Abstract: The electricity demand of industry users is usually affected by seasonal and cyclical factors, and sometimes the data obtained is incomplete, missing, or incorrect, which can have a negative impact on the accuracy of predictions. In order to achieve accurate prediction of industry user electricity demand, a data-driven and heterogeneous computing method for predicting industry user electricity demand is proposed. The Lagrange interpolation algorithm is used to fill in the missing part of user electricity data, the standardized preprocessing of electricity data is used to make electricity demand prediction accurate enough, denoising autoencoders and sparse constraint functions are used to extract electricity data features. The long-term memory neural network’s forgetting gate layer, input gate layer, update gate layer, and output gate layer are used to obtain the future trend of electricity demand, the task of industry user electricity demand prediction is completed. The experimental results show that the proposed method is suitable for long-term and short-term industry user electricity prediction, and the prediction results have high accuracy and short time consumption. 

Key words: data driven, heterogeneous computing, user electricity demand, denoising autoencoder, bias term, weight matrix