Lai JianWun
School of Engineering, University of California San Diego, San Diego, CA, 92093, USA.
Sci Rep. 2025 Jun 2;15(1):19257. doi: 10.1038/s41598-025-03612-5.
Research is implemented to protect the environment from an epidemic of chemical materials that could render living conditions hazardous. In order to efficiently use productivity while maintaining a constant and reliable level of waste quality, severe regulations regarding Waste-Water Treatment and Control Systems (WWTCS) must be adopted to mitigate the serious nature of water pollution and impure performance. Suboptimal treatment efficiency and use of resources are the results of the methods used for WWTCS, which are not highly susceptible to changing impact features and complex biological systems. The present study presented a prediction algorithm and an Integrated Control System (ICS) to address the problems of conventional methods. This research proposes a Deep Learning (DL) for the quality of wastewater prediction that employs a Quantile Regression-Random Forest (QR-RF) meta-learner when combined with Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). The proposed method has been implemented into practice and tested at Asia's Jiangsu Province Metropolitan Waste-Water Treatment Plant (WWTP). With a Root Mean Absolute Error (RMSE) of 4.76 mg/L for 24-h horizons and a Mean Absolute Error (MAE) of 0.85 mg/L for 1-h predictions, the proposed model outperforms conventional methods in terms of prediction accuracy. The ICS is superior to standard WWTCS by a vital error boundary, minimizing energy consumption by 17% and boosting chemical-based consumption optimization by 24%. With an average removal rate of 94.23% for Chemical Oxygen Demand (COD) compared to 88.76% for standard systems, the findings from experiments exhibited significant performance improvements.
开展研究是为了保护环境免受可能使生活条件变得危险的化学物质流行的影响。为了在保持废物质量恒定可靠水平的同时有效利用生产力，必须采用关于废水处理与控制系统（WWTCS）的严格规定，以减轻水污染的严重性和不纯性能。WWTCS所采用方法的结果是处理效率欠佳和资源利用不合理，这些方法对不断变化的影响特征和复杂生物系统不太敏感。本研究提出了一种预测算法和一种综合控制系统（ICS）来解决传统方法的问题。这项研究提出了一种用于废水质量预测的深度学习（DL）方法，当与卷积神经网络（CNN）、长短期记忆网络（LSTM）和门控循环单元（GRU）相结合时，采用分位数回归 - 随机森林（QR - RF）元学习器。所提出的方法已在实践中实施，并在亚洲江苏省城市污水处理厂（WWTP）进行了测试。对于24小时预测，均方根误差（RMSE）为4.76毫克/升，对于1小时预测，平均绝对误差（MAE）为0.85毫克/升，所提出的模型在预测准确性方面优于传统方法。ICS在关键误差边界方面优于标准WWTCS，能耗降低了17%，基于化学物质的消耗优化提高了24%。与标准系统的88.76%相比，化学需氧量（COD）的平均去除率为94.23%，实验结果显示出显著的性能提升。
