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

Automated summary

This paper proposes a performance-oriented method for active distribution network planning, which improves the performance of the planning through dynamically updating the solution space using deep neural networks.