Municipal Water Demand Forecasting in the Short and Long Term with ANN and SD Models

• Author / Creator

  Liu, Hanyu

• Decision makers require accurate estimates of water demand for the planning and operation of water resource systems. Short-term water demand forecasting can offer immediate information to assist daily or weekly operation while accurate long-term forecasting allows water utilities or governments to make informed decisions for water management and planning. These needs have prompted the development of statistical regression analysis and time series models for water demand forecasting over the past decades. More recently, artificial neural networks (ANNs) have increasingly been used to forecast short-term water demand due to their high prediction accuracy and independence from statistical assumptions. System dynamics (SD) models are superior for long-term forecasting because of their structure-based approach that permits detailed simulation of individual end uses, and the ease of running multiple scenarios for assessment of alternative management policies. This study developed both short-term and long-term water demand forecasting models for Edmonton, Canada, using ANN and SD models. The first part of the work explored the capability of ANNs for forecasting short-term – daily and weekly – water demands. Model development followed the conventional approach that includes model configuration, training, and testing for the study area; the performance of the best resulting ANNs was also compared to results from a conventional regression approach. The second part of the work produced a novel hybrid model composed of an ANN, several regression models and a system dynamics model for projection of long-term water demands. Several scenario groups were built based on the validated model to 1) investigate the relative importance of key demand drivers to long-term demands: population growth, climate change and policy implementation, and 2) assess the combined effect of multiple drivers to provide useful best-case and worst-case information, such as estimates of water demand from 2020-2100 and the year that water demand will double, of interest to water managers and decision makers for water demand management and planning. The study demonstrated the excellent ability of ANN for short-term forecasting, and the feasibility and accuracy of hybrid system-dynamics/data-driven models. The optimum ANNs for daily and weekly forecasting (with R2 = 0.92 and R2 = 0.89) consistently outperformed the conventional regression models. For short-term water demand forecasting, previous water demand was found to be the most effective predictor. Daily and weekly forecasting were found to depend relatively more on maximum air temperatures and mean air temperatures, respectively. Precipitation predictors were important only in conjunction with air temperature data, and precipitation amount was a better predictor for the Edmonton and region water demand than precipitation occurrence. For long-term forecasting, the hybrid model significantly outperformed an earlier SD model developed for Calgary over the whole simulated period, with an NRMSE that decreased significantly from around 7.9% to 4.7%. Simulations revealed that population growth produced the greatest change in water demand by 2100. Even with a slow population growth rate, the water demand under current policy conditions and medium climate change (RCP 4.5) increased by 162% by 2100, with doubling at 2079. The difference in water demand between high and low population growth scenarios was 20%, while climate change alone produced the least significant change – the difference between the high and low climate change scenarios was a 12% difference in water demand. The implementation of xeriscaping, greywater reuse and a best technology decreased the water demand by 17% compared to the reference scenario by 2100. Under the best-case scenario, with low population growth, low climate change and implementation of three water conservation policies, water demand doubled 30 years later than in the worst case, which included both high population growth and climate change, and no additional water conservation policies implemented.

• Subjects / Keywords

  • Municipal Water Demand
  • Water Demand Forecasting
  • Artificial Neural Network
  • System Dynamics

• Graduation date

  Spring 2020

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-w2pv-8s81

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.