工业水处理 ›› 2025, Vol. 45 ›› Issue (6): 159-166. doi: 10.19965/j.cnki.iwt.2024-0428

• 试验研究 • 上一篇    

基于PSO-LSTM-SATN模型的污水水质预测研究

杨潞霞1(), 王智瑜1, 沈帅杰1, 马永杰2, 付一政3,4()   

  1. 1. 太原师范学院计算机科学与技术学院,山西 晋中 030619
    2. 西北师范大学物理与电子工程学院,甘肃 兰州 730070
    3. 中北大学材料科学与工程学院,山西 太原 030051
    4. 山西省煤矿矿井水处理技术创新中心,山西 太原 030006
  • 收稿日期:2024-10-10 出版日期:2025-06-20 发布日期:2025-06-19
  • 通讯作者: 付一政
  • 作者简介:

    杨潞霞(1979— ),博士,教授,E-mail:

  • 基金资助:
    国家自然科学基金项目(62066041); 山西省重点研发计划项目(202102010101008)

Research on sewage water quality prediction based on the PSO-LSTM-SATN model

Luxia YANG1(), Zhiyu WANG1, Shuaijie SHEN1, Yongjie MA2, Yizheng FU3,4()   

  1. 1. College of Computer Science and Technology, Taiyuan Normal University, Jinzhong 030619, China
    2. School of Physics and Electronic Engineering, Northwest Normal University, Lanzhou 730070, China
    3. School of Materials Science and Engineering, North University of China, Taiyuan 030051, China
    4. Shanxi Coal Mine Water Treatment Technology Innovation Center, Taiyuan 030006, China
  • Received:2024-10-10 Online:2025-06-20 Published:2025-06-19
  • Contact: Yizheng FU

摘要:

为解决工业废水处理领域进水水质波动性大、随机性强、不具有周期性导致无法精准预测其水质的问题,提出粒子群优化算法(Particle Swarm Optimization,PSO)-长短期记忆网络模型(Long Short-Term Memory,LSTM)-自注意力机制(Self-Attention,SATN)污水水质预测模型。以山西省某煤炭水处理工厂7 357组历史污水水质数据为基础,首先采用LSTM捕获进水水质中COD数据的长期依赖关系,然后采用SATN解决水质信息分布不均匀的问题,最后结合PSO对LSTM-SATN模型进行优化,帮助网络自动获取最佳参数和模型配置。评价结果显示,模型均方误差(Mean Square Error,MSE)、平均绝对误差(Mean Absolute Error,MAE)和平均绝对百分比误差(Mean Absolute Percentage Error,MAPE)分别为0.528 4 (mg/L)2、0.236 9 mg/L和4.127 7%,与LSTM、门控循环单元结构(Gated Recurrent Unit,GRU)、双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)相比,MSE、MAE、MAPE均有大幅降低,即该PSO-LSTM-SATN模型能够更准确地预测进水水质,为工厂日常运营管理方案提供合理的指导意见。

关键词: 污水水质预测, 长短期记忆网络模型, 粒子群优化算法, 自注意力机制

Abstract:

In order to solve the problem of large fluctuations, strong randomness, and lack of periodicity in the inflow water quality of industrial wastewater treatment, Particle Swarm Optimization(PSO)-Long Short-Term Memory (LSTM)-Self-Attention(SATN) sewage quality prediction model was proposed. Based on 7 357 sets of historical wastewater quality data from a coal water treatment plant in Shanxi Province, LSTM was firstly used to capture the long-term dependence relationship of COD datas in the influent water quality. Then SATN was used to solve the problem of uneven distribution of water quality information. Finally, PSO was combined to optimize the LSTM-SATN model, which helped the network automatically to obtain the best parameters and model configuration. The results showed that the Mean Square Error(MSE) of the model was 0.528 4 (mg/L)2, the Mean Absolute Error(MAE) was 0.236 9 mg/L, and the Mean Absolute Percentage Error(MAPE) was 4.127 7%. Compared with the LSTM, Gated Recurrent Unit(GRU) and Bidirectional Long Short-Term Memory Network(BiLSTM), the MSE, MAE, and MAPE of the proposed model were significantly reduced. The PSO-LSTM-SATN model could more accurately predict the inflow water quality and provide reasonable guidance for the daily operation and management plan of the factory.

Key words: sewage water quality prediction, Long Short-Term Memory, Particle Swarm Optimizatio, Self-Attention

中图分类号: