Abstract: In order to deal with the energy crisis and improve the energy utilization, the integrated energy system (IES) has become an important direction for developing the innovative energy systems. The accurate IES multivariate load forecasting is of great importance for the economic dispatch and optimal operation of the IES, and the coupling characteristics of the IES multivariate loads may be further explored by the chaos theory. In this paper, an IES ultra-short-term electrical, cooling and heating load forecasting model based on the combination of the multivariate phase space reconstruction (MPSR) and the radial basis function neural network (RBFNN) is proposed. First, the coupling relationship between the energy subsystems in the IES is analyzed, and the Pearson correlation analysis is applied to quantitatively describe the correlation between the electrical, cooling and heating loads and the meteorological characteristics. Then, the C-C method is used to perform the MPSR on the time series to further reflect the coupling characteristics of the electrical, cooling and heating loads and the meteorological characteristics on time. Finally, the RBFNN model is used to learn the coupling relationships among the multivariate loads for forecasting. The experimental results show that the proposed method effectively explores and learns the coupling characteristics of the multivariate loads on time, and has a good and stable forecasting effect under different sample sizes.