Water Distribution Systems Demand Forecasting with Pattern Recognition

Abstract

The ability to accurately forecast water demand is an integral part of water utility planning for water production levels and the associated finances. Although experienced people are capable of predicting water demand with a reasonable level of accuracy, systematic prediction methods are required to minimize the randomness introduced by human subjectivity, define the quantifiable factors that affect water consumption, and pave the way for further analysis and understanding of these factors by providing the power of what if capability. Statistical models can be used to explain fluctuations in water use and, subsequently, the gained knowledge can be applied to forecasting total water usage. For more accurate studies, the forecasting models may be further refined, in particular by time frame. Short term water forecasting is for water consumption on a daily basis and is useful for decisions related to the daily operation of water treatment plants and distribution systems. Long term forecasting looks at water consumption on a monthly basis, or even longer span of time. These longer term forecasts are useful for revenue forecasting and water treatment plant expansion planning. The factors influencing water consumption are complex. The relationships between known factors and water consumption are often inconsistent and indirect. Furthermore, consumption analysis is limited to the information that is available and quantifiable. There are other factors that are difficult to quantify and difficult to forecast, such as economic factors, social factors, water usage habits, accidents, leaks, fires and climate changes. These factors contribute to the incidence of deviations between historic and current water usage patterns. At EPCOR Water Services Inc. (EWSI), several water demand forecast techniques have been investigated to facilitate better planning of the entire water utility cycle. Among the techniques investigated, pattern recognition has proven to be useful, in particular because it takes advantage of available historical data. This paper closely examines pattern recognition technology, and how its advantages are exploited in both short term and long term forecasting. Furthermore, this paper discusses other techniques used for short and long term forecasting and compares them to pattern recognition. This paper was presented at the 8th Annual Water Distribution Systems Analysis Symposium which was held with the generous support of Awwa Research Foundation (AwwaRF).