A Short-term Traffic Flow Prediction Approach of Neural Network Based on Cluster Analysis

Abstract

Given the principle of different dates having different traffic flow patterns, the research clusters the historical data based on the results of clustering, predicts the class, which flow pattern the date is in, and then outputs all data of this pattern to forecast the date’s traffic flow. On the choice of clustering method, we uses k-means algorithm. Since the k-means algorithm is not unity in the choice of the optimal clustering number, we combine KNN algorithm to determine the optimal cluster number k, then use it as the classification method, and choose BP neural network to predict short-term traffic flow. The experiment proved that the approach is better than that of without clustering analysis in prediction accuracy and the method of k-means algorithm combining with KNN algorithm to determine the optimal cluster number k is effective.