Key result
Evolutionary Strategy-assisted Deep Reinforcement Learning enhances exploration and reduces fine-tuning for volt/var control.
An Evolutionary Strategy-assisted Deep Reinforcement Learning method improves volt/var control in unbalanced distribution networks with photovoltaics by enhancing exploration and reducing reliance on expert fine-tuning.
With the rapid growth of photovoltaics (PVs), the inverter is widely used to provide flexible reactive power compensation and loss reduction for the distribution network. Due to their attractive features, data-driven methods, such as Deep Reinforcement Learning (DRL), have been widely applied to improve the performance of the volt/var control (VVC). However, the practical application of the current DRL-based VVC method is handicapped due to its insufficient exploration capability for the complex VVC environment and dependence on domain knowledge for fine-tuning or tailored algorithm selection. This paper proposes an Evolutionary Strategy (ES)-assisted tuning-friendly method to optimize the reactive power dispatch for inverter-based PVs in an unbalanced network considering nonlinear voltage unbalance constraints. It is confirmed by a case study on the IEEE 123 unbalanced network that the proposed method can enhance the exploration capability and mitigate the reliance on DRL fine-tuning and tailored algorithm selection with domain expert knowledge.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2024) studied Volt/var control in unbalanced distribution networks. Evolutionary Strategy (ES)-assisted tuning-friendly Deep Reinforcement Learning method vs. Current DRL-based VVC methods was evaluated on Reactive power dispatch optimization and exploration capability enhancement. An Evolutionary Strategy-assisted Deep Reinforcement Learning method enhanced exploration capability and mitigated reliance on fine-tuning for volt/var control in the IEEE 123 unbalanced network.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: