人工神经网络模型在水质预警中的应用研究进展

水质预警模型是大数据时代构建环境智能决策与管理体系的关键技术.近年来,水质自动化监测能力的提升以及测管协同对环境模型的强烈需求,激发了研究人员探索新的建模方法并努力提高模型预测性能.其中,人工神经网络(Artificial Neural Network, ANN)模型发展迅速.本文综述了3大类ANN模型的发展历史和模型结构特点,梳理了ANN模型在水质数据软测量、数据异常检测和时间序列预测等方面的研究进展,归纳了一般建模流程、技术建议和常用的模型性能指标,发现ANN模型的应用依赖于监测数据质量,存在模型可解释性差、模型运行硬件资源要求较高等不足,提出未来水质预警模型的研发思路和重点,需要加快推进水环境监测技术与预警模型的协同发展和业务化应用,通过多种应用场景检验实现技术迭代,形成大数据驱动的水质在线监测-智能预警-应急管理支撑体系,助力我国环境治理能力现代化.

Artificial neural network models for water quality early warning: A review

Water quality early warning models are a key component of intelligent environmental decision-making and management systems in the era of big data. In recent years, the increasing demand for early warning of water quality deterioration has stimulated researchers to develop new modeling approaches and improve prediction reliability, and artificial neural network (ANN) models are developing rapidly. In this paper we review the development history of three group ANN model and model structure characteristics. The research progress of ANN models for the purpose of soft measurement, data quality control and time series prediction of water quality are summarized. We summarized the general modeling procedure, technical recommendations, and performance indexes that are commonly used. We found that the application of ANN models has been limited by the poor quality of measured data, weak interpretability of model outputs and the substantial requirements in terms of hardware and computing resources. We emphasize that future efforts should be made to develop and apply early warning models in the field of water quality prediction. There is an urgent need to promote the coordinated development of innovative technologies for environmental monitoring and early warning, through constant validation and upgrading of models after their application in a variety of situations. The long-term goal is to form an online water quality monitoring system, incorporating intelligent early warning and emergency management, driven by big data, to support environmental governance.