In order to address the supply-demand imbalance in the increasingly complex and changing market environment, improving the accuracy of air cargo volume forecasting is of great significance for route planning and supply chain optimization. Firstly, based on monthly air cargo data from January 2000 to December 2022 as the training set, seasonal fluctuations and long-term trends were captured using seasonal and trend decomposition using loess (STL). Secondly, a deep learning time series prediction model (LSTM-SVR) was used to fit the nonlinear changes in cargo volume due to emergencies. Finally, the prediction model was tested based on monthly data for the entire year of 2023. The results indicate that the seasonal and combination prediction model (STL-SVR-LSTM) is more accurate in predicting air cargo volume during emergencies compared to traditional methods such as ARIMA, SVR, or LSTM. The data validation in 2023 shows that the root mean square error and average absolute percentage error of the seasonal and combination prediction models are 3.53 and 3.53%, respectively, with a goodness of fit score of 0.79. The LSTM model has the second best prediction results, with root mean square error and average absolute percentage error of 5.66 and 7.73%, respectively, and a goodness of fit score of 0.58, significantly better than the other two traditional models. It can be seen that this prediction model can adapt to the prediction of air cargo volume in complex environments, which is helpful in providing reference suggestions for enterprise operation and enhancing supply chain stability in case of emergencies.