Energy Consumption Forecasting Using Echo State Network Supported By Improved Frog-Leaping Algorithm

Energy consumption is an important source of the survival and development of human society.Along with the continuous improvement of people’s economic level and the continuous development of global industry,we are faced with the contradiction between the growing demand for energy,and the global energy reserves continued to decline.Reliable energy consumption forecasting can provide us effective decision making support for planning energy consumption and for establishing right national energy policy.This thesis constructed an energy prediction model based on the Echo State Network(ESN)and designed an improved frog leaping algorithm to calculate the output weight matrix of the ESN.Firstly,the standard frog leaping algorithm is improved by introducing the crossover operator and mutation operator from the differential evolution algorithm into the renewal strategy of the frog leaping algorithm,and added a selection strategy and resource pool structure were added into the improved algorithm.The improved algorithm was called Selected Differential Shuffled Frog Leaping Algorithm(SDSFLA).Secondly,to test the performance of the SDSFLA,a total of sixteen widely used benchmark test function and four kinds of algorithms were considered.The results verified the effectiveness and efficiency of the SDSFLA.Then the SDSFLA was used to calculate the output weight matrix of the ESN,the proposed prediction model called SDSFLA-ESN can effectively prevent the disease solution,and improve the goodness of fit.Finally,the SDSFLA-ESN was applied to the two cases of US energy consumption forecasting and China’s energy consumption forecasting.Compared with other predictive models,the SDSFLA-ESN shows high practicality in energy consumption forecasting.