Integrating water quality and operation into prediction of water production in drinking water treatment plants by genetic algorithm enhanced artificial neural network

• Published in: Water research (Oxford) Vol. 164; p. 114888
• Main Authors: Zhang, Yanyang, Gao, Xiang, Smith, Kate, Inial, Goulven, Liu, Shuming, Conil, Lenny B., Pan, Bingcai
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.11.2019
• Subjects: algorithms; analysis of variance; Artificial neural network; chemical oxygen demand; China; data collection; decision making; drinking water; Drinking water treatment; electric energy consumption; Genetic algorithm; neural networks; planning; prediction; principal component analysis; statistical models; supply balance; temperature; Water production; water quality; China; Drinking water treatment; Artificial neural network; Genetic algorithm; Water production
• ISSN: 0043-1354; 1879-2448; 1879-2448
• DOI: 10.1016/j.watres.2019.114888

Stringent regulations and deteriorating source water quality could greatly influence the water production capacity of drinking water treatment plants (DWTPs). Using models to predict the performance of DWTPs under stress provides valuable information for decision making and future planning. A hybrid statistic model named HANN was established by combining artificial neural network (ANN) with genetic algorithm (GA) aiming at forecasting the overall performance of DWTPs nationwide in China. Monthly data from 45 DWTPs across China was employed. Water quality parameters like temperature and chemical oxygen demand (COD) and operational parameters like electricity consumption and chemical consumption were selected as input variables, while drinking water production was employed as the output. Both preliminary data analysis and principal component analysis (PCA) suggested a clear non-linear relationship between the input and output variables. The structure of the HANN model was optimized by employing the lowest mean squared error (MSE) as the indicator. The resultant HANN model performed well when simulating the training datasets. Its predictive accuracy for the independent test datasets was enhanced when feeding more training datasets and the performance was constantly higher than the independent multi-layered ANN models using the coefficient of determination (R2) as the indicator, indicating the HANN model was capable of capturing complex non-linear relationship and extrapolation. Results from Accuracy test, Garson sensitivity analysis and Analysis of Variance (ANOVA) suggested the quantity of water produced by DWTPs was closely linked to water quality and operational parameters. The scenario analysis showed that the HANN model was capable of predicting water production variation based on the parameter variations, indicating that the HANN model could be a general management tool for decision makers and DWTP managers to make plans in advance of regulatory changes, source water quality variations and market demand. [Display omitted]