To address the issues that the existing traffic status prediction models cannot handle the fuzziness of highway traffic status effectively and fail to utilize real-time data streams after model training， a real-time prediction method for traffic status based on clustering multivariate time series model was proposed. Firstly， traffic flow parameters were analyzed， and a classification model based on the improved Fuzzy C-Means （FCM） algorithm and entropy weight method was developed to define traffic status and establish classification standards， and a Status Index （SI） indicator was employed to address the issue of classification boundary fuzziness. Secondly， a multivariate time series prediction model was constructed on the basis of the classification model. By combining convolutional networks and attention mechanisms， this model was able to capture both long-term and short-term dependencies in time series data effectively. Thirdly， a back-propagation update mechanism was applied for online learning to realize real-time prediction. Finally， the model was tested on the California Traffic Management Center Performance Measurement System （PeMS） dataset， the dataset was divided into training， validation， and test sets in time order， and real-time data stream simulations were conducted for online learning and prediction. Experimental results show that with the prediction step of 6， compared to the classic LightTS （Light Sampling-oriented MLP Structures） model， the proposed model reduces the Mean Squared Error （MSE） by 22.81% and the Mean Absolute Error （MAE） by 14.64%. It can be seen that the proposed model distinguishes traffic status levels effectively and achieves real-time traffic status prediction.