To predict the building energy consumption, a semi-supervised learning outlier detection algorithm is developed to effectively detect and process abnormal energy consumption values. After detecting and processing abnormal energy consumption values, a building energy consumption prediction model based on multi-granularity feature extraction is proposed. The prediction performance of the proposed method is compared with others to explore the accuracy and superiority of the proposed prediction model. The findings indicated that the mean square error and mean absolute error of the local anomaly factor algorithm based on semi-supervised learning were 0.0073 and 0.063, respectively. Compared with the long short-term memory network model, the mean square error, root mean square error, and mean absolute error of the proposed model were significantly reduced by 41.07%, 17.86%, and 30.50%, respectively. Accurately predicting building energy consumption is beneficial for fine planning and management of building energy consumption, making certain contributions to the green development of the building industry and achieving energy conservation and emission reduction.
