Abstract: This paper addresses the issues of complex and diverse spatiotemporal characteristics, and the insufficiencies in robustness and adaptability of traffic flow. We propose an improved model based on the convolutional neural network (CNN) and long short-term memory (LSTM) network for highway traffic flow prediction. The model aimed to resolve the correlations in time series and spatial network by extracting relevant features and conducting perturbation analysis during the model training process, introducing an error compensation mechanism to enhance the performance of traffic flow prediction. Experimental results indicate that the model can effectively predict traffic flow in highway networks, demonstrating good accuracy and robustness, which holds significant implications for the construction of intelligent transportation systems.