Comparison of EEMD-ARIMA, EEMD-BP and EEMD-SVM algorithms for predicting the hourly urban water consumption. Issue 3 (29th March 2022)

• Record Type: Journal Article
• Authors: Liu, Xingpo; Zhang, Yiqing; Zhang, Qichen
• Is Part Of: Journal of hydroinformatics. Volume 24:Issue 3(2022)
• Page Start: 535
• Page End: 558
• Publication Date: 2022-03-29
• Subjects: autoregressive integrated moving average (ARIMA) -- back-propagation (BP) neural network -- ensemble empirical mode decomposition (EEMD) -- short-term water consumption prediction -- support vector machine (SVM); Hydrology -- Data processing -- Periodicals; Geographic information systems -- Periodicals; Geographic information systems; Hydrology -- Data processing; Electronic journals; Periodicals; 551.480285
• DOI: 10.2166/hydro.2022.146
• Languages: English
• ISSNs: 1464-7141

Abstract

Abstract: Short-term (e.g., hourly) urban water consumption (or demand) prediction is of great significance for the optimal operation of the intelligent water distribution pump stations. In this study, three single models (autoregressive integrated moving average (ARIMA), back-propagation (BP), support vector machine (SVM)) and three hybrid models (ensemble empirical mode decomposition (EEMD)-ARIMA, EEMD-BP and EEMD-SVM) were developed and compared in terms of prediction accuracy and application convenience. 31-day (1 month) hourly flow series from a water distribution division in Shanghai were used for the demonstration case study, among which 30-day data used for model training and 1-day data used for model verification. Finally, the effects of historical data length on the prediction accuracy of three hybrid models were also analyzed, and the optima of the historical data length for three hybrid models were obtained. Results reveal that (1) the mean absolute percentage errors (MAPE) of EEMD-ARIMA, EEMD-BP, EEMD-SVM, ARIMA, BP and SVM are 5.2036, 1.4460, 1.3424, 5.7891, 4.3857 and 3.8470%, respectively. (2) In terms of prediction accuracy and actual practice convenience, EEMD-SVM performs best among the above six models. (3) The EEMD algorithm is effective for improving the prediction accuracy of six models. (4) The optimal historical data length of EEMD-ARIMA, EEMD-BP and EEMD-SVM are 11, 11 and 10 days, respectively. HIGHLIGHTS: Three single models (ARIMA, BP and SVM) and three hybrid models (EEMD-ARIMA, EEMD-BP and EEMD-SVM) were compared for the prediction of hourly water demand. EEMD-SVM performs best among the six prediction models. The EEMD algorithm is significant for improving prediction accuracy. The optimal historical data length for intelligent algorithms should be greater than a week. Graphical Abstract