随着中国各省电力现货市场与需求响应交易的推进，售电公司与虚拟电厂面临电价波动风险与偏差罚金风险。针对售电与虚拟电厂业务融合的新模式，本文构建基于条件风险价值的多场景风险厌恶模型，提出售电公司与虚拟电厂一体化运营的日前-实时双阶段电力交易决策方法。为同时度量价格波动与尾部亏损的双重影响，所提模型在目标函数中显式引入条件风险价值约束，并拓展为多场景随机规划，可基于原对偶内点法进行求解。目前常用的点预测确定性模型基于单点的电价预测进行优化决策，忽略了市场价格波动和新能源出力偏差等不确定性。相比之下，本文所提的基于条件风险价值的多场景风险厌恶模型考虑了价格分布和多场景风险，既可保留期望收益最大化的动机，又可有效抑制日前-实时价差和需求响应不确定性带来的极端亏损。仿真结果表明，基于条件风险价值的多场景风险厌恶模型可在不完美信息的条件下将期望收益提升至理论上限，同时将风险价值、条件风险价值由亏损转为收益，极大减少了尾部亏损。结果验证了基于条件风险价值的多场景风险厌恶模型在广东日前-实时双阶段交易中的收益-风险双优性能，可为售电公司与虚拟电厂一体化运营的电力交易决策与风险管理提供技术支持。

With the advancement of the electricity spot market and demand response trading in various provinces of China, electricity retailers and virtual power plants face the risk of electricity price fluctuations and deviation penalties. In response to the new pattern of the integration of electricity sales and virtual power plant business, this paper constructs a multi-scenario risk aversion model based on conditional value at risk (CVaR), and proposes a day-ahead and real-time two-stage power trading decision-making method for the integrated operation of the electricity retailer and virtual power plant. To simultaneously measure the dual impacts of price fluctuations and tail losses, the proposed model explicitly introduces CVaR constraints in the objective function and extends to multi-scenario stochastic programming, which can be solved based on the primal-dual interior-point method. The currently commonly used deterministic point prediction model is optimized based on single-point power price predictions, ignoring uncertainties such as market price fluctuations and deviations in renewable power generation. In contrast, considering the price distributions and multiple scenarios of risks, the proposed multi-scenario risk aversion model based on CVaR not only retains the motivation of maximizing expected returns, but also effectively suppresses extreme losses caused by day-ahead and real-time price differences and uncertainties in demand response. The simulation results show that the multi-scenario risk aversion model based on CVaR can increase the expected return to the theoretical upper limit in the absence of perfect information, and at the same time convert the loss of the VaR and the CVaR into a return, greatly reducing the tail loss. The results verified the dual-excellent performance of return and risk of the multi-scenario risk aversion model based on CVaR in the day-ahead and real-time two-stage trading in Guangdong, which can provide technical support for the power trading decision-making and risk management of the integrated operation of the electricity retailer and virtual power plant.