电网技术2023，Vol.47Issue(12)：5147-5157,11.DOI:10.13335/j.1000-3673.pst.2023.0548
基于可迁移强化学习的断面输电极限计算方法
Transmission Limit Calculation of Corridors Based on Transferable Reinforcement Learning
摘要
Abstract
The corridor's transmission limit is defined as the downscaled projection of the grid's security boundary onto the cut set of the corridor,which is essentially a complex mixed integer nonconvex nonlinear problem considering voltage-reactive power optimization and multiple types of stability constraints.Furthermore,the increasing integration of new energies into the grid further expands the corridor's transmission limit computational dimension,making it diffiicult to solve by using the traditional methods.To this end,a method for calculating the transmission limit of corridors based on the transferable reinforcement learning is proposed.In the first place,a hybrid integer model of transmission limit with the differential-algebraic equations is established,which takes into account the constraints related to the transient power angle,the voltage stability,and the reactive power resources like the capacitor switching.Subsequently,the model is transformed into a Markov decision process with the mixed integers,and a method of proximal policy optimization based on the mixed Categorical distribution is proposed.Ultimately,the policy distribution entropy is introduced to maximize the objective to ensure the transferability of the intelligent computing model in the unseen operating modes,realizing the fast calculation of the transmission limit of the corridors under the implementation of the operating modes or the boundary condition switching.The verification of the IEEE 39-node system shows that compared with the traditional meta-heuristic black-box optimization algorithm,the proposed method improves the calculation effiiciency by 97.15％without sacrificing the accuracy.关键词
输电极限/无功优化/可迁移强化学习/近端策略优化/策略分布熵Key words
transmission limit/reactive power optimization/transferable reinforcement learning/proximal policy optimization/policy distribution entropy分类
信息技术与安全科学引用本文复制引用
李康文,邱高,刘挺坚,刘友波,刘俊勇,丁理杰..基于可迁移强化学习的断面输电极限计算方法[J].电网技术,2023,47(12):5147-5157,11.基金项目
国家自然科学基金资助项目(52307124) （52307124）
中央高校基本科研业务费专项资金资助项目(YJ2021162) （YJ2021162）
四川省科技厅项目(2021LDTD0016-LH).Project Supported by National Natural Science Foundation of China(52307124) （2021LDTD0016-LH）
the Fundamental Research Funds for the Central Universities(YJ2021162) （YJ2021162）
the Science and Technology Department of Sichuan Province(2021LDTD0016-LH). （2021LDTD0016-LH）
