Research on repair method of abnormal energy consumption data of lighting and plug based on similar features

•A kNN-BS method is proposed to repair lighting and plug energy data.•Using cross-validation method to select the optimal k.•Using 2-step data classification method to decide day type.•Achieved higher accuracy than traditional kNN. The public building energy consumption monitoring platforms (BECMPs) play an important role in building energy conservation and energy-related decision-making. However, problems such as data missing and data mutation are quite common that they impede the effectiveness of the BECMPs. Based on the kNN algorithm, this paper proposes a kNN-BS data repair algorithm with the characteristics of electricity consumption trends taken into consideration. After data classification and cleaning, the data of hourly lighting and plug power consumption of an office building of Dalian University of Technology is repaired using the kNN-BS algorithm. The cross-validation method is used to determine the optimal k value. The relative repair error of the total daily electricity consumption of kNN-BS is less than 4% for working days and less than 6% for non-working days. Compared with kNN, the average CVRMSE of the kNN-BS algorithm is reduced by 1.36%∼1.59% for working days and 5.49%∼8.17% for non-working days. The kNN-BS algorithm shows excellent stability and robustness compared with kNN, polynomial and BPNN, which makes it very suitable for practical use, especially in efficiency-demanding scenarios. The kNN-BS algorithm can effectively improve the quality of public building energy consumption monitoring data and provide a theoretical basis and technical means for solving data quality problems in the BECMPs.