This paper reviews recent works applying machine learning techniques in the context of energy systems reliability assessment and control. We showcase both the progress achieved to date as well as the important future directions for further research, while providing an adequate background in the fields of reliability management and of machine learning. The objective is to foster the synergy between these two fields and speed up the practical adoption of machine learning techniques for energy systems reliability management. We focus on bulk electric power systems and use them as an example, but we argue that the methods, tools, etc. can be extended to other similar systems, such as distribution systems, micro-grids, and multi-energy systems.

Abstract-In this paper we study how supervised machine learning could be applied to build simplified models of realtime (RT) reliability management response to the realization of uncertainties. The final objective is to import these models into look-ahead operation planning under uncertainties. Our response models predict in particular the real-time reliability management costs and the resulting reliability level of the system. We tested our methodology on the IEEE-RTS96 benchmark.Among the supervised learning algorithms tested, extremely randomized trees, kernel ridge regression and neural networks appear to be the best methods for this application. Furthermore, by using feature "importances" computed by tree-based ensemble methods, we were able to extract the most relevant variables to predict the response of real-time reliability management, and thus obtain a better understanding of the system properties.

Abstract-In the context of operation planning, probabilistic reliability assessment essentially boils down to predicting, efficiently and with sufficient accuracy, various economic and reliability indicators reflecting the expected performance of the system over a certain look-ahead horizon, so as to guide the operation planner in his decision-making. In order to speedup the crude Monte Carlo approach, which would entail a very large number of heavy computations, we propose in this paper an approach combining Monte Carlo simulation, machine learning and variance reduction techniques such as control variates. We provide an extensive case study testing this approach on the three-area IEEE-RTS96 benchmark, in the context of day-ahead operation planning while using a security constrained optimal power flow model to simulate real-time operation according to the N-1 criterion. From this case study, we can conclude that the proposed approach allows to reduce the number of heavy computations by about an order of magnitude, without sacrificing accuracy.

A community microgrid is a microgrid composed of several entities, or members, that can share energy among themselves. The members of the community can match their demand and supply through an internal local market with a significant reduction of the exchanges with the main grid. As a consequence each participant can benefit from a reduction of its energy costs when the energy available locally is cheaper than the energy from the grid, from a drop of the energy peak demanded from the main grid, and from the new capability to provide energy reserve at aggregate level. In this paper, we analyze how the changes of the community market model parameters can affect both the community as a whole, and the welfare of each participant. The analysis is performed by varying the main drivers of the community market model, the community and storage fees, and the storage capacity. The numerical results are obtained by using real data based on the MeryGrid project.

In operation planning, probabilistic reliability assessment consists in evaluating, for various candidate planning decisions, the induced probability of meeting a reliability target and the expected operating cost over a certain future time period. In this paper, we propose to exploit Monte-Carlo simulation and machine learning to predict operation costs for various day-ahead unit commitment and economic dispatch decisions and a range of realisations of uncertain loads and renewable generations over the next day. We describe how to generate a database, how to apply supervised machine learning to it, and how to use the learnt proxies to rank candidate day-ahead decisions in terms of the expected operating cost they induce over the next day. We illustrate the approach on the IEEE-RTS96 benchmark where we use the DC power-flow approximation and the N-1 criterion to simulate real-time operation and to generate generation schedules in the day-ahead operation planning stage.