It is very important for electricity sales forecasting to optimize electricity supply structure and to understand economic trends.However,traditional electricity sales forecasting models cannot extract good enough data features from electricity sales data,which will generally result in lower forecasting accuracy.In this paper,a novel electricity sales forecasting model based on long-short term memory (LSTM) network is proposed.This model can analyze the correlation between electricity sales data and the related data.The original data are firstly clustered according to the features from different industries.Then,the LTM network is trained with the clustered data results.This model can automatically learn much better data features and intrinsic relationships between electricity sales data and other related data.Experimental results show the proposed model has achieved much higher forecasting accuracy than the traditional forecasting models.