Abstract: With the advance of the electricity market reform on the electricity selling side, the number of consumers participating in the demand response is increasing. Due to the different characteristics and great differences in response ability of the consumers participating in the interaction, the combination of multiple types of the consumers results in the complex features of high dimension, nonlinear and non-convex in the overall response characteristics, which brings great challenges to the traditional model-driven interaction modeling and pricing optimization strategies. In this paper, a new data-driven modeling method of the consumer demand response behavior is proposed. Using the environmental meteorological data, the electricity price data and the historical interaction data of the consumer aggregates that do not involve their privacy and combined with the deep LSTM network method, the complex response characteristics of the user aggregates can be effectively characterized. A feature encapsulation model is established to reflect the differentiated demand response capabilities of the consumer aggregates, which is tested in the retail market pricing scenario. The results show that the proposed deep learning encapsulation modeling method can better approximate the theoretical value of the user demand response characteristics and has good accuracy. At the same time, it can significantly reduce the time required for iteration clearance of the retail pricing. This proposed method can provide some reference for the operation of power market with complex consumer participation.