Empirical Mode Decomposition–Least Squares Support Vector Machine Based for Water Demand Forecasting

Accurate forecast of water demand is one of the main problems in developing management strategy for the optimal control of water supply system. In this paper, a hybrid model which combines empirical mode decomposition (EMD) and least square support vector machine (LSSVM) model is proposed to forecast water demand. This hybrid is formulated specifically to address in modelling water demand that has high non-linear and nonstationary time series which can hardly be properly modelled and accurately forecasted by traditional statistical models. EMD is used to decompose the water demands into several intrinsic mode functions (IMFs) component and one residual component. LSSVM is built to forecast these IMFs and residual series individually, and all of these forecasting values are then aggregated to produce the final forecasted value for water demand series. To assess the effectiveness and predictability of proposed models, monthly water demand record data from Batu Pahat city in Johor of Peninsular Malaysia, has been used as a case study. Empirical results suggest that the proposed model outperforms the single LSSVM and artificial neural network (ANN) model without EMD preprocessing and EMD-ANN model. Thus, the EMD-LSSVM model is an effective method for water demand forecasting.