In the distribution network,the participation of flexible resources such as distributed new energy and controllable load in demand response has become an important measure of load adjusting in the new power system. How to consider the uncertainty of user responsiveness and balance the interests of multiple participants is very critical. In this paper,the fuzzy control model of user responsiveness is established firstly,and the benefit function model of multi-agent demand response participants such as power supplier,load aggregator and user is given considering the fuzziness of user responsiveness. Furthermore,with the objectives of minimizing the deviation of daily load curve and minimizing the system cost,the upper optimization adopts the optimal demand response scheme of the power supplier. The lower optimization obtains the best task allocation between the power supplier and the load aggregator,so as to establish the two-layer model of multi-agent collaborative demand response of the power supplier,the load aggregator and the user. A solution method based on Stackelberg game theory and k-means clustering algorithm is proposed. Finally,the simulation results of historical data in an area show that the model can effectively screen high-quality demand response resources to suppress load fluctuation under the consideration of user responsiveness and collaborative of multi-agent interests.