Learning-Based Situational Awareness, Decision Making, And Flexibility Aggregation for Power Distribution Systems with Uncertainty

Abstract

The ever-growing penetration of distributed energy resources in both the generation and demand-side brings environmental benefits and technical challenges to electric power distribution systems. Specifically, due to the inherently intermittent nature of renewable energy resources and the invisible behaviors of customers in electricity use, there exists a high level of uncertainty, which has significantly threatened the stable, secure, and dedicated operation of distribution systems.The conventional operation strategies tend to be model-driven based on offline studies or historical experiences, leading to an over-conservative or risky operation solution especially when the system encounters considerable uncertainty. That is, such a deterministic solution is highly difficult to adapt to the various unknown system operating conditions. Therefore, it is imperative to find a proper operation strategy for distribution systems under uncertainty. With the high volume of the real-time measurement data available to the distribution system operator and the huge success of ML technologies in the data-intensive industry, it is promising to marriage the knowledge representation and reasoning power of ML to analyze, understand and reveal the potential effects of uncertainty from data itself, finally solving optimal operation problems under uncertainty more efficiently and accurately. This dissertation aims at developing learning-based approaches for three representative and challenging operation problems to have accurate situational awareness, optimal decision making, and efficient flexibility aggregation under uncertainty. The problems are focused on identifying the behind-the-meter Electric Vehicle charging load, scheduling energy storage systems for voltage regulation, and estimating a feasible active-reactive power flexibility region for capacity support.