Accurate multi-component load forecasting is helpful to the rational planning and optimal operation of integrated energy systems. Aiming at the problems that input parameters are difficult to determine and model network parameters are difficult to set reasonably in multi-component load forecasting, a collaborative forecasting method of building electricity, cold and heat multi-load is proposed. Firstly, considering the influence of different input parameters on multiple loads, grey relational degree analysis (GRA) is used to calculate the correlation between input parameters and loads, and the parameters with grey correlation degree greater than 0.6 are selected as model inputs. At the same time, Grey Wolf optimization algorithm (GWO) is used to optimize the key network parameters of the short-time memory neural network (LSTM), and the multi-component load forecasting model of GRA-GWO-LSTM is established. Finally, taking a campus in a cold area as an example, compared with the single neural network model and the hybrid neural network model GWO-LSTM, the proposed prediction model has higher prediction accuracy in the long-term prediction of electric, cold and heat loads. Compared with the mean absolute percentage error (MAPE) of the LSTM model and the GWO-LSTM model, it is reduced by 31.64 % and 23.47 % respectively, and it also has good prediction performance for short-term load forecasting, which can be used to guide the planning and intelligent operation of the integrated energy system.