Abdullah Al Zishan, Moosa Moghimi Haji, Omid Ardakanian
{"title":"基于强化学习的插电式电动汽车充电自适应控制","authors":"Abdullah Al Zishan, Moosa Moghimi Haji, Omid Ardakanian","doi":"10.1145/3396851.3397706","DOIUrl":null,"url":null,"abstract":"This paper proposes an adaptive additive-increase multiplicative-decrease (AIMD)-like algorithm for controlled charging of plug-in electric vehicles in a power system. The proposed algorithm is decentralized and model-free, and relies on congestion signals received from sensors deployed across the network to avoid congestion. We use multi-agent reinforcement learning to dynamically adjust the parameters of the adaptive AIMD algorithm assuming that charging points are independent agents. We adopt imitation learning to pre-train these agents and an off-policy actor-critic deep reinforcement learning algorithm to determine the optimal control in the online setting. Simulation results obtained in a parking station with several charging points corroborate that the proposed algorithm closely tracks the available capacity of the network while avoiding line or transformer overloading, and outperforms the AIMD algorithm and other baselines in terms of utilization.","PeriodicalId":442966,"journal":{"name":"Proceedings of the Eleventh ACM International Conference on Future Energy Systems","volume":"57 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-06-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"12","resultStr":"{\"title\":\"Adaptive Control of Plug-in Electric Vehicle Charging with Reinforcement Learning\",\"authors\":\"Abdullah Al Zishan, Moosa Moghimi Haji, Omid Ardakanian\",\"doi\":\"10.1145/3396851.3397706\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"This paper proposes an adaptive additive-increase multiplicative-decrease (AIMD)-like algorithm for controlled charging of plug-in electric vehicles in a power system. The proposed algorithm is decentralized and model-free, and relies on congestion signals received from sensors deployed across the network to avoid congestion. We use multi-agent reinforcement learning to dynamically adjust the parameters of the adaptive AIMD algorithm assuming that charging points are independent agents. We adopt imitation learning to pre-train these agents and an off-policy actor-critic deep reinforcement learning algorithm to determine the optimal control in the online setting. Simulation results obtained in a parking station with several charging points corroborate that the proposed algorithm closely tracks the available capacity of the network while avoiding line or transformer overloading, and outperforms the AIMD algorithm and other baselines in terms of utilization.\",\"PeriodicalId\":442966,\"journal\":{\"name\":\"Proceedings of the Eleventh ACM International Conference on Future Energy Systems\",\"volume\":\"57 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2020-06-12\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"12\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Proceedings of the Eleventh ACM International Conference on Future Energy Systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3396851.3397706\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the Eleventh ACM International Conference on Future Energy Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3396851.3397706","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Adaptive Control of Plug-in Electric Vehicle Charging with Reinforcement Learning
This paper proposes an adaptive additive-increase multiplicative-decrease (AIMD)-like algorithm for controlled charging of plug-in electric vehicles in a power system. The proposed algorithm is decentralized and model-free, and relies on congestion signals received from sensors deployed across the network to avoid congestion. We use multi-agent reinforcement learning to dynamically adjust the parameters of the adaptive AIMD algorithm assuming that charging points are independent agents. We adopt imitation learning to pre-train these agents and an off-policy actor-critic deep reinforcement learning algorithm to determine the optimal control in the online setting. Simulation results obtained in a parking station with several charging points corroborate that the proposed algorithm closely tracks the available capacity of the network while avoiding line or transformer overloading, and outperforms the AIMD algorithm and other baselines in terms of utilization.