Short-term demand forecasting for distributed water supply networks: A multi-scale approach

As one of the most important parts in the operation and management of modern metropolis, distributed water supply networks is directly related to the residents' quality of life. Water demand forecast, as a classic problem in water management, is an efficient method to schedule urban water facilities. In this research, a multi-scale method based on Empirical Mode Decomposition (EMD) and a dynamic architecture of Artificial Neural Network (DANN) is proposed to forecast the daily water demand in Shanghai. For the sake of achieving more accuracy and dealing with the nonlinearity and non-stationary in time series, EMD is adopted in the method. Due to DANN's good performance in forecasting the urban water demand, improvement in our research is appreciable. To meet requirements of the distributed system, the idea of multi-scale has congenital superiority in data process. There are four criteria utilized to evaluate the accuracy. In addition, a measurement of forecasting trend is defined to evaluate model's ability in forecasting the dynamic change in time series.