Abstract: By imputing missing data to enhance the quality of traffic flow data, accurate real-time predictions for highway traffic flow can be achieved. This study adopted the low-rank tensor decomposition theory to repair the missing data and used the convolutional long and short term memory neural network based on the attention mechanism to build a prediction model. The actual traffic flow data of the Shandong expressway toll system were used to verify the method proposed in this study. The experimental results showed that: by classifying and identifying missing data, using the confusion matrix to determine the threshold of identifying missing data, introducing the three-dimensional tensor model, and using the low-rank tensor completion method to repair missing data, the data repair effect was the best compared with Lagrange interpolation, KNN, and SVR methods. Compared with SVR, CNN, LSTM, and CNN-LSTM, the combined traffic flow prediction model proposed in this study had higher prediction accuracy in different road sections, working days, and non-working days, and the M_SE and M_APE of traffic prediction decreased by 22.47% and 8.41%, respectively. The average speed forecast for M_SE and M_APE decreased by 42.83% and 6.32%, respectively. The proposed model provided new ideas for research on highway traffic state prediction methods.

Key words: expressways, traffic data repair, spatiotemporal correlation, traffic flow parameter prediction, deep learning