EIMC2 seminar: Phase Management of Power Distribution Systems

The power grid is designed as a three-phase alternating current (AC) system. To operate and control the grid effectively, it is crucial to know precisely how power sources and consumers are connected to its phases. Additionally, the three phases of the grid must be balanced as much as possible; otherwise, system stability, power quality, and equipment safety will be significantly compromised. This talk will cover two important topics for phase management in distribution systems: phase identification for loads without phase labels, and load phase reconfiguration for phase imbalance mitigation. First, we will present a data-driven method for phase identification, inspired by concepts from information theory. This method analyzes event signatures from customer locations and the feeder head, using data from smart meters. By examining the cross-entropy of their mutual information, it accurately identifies the correct phases for both wye and delta-connected loads. Next, we will introduce a data-driven approach for scheduling phase rebalancing over a finite horizon. Phase scheduling is formulated as a Markov Decision Process (MDP), and an Imitation Learning (IL) framework, implemented through Ensemble Random Forests, is used to streamline the process of sequential load phase swapping. The previously costly multi-objective optimization is now limited to steps where phase swapping is needed.