Identifying Features that Impact Real-Time Power Flow Performance Using Data Analysis Methods

Load estimation is critical in near real time monitoring of distribution systems due to limited real-time data. Limited real-time data makes it also challenging to determine the cause when the load estimation (called Bus Load Allocation (BLA)) provides poor results on a consistent basis. This paper focuses on this issue of poor BLA performance and proposes a two-step data analytics-based method to help identify features of a system that correlate to poor BLA performance. The proposed method first utilizes clustering on a sample of cases to check if the cases with poor performance can be separated from cases with good performance by using the selected set of features/parameters. The method then uses two-sample t-test to determine the features that are different between good and poor performing clusters. A sample study using a set of cases from a real system is given to illustrate the effectiveness of the proposed method.