基于卷积层-注意力机制的长短期记忆网络出水氨氮浓度预测方法  

A prediction method for effluent ammonia nitrogen concentration based on convolutional layer and attention mechanism long short-term memory network

作　　者：张勇 赵景波 权利敏 ZHANG Yong;ZHAO Jingbo;QUAN Limin(School of Information and Control Engineering,Qingdao University of Technology,Qingdao 266520,Shandong,China)

机构地区：[1]青岛理工大学信息与控制工程学院,山东青岛266520

出　　处：《化工学报》2024年第12期4679-4688,共10页CIESC Journal

基　　金：山东省重点研发计划(软科学项目)重大项目(2023RZA02017);青岛市科技惠民计划项目(22-3-7-xdny-18-nsh)。

摘　　要：为解决城市污水处理过程出水氨氮浓度难以实时精准测量的问题,构建了一种融合卷积层和注意力机制的长短期记忆网络(convolutional layer and squeeze-and-excitation attention mechanism based long short-term memory network,CSA-LSTM)模型。首先,通过引入卷积层(convolutional layer,CL)深度提取数据中的非线性特征,并通过注意力机制(squeeze-and-excitation attention mechanism,SEAM)自适应分配特征通道的权重,实现特征解耦;其次,长短期记忆网络(long short-term memory network,LSTM)提取时间序列数据长期依赖关系,实现出水氨氮浓度的实时预测;然后,提出一种具有自适应采集函数的贝叶斯优化算法,实现模型参数优化,进一步提高模型精度;最后,基于基准实验和实际污水处理厂数据测试CSA-LSTM的有效性。结果表明,模型具有较高的出水氨氮浓度预测精度,能够有效应对城市污水处理中数据的强非线性、耦合性以及时间依赖性问题,具有良好的泛化能力。To address the issue of accurately measuring the effluent ammonia nitrogen concentration in real-time during urban wastewater treatment processes,this paper constructs a convolutional layer(CL)and squeeze-and-excitation attention mechanism(SEAM)based long short-term memory network(CSA-LSTM)model.First,by introducing the CL,nonlinear features within the data are deeply extracted.The SEAM adaptively allocates weights to feature channels,achieving feature decoupling.Secondly,the long short-term memory network(LSTM)extracts long-term dependencies in time-series data to realize real-time prediction of effluent ammonia nitrogen concentration.Then,a Bayesian optimization algorithm with an adaptive acquisition function is proposed to optimize model parameters,further enhancing model accuracy.Finally,the effectiveness of CSA-LSTM is tested based on benchmark experiments and actual wastewater treatment plant(WWTP)data.The results show that the model has high predictive accuracy for effluent ammonia nitrogen concentration,can effectively handle the strong nonlinearity,coupling,and time-dependency of data in urban sewage treatment,and has good generalization ability.

关 键 词：城市污水处理 氨氮浓度预测 神经网络 特征提取 优化 算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]