Closing the Loop: A Framework for Trustworthy Machine Learning in Power Systems

Abstract

Deep decarbonization of the energy sector will require massive penetration of stochastic renewable energy resources and an enormous amount of grid asset coordination; this represents a challenging paradigm for the power system operators who are tasked with maintaining grid stability and security in the face of such changes. With its ability to learn from complex datasets and provide predictive solutions on fast timescales, machine learning (ML) is well-posed to help overcome these challenges as power systems transform in the coming decades. In this work, we outline five key challenges (dataset generation, data pre-processing, model training, model assessment, and model embedding) associated with building trustworthy ML models which learn from physics-based simulation data. We then demonstrate how linking together individual modules, each of which overcomes a respective challenge, at sequential stages in the machine learning pipeline can help enhance the overall performance of the training process. In particular, we implement methods that connect different elements of the learning pipeline through feedback, thus "closing the loop" between model training, performance assessments, and re-training. We demonstrate the effectiveness of this framework, its constituent modules, and its feedback connections by learning the N-1 small-signal stability margin associated with a detailed model of a proposed North Sea Wind Power Hub system.

Towards an AI Assistant for Power Grid Operators

“…Finally, [98,99,100] let operators explore interactively and iteratively historical explainable factors across similar situations and decisions for augmenting and keeping up-to-date the system knowledge and proper labels. This should all be carefully developed within a trustworthy framework [101]. This is an illustrative sample of today's AI potential [102,103] to provide effective assistance functions and interactions which needs to be developed further.…”

Split-Based Sequential Sampling for Realtime Security Assessment

“…For instance, Yan [22] uses entropy as a metric to generate 'relevant' samples closer to the decision boundary. [24] uses 'directed walk' methods to samples around the decision boundary. The third type of approach, generic sampling, generates points uniformly distributed in the feasible space to explore all possible OCs.…”

“…Thus, a first point of comparison to generate pre-fault OCs is with methods that consider the OPF to generate initial OCs. The other approaches in the literature explore the entire feasible space in a generic way via random sampling, often using the Latin Hypercube sampling to generate initial OCs [37,39,40,24,22]. As a consequence, a second and more pivotal comparison is with those methods that aim to uniformly cover the search space using techniques like the Latin Hypercube sampling.…”

“…Consequently generic sampling allows the exploration of the full physical feasible space. While existing generic sampling approaches currently in the literature attempt to discard sections of the search space via rapid rejection sampling [41,39,40,24], our proposed method differs fundamentally from state-of-the-art approaches by optimally exploring the feasible space in an iterative fashion. This exploration is done by varying the objective function and active constraints while respecting all the physical feasible constraints.…”

“…Using only historical observations as training data is insufficient [22,23]. Therefore, the generation (sampling) of data is highly relevant for the success of all aforementioned approaches [24]. Many papers that focus on developing ML-based models consider the generation of the data, a few have focused only on the generation of database.…”