Abnormal data processing is the premise of accurate prediction of building energy consumption. In order to effectively deal with the abnormal data of building energy consumption, this paper uses the K-means and KNN algorithms based on machine learning to identify and repair the hourly energy consumption data for July 2017 of a hotel building in Shanghai. By establishing a long and short term memory neural network model, this paper uses the processed energy consumption data to predict the hourly energy consumption data for the first week of August 2017. The prediction results show that the method proposed in this paper can accurately identify and repair abnormal building energy consumption data, thus significantly improving the effect of subsequent energy consumption prediction.