面向混合增强智能的省级现货市场报价样本集增强及市场力识别技术  

Provincial Spot Market Quote Sample Set Enhancement for Hybrid Augmented Intelligence and Market Force Recognition

作　　者：宁龙飞 刘飞宇 王蓓蓓[1] 郑亚先 NING Longfei;LIU Feiyu;WANG Beibei;ZHENG Yaxian(College of Electrical Engineering,Southeast University,Nanjing 210018,Jiangsu Province,China;China Electric Power Research Institute,Nanjing 210037,Jiangsu Province,China)

机构地区：[1]东南大学电气工程学院,江苏省南京市210018 [2]中国电力科学研究院有限公司,江苏省南京市210037

出　　处：《全球能源互联网》2025年第1期110-123,共14页Journal of Global Energy Interconnection

基　　金：国家电网有限公司科技项目(涵盖极端形态的省级日前电能市场推演场景智能构建技术研究,SGZJJH00DKJS2310199)。

摘　　要：随着电力现货市场改革的推进,亟须对省级现货市场参与主体报价行为进行市场力识别以鼓励良性竞争。考虑到市场初期阶段,来源于现货市场实践的市场成员行使市场力而获取超额利润的样本数目相较于正常交易行为的样本较少,存在样本不平衡问题。首先基于现货市场实践的样本和专家经验进行市场力标签标记,并以此训练用于市场力识别的随机森林算法。然后利用样本抽样方法进行样本不平衡增强以提高市场力识别精度。此外考虑到随着市场成熟度的推进,评判市场力的标准不是一成不变,为了应对市场力评判标准的变化,将“人类智能”与“机器智能”结合构成混合增强智能算法。算例结果表明通过“人类智能”将市场力标签进行改变后,“机器智能”仍可以对市场力样本进行有效识别,体现了提出的混合增强智能的优势。With the advancement of electric power spot market reform,there is an urgent need to conduct market power identification on the quotation behavior of provincial spot market participants to encourage healthy competition.Considering that in the initial stage of the market,the number of samples derived from the spot market practice in which market members exercise market power to obtain excess profits is relatively small compared to the samples from the normal trading behavior,there is a sample imbalance problem.In this paper,first of all,the market power label tagging was labeled based on experts’experience.In this paper,the samples of spot market practice with market power labels based on experts’experience were labeled and used to train the Random Forest algorithm for market power identification,and then enhance the sample imbalance to improve the accuracy of market power identification by using the sample sampling method.In addition,it is considered that as the market matures,the criteria for judging market power are not static,and in order to cope with the changing criteria for market power,the“human intelligence”and the“machine intelligence”were combined in this paper.The results of the algorithm show that after the market power label is changed by“human intelligence”,and“machine intelligence”can still recognize the market power samples effectively,which reflects the advantages of the hybrid augmented intelligence proposed in this paper.

关 键 词：混合增强智能 多维度市场力样本集 样本不平衡 随机森林算法 

分 类 号：TM73[电气工程—电力系统及自动化]