基于CNN-LSTM混合神经网络的炼化污水处理场COD排放浓度预测

Prediction of COD Emission Concentration of Petroleum Refining Waste Water Treatment Plant Based on CNN-LSTM Hybrid Neural Network

何为 ¹岳留强 ²唐智和 ¹栾辉 ¹陈昌照 ²王若尧¹

扫码查看

作者信息

1. 中国石油集团安全环保技术研究院有限公司 QHSE检测中心,北京 102206
2. 中国石油天然气集团有限公司质量健康安全环保部,北京 100007
折叠

摘要

快速、准确预测含油污水中有机污染物的化学需氧量(COD),及时优化运行参数,对于石油炼化企业节能减排、污水达标排放极为重要和迫切.为此,以某炼化企业污水处理场生产要素和COD排放浓度数据为研究样本,提出一种增加先验知识的CNN-LSTM混合神经网络算法,建立炼化企业污水处理场COD排放浓度预测模型.结果表明,该模型既可发挥CNN较好刻画、提取局部特征信息的优势,又具有LSTM对连续时间序列数据较好继承性的特点,其均方根误差为22.321 7、决定系数为0.873 2,平均相对误差较LSTM网络构建的模型降低5.45%.

Abstract

It is important and urgent for petroleum refining enterprises to quickly and accurately predict the chemical oxygen demand(COD)of organic pollutants.To this end,a CNN-LSTMhybrid neural network algorithm with prior knowledge is proposed,a COD dis-charge concentration prediction model for refinery waste water treatment plants is established by using production factors and COD dis-charge concentration data from a refinery waste water treatment plant as research samples.The results show that the model can not only give full play to the advantages of CNN in better characterizing and extracting local feature information,but also has the characteristics of LSTM's good inheritance of continuous time series data.Its root mean square error is 22.321 7,the determination coefficient is 0.873 2,and the average relative error is reduced by 5.45%compared to the model constructed by LSTMnetwork.

关键词

炼化污水处理/混合神经网络(CNN-LSTM)/COD浓度/污染排放预测

Key words

refining waste water treatment/hybrid neural network/COD concentration/pollution emission prediction

引用本文复制引用

出版年

2025

西安石油大学学报(自然科学版)

西安石油大学

西安石油大学学报(自然科学版)

北大核心

影响因子：0.788

ISSN：1673-064X

段落导航