The demand response potential of flexible adjustable load cluster has certain characteristics of randomness and spatio-temporal interaction. Therefore, a time-varying demand response potential evaluation method of load cluster based on consumer psychology model and adaptive graph attention network is proposed. Firstly, a demand response model of consumer psychology is established taking account of the randomness of response. Then, according to the power consumption characteristics and model parameters of pilot users in the load cluster, parameter migration and adaptive graph attention network are used to extract the spatio-temporal characteristics of demand response of load cluster. Finally, based on the temporal and spatial characteristics of the cluster and historical response data, the deterministic and uncertain time-varying parameters of the model are estimated respectively, and the time-varying demand response potential of the load cluster is evaluated.