多源数据融合与人机混合实验驱动的两级电力市场全景式建模与决策理论研究

The new round of electricity market reform is undergoing in China. The modern electricity market system with the spot trading as its core will be established. The big data generated by the wide deployment of smart grid technologies enable market participants to accurately understand their customers and competitors, and therefore will deeply change the decision paradigm of bidding, pricing and rule design in electricity markets. However, traditional methods of decision making are based on incorrect assumptions such as complete rationality or information; thus they cannot effectively solve the market operation problems. In this project, we will follow a research method characterized by “bottom-up, cross domain data fusion, data driven and panoramic modeling”, to study the decision-making problems of market participants and policy makers under a big data environment. The key problems that will be studied include: 1) The micro modeling methods for different types of customers based on deep learning and cross domain data fusion. 2) Study the meso models of generators’ and retailers’ behaviors based on transfer learning and inverse reinforcement learning, and study the bidding and pricing problems accordingly. 3) Design hybrid economic experiment methods, and apply them to study the problem of designing the “wholesale - retailing” double layer market. The research outcomes of this project will likely be a theoretical framework of electricity market panoramic modeling and decision making driven by big data. The research outcomes will be applied by the policy makers to the design of electricity wholesale and retail markets.

我国新一轮电力市场改革正在推进，以现货交易为核心的现代电力市场体系即将建立。并且智能电网技术产生的电力大数据赋予了市场参与者精确刻画用户与竞争对手的能力，正深刻改变电力竞价、定价及规则设计的决策范式。然而，传统决策方法基于完全理性等不准确假定做出，难以有效解决智能电网环境下的市场运营问题。为此，本项目立足“自下而上、多源融合、数据驱动、全景建模”思路，研究大数据环境下市场参与者与政策制定者的决策问题，重点研究：1）基于深度学习与多源数据融合的居民、工业等多类型用户的微观建模方法；2）基于转移学习与逆向强化学习的发电与售电企业交易行为中观建模，研究数据驱动的竞价与定价决策；3）设计人机混合实验，研究融合真实交易与实验数据的批发-零售两级市场宏观建模及交易规则设计。研究成果有望形成数据驱动的电力市场“微观-中观-宏观”全景建模及决策理论体系，推动政府部门利用大数据开展电力市场规则设计。

我国新一轮电力市场改革正在推进，以现货交易为核心的现代电力市场体系即将建立。与此同时，智能电网环境下的电力大数据赋予了市场参与者精确刻画用户与竞争对手的能力，正深刻改变电力竞价、定价及规则设计的决策范式。然而，传统决策方法基于完全理性等不准确假定做出，难以有效解决智能电网环境下的市场运营问题。为此，本项目立足“自下而上、多源融合、数据驱动、全景建模”思路，研究大数据环境下市场参与者与政策制定者的决策问题。按照微观(用户个体行为建模)→中观(发电与售电企业交易行为建模与决策)→宏观(市场规则设计)三个层次进行研究。.微观层面，基于深度学习与多源数据融合对居民、工业等多类型用户进行了微观建模。提出的高分辨率感知技术可以有效重建高频数据，从而提高状态估计与负荷识别效果。基于非切入式负荷监测算法对各类用户用电情况进行监测，可以有效获取用户的用电行为，为电价制定提供参考。中观层面，系统研究大数据驱动的发电企业与售电企业的行为模型。基于逆向强化学习对发售电企业在批发市场中的竞价行为，利用数据挖掘技术为电力零售定价提供支持。同时利用逆强化学习对各类企业的风险偏好进行了研究。宏观层面，研究了电力市场的交易规则设计模型。提出了人机混合实验模型，对电力市场进行宏观建模，并应用对抗攻击网络研究了现货市场下燃气机组补贴机制的制定。针对配电市场的零价格竞标行为，提出了以平均成本出清代替边际成本出清的市场机制，以提高市场竞争性。.基于本项目的研究，形成了数据驱动的电力市场“微观-中观-宏观”全景建模及决策理论体系，推动政府部门利用大数据开展电力市场规则设计。