基于非合作博弈与分布式机器学习的多微网端对端交易

Design of Peer-to-peer Trading for Multi-microgrids Based on Non-cooperative Game and Distributed Machine Learning

针对当前端对端分布式交易方面的研究缺乏交易主体间细致的博弈关系分析,以及以机器学习为代表的人工智能方法在电力分布式市场交易方面的应用仍存在空白等问题,提出基于非合作博弈与分布式机器学习算法的多微网端对端交易方法。首先,基于多微网端对端分布式交易框架,构建市场交易主体,即微网的自治调度与端对端博弈交易模型;其次,提出基于弹性平均随机梯度下降算法的分布式机器学习框架,以及非合作博弈与分布式机器学习的多微网端对端交易流程;最后,通过实际算例仿真验证所提理论与方法在经济性、新能源消纳以及算法性能等方面的有效性与适用性。

The current research on peer-to-peer distributed trading lacks detailed analysis of game relationships between trading entities.There is still a gap in the application of artificial intelligence methods represented by machine learning in the distributed electricity market trading.Therefore,a multi microgrid peer-to-peer trading method based on non-cooperative games and distributed machine learning algorithms was proposed.Firstly,an autonomous scheduling and peer-to-peer game trading model of the market trading entity,i.e.,microgrid,was constructed based on the peer-to-peer distributed trading framework of multi microgrid.Then,a distributed machine learning framework based on elastic average stochastic gradient descent algorithm was proposed.And a multi-microgrid peer-to-peer trading process based on non-cooperative game and distributed machine learning was proposed.Finally,the effectiveness and applicability of the proposed theory and methodology in terms of economy,new energy accommodation and algorithm performance were verified through practical simulations.