR and BP-ANN multi parameter prediction models did not achieve ideal prediction results,with determination coefficients R^(2) both below 0.5;The SVR and BP-ANN time series models have higher prediction accuracy,and the parameter model has significantly improved.The coefficient of determination R2 has increased by an average of 45%,both higher than 0.7,the predicted values are consistent with the measured values and have a high degree of fit;The on-site verification results of the model indicate that when the upstream wastewaters quality fluctuates,model can accurately predict the trend of water quality fluctuations and effectively guide the on-site regulation of process parameters.