Echo State Networks With Adaptive Particle Swarm Optimization Algorithm For Energy Demand Forecasting

Energy is an important material foundation for economic development of countries.In recent years,under the influence of the continuous development of the global economy and the increase in heating and cooling demand in some regions,the total amount of global primary energy consumption has continued to grow.The growth rate in 2018 is the highest since 2010.At the same time,numerous factors affecting energy demand have higher inherent complexity and irregularities.Therefore,accurate forecasting of energy demand has become a challenging issue,which can make a significant contribution to energy security strategy formulation and economic development planning.Firstly,this thesis introduces adaptive operators to improve the inertial weights and learning factors of the standard particle swarm optimization algorithm(PSO)to improve the balance between the exploration and exploitation.At the same time,a nonlinear function is used to construct a nonlinear relationship between the internal states of Echo State Networks(ESN).Then,APSO is used to optimize the key parameters of NESN to propose an assembled prediction model named APSO-NESN.Then,the effectiveness of APSO-NESN in energy demand forecasting is verified through two comparative examples of US electricity demand forecasting and Hubei industrial electricity consumption forecasting.Experimental results show that the APSO-NESN model has higher prediction accuracy than the prediction models such as GD-ANN,ADE-BPNN,adaboost and adaboost＿sp.Finally,APSO-NESN is applied to industrial electricity consumption forecasting in Guangdong Province.Gradient Boosting Decision Tree(GBDT)is used for input variables selection.The results of APSO-NESN,ESN,PSO-NESN and other models are compared,which shows that APSO-NESN has excellent performance in industrial electricity consumption forecasting and can provide reliable auxiliary support to make decisions on energy policies for countries or regions.