Reinforcement learning for active distribution network planning based on Monte Carlo tree search

Active distribution network planning is of importance for utility companies in terms distributed generation investment, reliability assessment, optimal reactive power planning, substation evaluation, and feeder reconfiguration. However, it challenging current model-based optimization problems to guarantee the performances active due an empirically pre-defined solution space. To overcome this issue, paper proposes a performance-oriented method planning. The space model dynamically updated through using deep neural networks which are trained by Monte Carlo tree search-based reinforcement learning until desired satisfied. Simulation results based on standard IEEE 33-bus test system demonstrate that proposed can successfully improve level at lower investment cost compared other cases.