The power management system (PMS) plays a critical role in microgrid (MG) management, ensuring stable and reliable operation while minimizing energy costs by effectively managing local generation resources. This paper proposes a novel approach for power management in a grid-connected microgrid using deep reinforcement learning (DRL) algorithm. The proposed approach uses a deep neural network (DNN) trained by a covariance matrix adaptation evolution strategy (CMA-ES) to generate optimal policies for the microgrid power management. To demonstrate the effectiveness of the approach, a case study is conducted where the DNN is trained using CMA-ES on representative load and meteorological data to generate an optimal PMS policy. The results show the effectiveness of the proposed methodology in achieving efficient microgrid power management by reducing energy cost in grid connected mode and ensuring a stable operation in islanded mode.

Electrical power transmission is provided by the high-voltage alternatif current (HVAC) system, but in long lines, the HVAC system creates huge problems that can affect the electrical system. For example, over distances between 400and 800km, the HVAC system can produce capacitive effects that lead to power surges at the ends of the line. For longer distances, the high-voltage direct current (HVDC) transmission system plays an increasingly important role in power transmission, as HVDC power transmission is more economical and technically more reliable than HVAC transmission, especially for connecting off-shore renewable energy generation to the grids. The main objective of this work is to develop an optimal two-level power flow to ensure stable and economical operation for the hybrid HVDC/HVAC power system. To show the effectiveness of the proposed approach, simulations are performed on the 30-bus IEEE power system. The obtained results show that for long distance transmission, HVDC is more advantageous with 6.26Mw of active losses and 18.33Mw for HVAC. The results also show the evolution of the power transited in the HVDC line during 24 hours as a function of the electrical load and the production cost. The proposed approach allows optimizing the parameters of the HVDC link in order to minimize the active losses, the voltage deviation, the generation cost and the transited power.

This paper present several cases of study. The network used is the IEEE 10 generators in two different seasons, summer and winter. Two metaheuristic methods were chosen and examined (GA and PSO) for the study of the optimization processed in this paper, which presents the minimization of photovoltaic costs, NOx emission cost, and the cost of the fuel with the effect of the valve.