面向强不确定性供需波动的新能源电网调度强化学习算法OA

As the new energy industry gradually scales up,dynamic economic dispatch in power systems aimed at enhancing new energy absorption capacity is becoming increasingly crucial.This paper proposes a distributed multi-agent reinforcement learning algorithm to address the dynamic economic dispatch problem in systems with high proportions of new energy.First,the total power demand of the grid at each time step is obtained through an average consensus algorithm,and a feasible power output that meets coupling constraints and fully absorbs new energy is determined using a projection optimization method.Additionally,by employing a quadratic function to approximate the state-action value function of the evaluation network,another near-optimal solution for dispatch is derived by solving a convex optimization problem.Subsequently,an action network is constructed to directly learn the relationship between total power demand,real-time maximum available output of new energy units,and the active power output decisions of each thermal power unit.Leveraging experience from extensive training,the model swiftly predicts optimal output power in dynamic power systems,thus improving generator decision efficiency.Finally,the effectiveness and robustness of the proposed algorithm are verified through tests on the IEEE 39-bus system.