Abstract:

Traditional water quality monitoring methods are time-consuming, costly, and have poor data timeliness, leading to long feedback and adjustment cycles in wastewater treatment systems. With the rapid advancements in artificial intelligence, developing data-driven water quality prediction technologies is of significant importance. This study focused on the collection and processing of big data, the characteristics and applications of water quality data collection, cleaning strategies, and feature engineering methods were summarized. Based on this, the predictive performance and characteristics of different types of water quality prediction models were introduced. Statistical regression models, machine learning models, and deep learning models all demonstrated certain advantages. However, significant differences in data quality across different datasets made it difficult to obtain universal prediction models. By considering the features of big data and data quality, applying reasonable data preprocessing techniques, and utilizing various prediction methods or combinations, the accuracy of model predictions could be significantly improved. Finally, this review summarized the current state of water quality prediction models, existing challenges, and future development directions, aiming to provide a reference for the research, development, and application of water quality prediction models.

Key words: water quality prediction, feature engineering, machine learning, deep learning, ensemble learning

摘要：

传统水质监测手段耗时长、成本高且数据时效性差，污水处理系统参数反馈和调整周期长，在人工智能迅猛发展的背景下，构建基于数据驱动的水质预测技术有重要意义。从大数据收集与处理层面出发，梳理了国内外水质数据收集、清洗策略以及特征工程等方法的特点和应用状况。在此基础上介绍了不同类型水质预测模型的预测效果与特点，统计回归模型、机器学习模型和深度学习模型都展现出一定的优势，但不同数据集质量上存在显著差异，难以获得普适的预测模型。结合大数据特征和数据集质量，采取合理的数据预处理手段，应用不同类型的预测方法或组合，可以显著提高模型预测准确性。最后综述了现阶段水质预测模型的应用现状、存在的问题以及未来发展方向，以期为水质预测模型研究、开发与应用提供参考。

关键词: 水质预测, 特征工程, 机器学习, 深度学习, 集成学习

CLC Number: 

• X703