Detecting Load Transfers

Power distribution companies use load transfers as one way to improve systems reliability. These load transfers are not always logged accurately. As a result, the load data at the distribution level often include anomalies that are barriers to further data analysis such as load forecasting. This paper focuses on detecting load transfers between two meters at the distribution level. We propose two methods, a model-free method and a model-based method, inspired by a simple idea: aggregating meters with load transfers offsets the transfers. We first explore the features of load profiles to screen the meters with abnormal shapes. We then detect the best match for a load transfer by pairwise grouping the meters. We evaluate the proposed methods using the data from 428 meters of a U.S. utility company that has 128 load transfers. We simulate the load transfers based on these 128 known transfers to test the precision of the proposed methodology. The results demonstrate the effectiveness of both methods, while the model-based method can detect up to 81% of the simulated load transfers with a low false positive ratio. In addition, by grouping the meters with a load transfer, we improve the input data quality for a load forecasting process. The proposed solution significantly enhanced the accuracy of the load forecasts for the meters with transfers.