To facilitate the rapid calculation of fatigue damage and accurate assessment of the remaining life of the integrated transmission system， a real-time evaluation model for the running load of the transmission main shaft under multiple operating conditions is proposed. The model first extracted characteristic parameters of typical operating conditions for special vehicles and realizeed real-time discrimination of these conditions using K-means clustering and support vector machine methods. Then， a transmission main shaft load evaluation model based on convolutional neural networks-long short-term memory was constructed. Bayesian algorithm was employed to optimize hyperparameters， such as learning rate and network depth， aiming to improve the accuracy of the transmission main shaft load evaluation. Finally， the model was validated through experimental data collected under typical operating conditions of special vehicles. The results show that under shifting， steering， and climbing conditions， the average relative errors of the model’s load evaluation for the transmission main shaft are 0.150， 0.014， 0.006 （5-degree slope）， and 0.004 （10-degree slope）， respectively， indicating the effectiveness of the model in assessing the load of the integrated transmission system under variable operating conditions.