Considering the demand response mechanism and the dynamic coupling characteristics of multi loads in integrated energy systems, a multi load forecasting model based on an AGCN-TimeXer framework that incorporates integrated demand response is proposed. Firstly, drawing on consumer psychology theories, the electricity demand response signals are modeled and their uncertainty is quantified. Comprehensive demand response signals are then constructed by calculating cooling and heating load responses using the coupling response principle. Secondly, an adaptive graph convolutional network (AGCN) is employed to construct a dynamic adaptive graph structure, thereby facilitating the extraction of dynamic coupling relationships among multi loads. Finally, the self-attention mechanism of TimeXer is utilized to explore the temporal dependence characteristics of load data, and the cross-attention mechanism is adopted to analyze the relationships between multivariate loads and external influencing factors, including integrated demand response signals, dynamic coupling relationships among multivariate loads, and meteorological factors. Simulation results demonstrate that the proposed model significantly outperforms traditional forecasting methods in terms of prediction accuracy.