AWWA ACE59835

This paper discusses how Artificial Neural Network (ANN) technology served as a valuable tool in making conservation program choices in an ongoing residential water conservation program for Edmonton, Alberta, Canada. The focus of this study was to identify significant influencing factors for residential water consumption and use this information to adopt public education programs that would best influence the behavior of residential water users. Artificial Intelligence (AI) was used in-house to create an ANN model that forecasts residential water consumption levels. One of the most important steps in creating the model was determining the statistically significant factors that have affected residential water consumption in Edmonton over the last 30 years. Now complete, the model can be used to predict consumption forecasts based on various combinations of these factors. The ANN model provided the list of the main factors that influence water consumption as well as a means for running multiple scenarios based on a range of values for the various factors. Each scenario produced a consumption prediction. The degree of variability of these consumption predictions was analyzed to determine which factors have the greatest impact on consumption. The order of significance of impact of these factors on residential consumption proved to be: base consumption index (a measure of the lowest year-round minimum usage); weather (summer and winter months); and, customer count (can be most or least significant, depending on the span of years studied). Determining the order of significance of the main residential consumption drivers helped identify which factors could be targeted through public water conservation initiatives. ANN technology provided the information necessary to eliminate options and to focus resources on the most effective and cost efficient programs for the community. In order to create a targeted residential water conservation program, the study researched these 5 questions: is the ANN model predicting consumption with a reasonable degree of accuracy; what is the range of possible residential water consumption (these values were used as a check on later calculations); what factors drive residential water consumption (these factors were determined during development of an ANN residential water consumption forecasting model); how much can the changes in each consumption driver affect total residential water consumption (the degree of variability was assessed by running scenarios with the ANN model); and, which factors can be influenced by water conservation campaigns? Includes 6 references, figures.