Short‐term water demand forecast modeling techniques—CONVENTIONAL METHODS VERSUS AI

TLDR

The study examined conventional regression and time‑series methods alongside AI approaches such as expert systems and artificial neural networks for short‑term water demand forecasting. Using daily water demand, temperature, and rainfall data from Lexington, KY (1982‑92), the authors developed and tested multiple models, assessing performance with two standard statistical metrics. AI models outperformed conventional ones, indicating that expert systems and ANNs warrant further exploration by water utilities to improve operational performance.

Abstract

A variety of forecast modeling techniques, from conventional techniques such as regression and time series analyses to relatively new artificial intelligence (AI) techniques such as expert systems and artificial neural networks (ANNs), were investigated for use in short‐term water demand forecasting. Daily water demand, daily maximum air temperature, and daily total rainfall data from Lexington, Ky., for 1982–92 were used to develop and test several forecast models. The performance of each model was evaluated using two standard statistical parameters. On the basis of the measured statistical parameters, the AI models outperformed the conventional models. Both expert system and ANN technologies should be further explored by water utility engineers and managers because these techniques have the potential to enhance the operational performance of various water supply and delivery systems.