Application of Key Node Identification Based on Revised Penalized Local Structural Entropy in Traffic Flow Prediction

In intelligent transportation systems (ITS), traffic flow prediction is a necessary tool for effective traffic management. By identifying and extracting key nodes in the network, it is possible to achieve efficient traffic flow prediction of the whole network using “partial” nodes, as the key nodes contain essential information about changes in the state of the traffic network. This paper proposes a key node identification method based on revised penalty local structure entropy (RPLE) for specific traffic networks. This method takes into account the influence of node distance and traffic flow on identifying important nodes within the traffic network. By introducing a modified penalty term and a comprehensive weight, it achieves a certain level of accuracy in traffic flow prediction using data from key nodes in the network. We compared the RPLE method with different key node identification methods and combined it with different prediction models to compare the traffic flow prediction performance under key node coverage rates of 50% to 100%. Experimental results on the PeMS04 dataset show that using the RPLE method to screen out the top 60% of key nodes, the importance of these nodes accounts for over 80% of the overall importance. As the coverage rate of key nodes increases, compared with other important node identification methods, the key nodes selected using RPLE have consistently shown the best prediction performance, with a maximum reduction of 14.93% in MAE and 8.33% in MAPE. Moreover, when the coverage rate of key nodes reaches 75%, the prediction accuracy of traffic flow has already reached 80%. Therefore, it is possible to consider using “partial” key nodes for predicting the overall traffic flow of the whole traffic network.