Management of a Microgrid using Deep Learning Techniques

Panduro-Ramirez J.; Joshi K.; Chakravarthi M.K.; Singh P.; Bind N.K.; Banik A. · Proceedings of 5th International Conference on Contemporary Computing and Informatics, IC3I 2022 (2022) · DOI: 10.1109/ic3i56241.2022.10073395

Resumen

Our power generation grid is experiencing a dramatic modernization as a result of recent developments in smart-grid advancements. By offering a flexible method of sort of technology solar and wind power (RES) into the electricity system, microgrid (MG) plays a significant role in the modernity. Spread RES, like solar and wind, can, nevertheless, be very essentially random and occasional. It is challenging to effectively execute an MG because of these erratic resources, state of charge, and the like but on the sides of the market. This paper addressed this issue and presented a novel strategy for the real-time timetabling of an MG taking into account the adequacy of renewables, energy cost, and utility grid confusion. The suggested solution is studying and therefore does not involve an accurate model of the confusion, in contrast to traditional model-based methods that call for an indicator to estimate the confusion. The goal of the MG power process is to reduce the daily operational expenses, and it is specifically modelled as a Markov Decision Process (MDP). To resolve the MDP, a deep encouragement learning (DRL) strategy has been developed. The deep Q-network (DQN) algorithm is used to find the human brain in the DRL framework, which uses a deep fnn network to measure the ideal response role. The suggested method directly generates real-time creation schedule using the MG's current state as inputs. Furthermore, case experiments are performed to show the usefulness of the suggested strategy using total power data from the Cal Organization For Standardization (CAISO).

Instituciones: Universidad Tecnológica del Perú