基于安全图强化学习的柔性互联配电网分布式电源承载力动态评估OA

Aimed at the problem that it is difficult to achieve dynamic and accurate evaluation of distributed generation(DG)hosting capacity based on the mathematical optimization methods at present,a dynamic evaluation method for the hosting capacity of DG in flexible interconnected distribution network is proposed,which is based on safe graph rein-forcement learning(RL).First,considering the dynamic hosting capacity enhancement and network loss optimization of DG under the regulation of soft open points(SOPs),a dynamic DG hosting capacity evaluation model is constructed,which is applicable to three-phase unbalanced distribution network.Second,the dynamic DG hosting capacity evalua-tion model is transformed into a standard form of constrained Markov decision process,so as to achieve a balance be-tween the optimal decision-making and safe actions.Third,a soft actor-critic(SAC)algorithm with embedded graph convolutional neural network is proposed for offline training and online execution.The graph convolutional neural net-work is embedded into the SAC algorithm strategy network,so that the real-time and accurate evaluation for the dynam-ic DG hosting capacity in the distribution network is achieved.Finally,simulations are conducted with an IEEE 33 node network as an example to validate the effectiveness of the proposed model and method.