A Hybrid Prediction Framework for Water Quality with Integrated W-ARIMA-GRU and LightGBM Methods

• Published in: Water (Basel) Vol. 14; no. 9; p. 1322
• Main Authors: Zhou, Shenghan, Song, Chaofei, Zhang, Jingjing, Chang, Wenbing, Hou, Wenkui, Yang, Linchao
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.05.2022
• Subjects: Accuracy; Algae; Algorithms; Aquatic resources; ARIMA; Autoregressive models; Chemical oxygen demand; China; Climate change; Decomposition; Deep learning; Freshwater resources; GRU; Learning strategies; LightGBM; Machine learning; Medical laboratories; Methods; Neural networks; Performance evaluation; Predictions; Quality assessment; Quality control; Random variables; Regression analysis; Rivers; Surface water; Time series; Trends; Water quality; water quality prediction framework; Water supply; wavelet decomposition; Wavelet transforms; China
• ISSN: 2073-4441; 2073-4441
• DOI: 10.3390/w14091322

Water is the source of life, and in recent years, with the progress in technology, water quality data have shown explosive growth; how to use the massive amounts of data for water quality prediction services has become a new opportunity and challenge. In this paper, we use the surface water quality data of an area in Beijing collected and compiled by Zhongguancun International Medical Laboratory Certification Co., Ltd. (Beijing, China). On this basis, we decompose the original water quality indicator data series into two series in terms of trend and fluctuation; for the characteristics of the decomposed series data, we use the traditional time series prediction method to model the trend term, introduce the deep learning method to interpret the fluctuation term, and fuse the final prediction results. Compared with other models, our proposed integrated Wavelet decomposition, Autoregressive Integrated Moving Average (ARIMA) and Gated Recurrent Unit (GRU) model, which is abbreviated as the W-ARIMA-GRU model, has better prediction accuracy, stability, and robustness for three conventional water quality indicators. At the same time, this paper uses the ensemble learning model LightGBM for the prediction of water quality evaluation level, and the accuracy and F1-score reached 97.5% and 97.8%, respectively, showing very strong performance. This paper establishes a set of effective water quality prediction frameworks that can be used for timely water quality prediction and to provide a theoretical model and scientific and reasonable analysis reference for the relevant departments for advanced control.