A combination approach based on seasonal adjustment method and echo state network for energy consumption forecasting in USA

Qin Lan,Li Weide

DOI: https://doi.org/10.1007/s12053-020-09897-x

IF: 3.1339

2020-01-01

Energy Efficiency

Abstract:It is extremely significant to construct a scientific and accurate forecasting model for energy consumption of the USA, because it could help to formulate energy policies and allocate energy resources. In recent years, more and more hybrid models based on divided-and-conquer method have been applied in energy consumption prediction to obtain satisfactory results. However, owning to the obvious enhancing effect of decomposition methods, the issue concerning seasonal fluctuation existed in time series is rarely considered before modeling. It is a fact that seasonality indeed influences the performance of prediction. This paper proposes a hybrid forecasting model for energy consumption, which combines seasonal adjustment method, ensemble empirical mode decomposition (EEMD), echo state network (ESN), and grasshopper optimization algorithm (GOA). The seasonal adjustment method that inherits the idea of divide and conquer is used to decompose original time series into only two parts including seasonal subseries and remainder subseries rather than regular three parts(seasonality, trend, and residual) for avoiding the complex modeling task of the residual subseries. Then models ESN and EEMD-GOA-ESN are utilized to model and predict the seasonal subseries and remainder subseries respectively. Finally, two parts are summed to generate the final predictive results. The empirical studies of fossil fuels, nuclear electric power, and renewable energy consumptions show that the proposed model outperforms other alternative benchmarks with regard to effectiveness and scalability. Besides, the sample extrapolation forecasting displays that the technique could limit the error of monthly energy consumption to 3.3%.