A Deep Learning Based Multiobjective Optimization for the Planning of Resilience Oriented Microgrids in Active Distribution System

• Published in: IEEE access Vol. 10; pp. 84330 - 84364
• Main Authors: Vilaisarn, Youthanalack, Rodrigues, Yuri R., Abdelaziz, Morad Mohamed Abdelmageed, Cros, Jerome
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Back propagation; Back propagation networks; Deep learning; Distributed generation; Distributed power generation; Energy distribution; Energy sources; Environmental impact; Machine learning; Microgrids; microgrids formation; Multiple objective analysis; NSGA-II; Optimization; Regularization; Resilience; Resource management; stochastic linear optimal power flow; Stochastic processes; Switches; Upper bound; Weather
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3197194

When facing severe weather events, a distribution system may suffer from the loss or failure of one or more of its components, the so-called N-K contingencies. Nevertheless, taking advantage of the system's isolate switches and the increasing availability of distributed energy resources (DERs), a distribution system may be clustered into microgrids able to withstand such contingencies with minimal power interruption. In this perspective, this work proposes a novel bilevel optimization framework for planning microgrids in active distribution systems under a resilience-oriented perspective. For this, first, the outer level optimization features a multi-objective problem seeking to optimally allocate DERs and isolate switches in the distribution network while balancing the competing objectives of cost, resilience, and environmental impact. Next, the inner level handles the optimization problem pertaining to the optimal operation of the microgrids that can be created by harnessing local DERs and isolate switches allocated in the outer level. Further, given the proposed approach resilience-oriented focus, the developed framework employes deep learning models based on deep neural network (DNN) architectures trained using Bayesian Regularization Backpropagation (BRB) technique. This strategy allows for avoiding the modeling simplifications typically employed to alleviate the computational burden that can otherwise jeopardize planning solutions' feasibility. Thus, enabling the accurate consideration of microgrids' operational behavior, including hierarchal controls and the stochastic nature of loads, generation, and weather-induced line failures, especially critical aspects under resilience-oriented planning. Simulation case studies are developed to demonstrate the effectiveness of the developed planning framework.