Abstract: Existing long-term electricity consumption prediction methods are difficult to solve the problem of

variable selection，which leads to inaccurate prediction results of power consumption. Therefore，combining the

radio frequency variable selection in random forest（RF）algorithm with long short-term memory（LSTM）

regression in long-term and short-term memory networks，a long-term electricity consumption prediction model

based on RF variable selection and LSTM regression was designed. RF method was used to evaluate the

importance of a single variable，and the correlation coefficients between each influencing factors and electricity

consumption were obtained. Then，the variable with higher value was selected as the basis of electricity

consumption forecast. Combined with the selection results of RF variables，the theory of power system was

analyzed，and the relationship between power consumption and industrial development level，temperature and other factors was studied by using convergence cross mapping method. Based on the relationship between various factors and the LSTM regression method，a prediction model of electricity consumption was established，and the long-term prediction of electricity consumption was realized. The results show that，compared with the traditional methods，the designed model has higher prediction accuracy and efficiency，can predict the long-term electricity consumption in the growing period quickly and accurately，and has high application value.

      Key words: variable selection；random forest（RF）algorithm；long short-term memory（LSTM）regression；

long-term power consumption；prediction model