Urban Monthly Water Consumption Forecasting Based on Signal Decomposition and Optimized Extreme Learning Machine

With the development of information technologies, the smart water is getting to be popular. Water consumption forecasting plays a significant role in the smart water system. This paper proposes a PSO-ELM (Extreme Learning Machine Optimized by Particle Swarm Optimization) coupled with SDEMD (Secondary Decomposition of Empirical Mode Decomposition by Variational Mode Decomposition) for monthly water demand forecasting. Firstly, we use the EMD (Empirical Mode Decomposition) to decompose the water consumption data and the VMD (Variational Mode Decomposition) to process the first sub-time series. Then, PSO-ELM is employed to forecast the sub-time series separately, the forecasting results of every sub-time series are combined together to form the final forecasting result. The experimental results reveal that the SDEMD-PSO-ELM model can effectively forecast the monthly water consumption for strong nonlinearity and non-stationary data and obtains better performance than ELM (Extreme Learning Machine), RBFNN (Radial Basis Function Neural Network) and VMD-ELM models.