Renewable-based microgrids' energy management using smart deep learning techniques: Realistic digital twin case

In this research study, a novel demand response program (DRP) has been proposed for renewable-based microgrids (MGs) which takes into account the high penetration of tidal units and solar energy as important, prevalent renewable resources in the power systems. To this end, multi-objective problem (MOP) structure is a promising solution to reduce the whole operating costs of the scheduling problem as well as decrease the high risk of failure in electrical power transmission because of component increasing failure rates and extended repair time. The complexities and nonlinearity of the problem necessitate an innovative heuristic solution which is derived from the Grey Wolf optimization algorithm to help for resolving the problem without making any assumptions or compromising precision. This paper also proposes the dynamic 3-phase correction (DPC) formulation to boost the layout convergence ability by increasing the global search features. Through such a modification, the diversity of the members in the algorithm population increases which would result in low computational burden and low possibility of trapping in local optima. Moreover, it is necessary to have a clear and accurate estimation of the output power of renewable sources in the system. Considering that solar irradiance is hard to anticipate, this paper develops a deep learning layout that uses generative adversarial networks (GAN) to forecast the hourly power generation of the tidal and solar agents. GAN model consists of two competing networks which help to enhance the training process by increasing the accuracy of distinguishing real data from fake data. At the end, IEEE standard test system is used to evaluate the efficiency and effectiveness of the suggested multi-layer problem. The simulation results display that the suggested deep model outperforms other well-known algorithms in smart microgrids.