Association Rule Mining for Localizing Solar Power in Different Distribution Grid Feeders

Abstract

The increase in the photovoltaic generation on distribution grids may create problems, such as voltage-violations. To gain situational awareness for system operation, e.g., adjusting the tap-settings of the transformers or adjust capacitor banks, utilities need situational awareness about locations and amounts of photovoltaic powers being generated in different feeders. Unfortunately, many utilities not only lack observability of the distribution grid, e.g., no secondary grid schematics but also have no situational awareness on which feeders solar panels locate. To understand where the solar users are roughly, we propose to use the feeder measurements from utilities with solar panel measurements from third-party solar companies. Due to the property of active correlation detection, we propose several sequentially improved methods based on quantitative association rule mining (QARM), where we also provide a lower bound for performance guarantees based on the amount of available data and the size of the bin for clustering. However, the binning of data leads to information loss. So, we design a band to replace bin for creating a new data mining approach for robustness. We validate our result for the IEEE 4-, 8-, 123-, 8500-bus cases with the Pecan-Street dataset, and also for the IEEE 123-bus case under low/high penetration and with radial/weakly meshed configurations. For realistic validation, we also obtain real data from a utility and a solar power company in the same zip codes in a city of California. Numerical results show accurate associations of feeders and solar panels, leading to increased situational awareness of the secondary distribution grids.