Development of Water Demand Forecasting Simulator and Performance Evaluation

Generally, treated water or raw water is transported into storage reservoirs which are receiving facilities of local governments from multi-regional water supply systems. A water supply control and operation center is operated not only to manage the water facilities more economically and efficiently but also to mitigate the shortage of water resources due to the increase in water consumption. To achieve the goal, important information such as the flow-rate in the systems, water levels of storage reservoirs or tanks, and pump-operation schedule should be considered based on the resonable water demand forecasting. However, it is difficult to acquire the pattern of water demand used in local government, since the operating information is not shared between multi-regional and local water systems. The pattern of water demand is irregular and unpredictable. Also, additional changes such as an abrupt accident and frequent changes of electric power rates could occur. Consequently, it is not easy to forecast accurate water demands. Therefore, it is necessary to introduce a short-term water demands forecasting and to develop an application of the forecasting models. In this study, the forecasting simulator for water demand is developed based on mathematical and neural network methods as linear and non-linear models to implement the optimal water demands forecasting. It is shown that MLP(Multi-Layered Perceptron) and ANFIS(Adaptive Neuro-Fuzzy Inference System) can be applied to obtain better forecasting results in multi-regional water supply systems with a large scale and local water supply systems with small or medium scale than conventional methods, respectively.