计及线路电热耦合特性的配电网鲁棒强化学习动态重构方法OA

As the penetration of photovoltaics(PV)in distribution networks is increasing,the dynamic reconfiguration method based on the coordination of the soft open point(SOP)with sectionalizing and tie switches has become a key technological pathway for ensuring safe and stable operation of distribution networks.However,the electro-thermal coupling characteristics of lines are often overlooked during dynamic reconfiguration,leading to errors in resistance calculations that can skew reconfiguration results and compromise the safe and economic operation of the power grid.To address this,a robust reinforcement learning-based dynamic reconfiguration method considering the line electro-thermal coupling characteristics is proposed for SOP-equipped distribution networks.Firstly,to mitigate system modeling errors caused by the assumption of constant line resistance,a dynamic reconfiguration model is developed for SOP-equipped distribution networks considering the electro-thermal coupling of lines.Secondly,the original optimization problem is transformed into a Markov decision process,and a reward function based on first-order affine polynomials is constructed to evaluate the operation risks arising from PV and load fluctuations,thereby enhancing decision robustness.On this basis,a robust deep reinforcement learning algorithm is proposed,which realizes effective learning of robust optimization strategies through confidence-based action selection and robust updating mechanism of action network parameters.Finally,simulation tests on IEEE 34-bus and 123-bus systems show that the proposed method can better capture the dynamic variation of line resistance compared with traditional modeling approaches,improve decision reliability,and effectively reduce operation cost and risk under short-term PV generation and load fluctuations.