The Combination Of Seasonal Adjustment Method And Echo State Network For Energy Consumption Forecasting In USA

As the fast development of economy,society has a growing demand for energy.The conventional fossil energy including coal,petroleum and natural gas could provide the life of people with great convenience,while it also causes environmental pollution problems.To coordinate the contradiction between them,the renewable energy including hydroenergy,wind energy and geothermal energy gradually makes a figure in the energy system,occupying respectable market shares and eroding the consumption rate of fossil energy to a certain extent.However,it is a fact that the consumption of fossil energy is still rising.It is of significance for nation to build a scientific and accurate energy consumption prediction model,because it helps to develop the economy,enact the energy policies and allocate the energy resources.Energy consumption is a complex nonlinear system,it's hard to obtain satisfactory forecasting performance with single method.Recently,data mining methods are widely used in different fields and gain some success,which offer the mature technical support for the further study concerning the energy consumption prediction.This paper combines the seasonal adjustment method and neural network to model and predict the monthly energy consumption of USA including fossil energy and renewable energy.According to the proposed forecasting technique,the seasonal adjustment method decomposes the energy consumption time series into two parts including the seasonal subseries and the remainder subseries.Echo state network(ESN)is utilized to model and predict the seasonal subseries.The remainder subseries is decomposed into a series of intrinsic mode functions and the residual using ensemble empirical mode decomposition(EEMD).ESN optimized by grasshopper optimization algorithm(GOA)is used to forecast these series.Finally,two parts are summed to generate the final predictive results.The empirical studies of fossil energy and renewable energy consumptions are used to verify the effectiveness and scalability of the proposed method for the energy prediction problem.In addition,the sample-extrapolation forecasting results also reflect its good performance.