Abstract

With the gradual increase in the number of vehicles owned by people, the problem of urban traffic congestion is becoming more and more serious. Timely and accurate traffic flow prediction helps travel vehicles to plan their routes reasonably, avoid congested roads and reduce environmental pollution. In order to take full advantage of the daily periodicity of the traffic flow of a road section, K-means++ with improved initial clustering centroids is applied to analyze the similarity of the daily traffic flow patterns and classify the traffic flow patterns with day as the time interval. One can separate the traffic patterns affected by different factors, and the other can exclude individual days containing abnormal data trends formed by unexpected road conditions, which can save the deep learning model from the interference of other irrelevant data features. Finally, a Long Short-Term Memory (LSTM) model is combined to predict the traffic flow value at the next time point based on the data of the previous hour. In the experiment, the K-means-LSTM model can significantly reduce the prediction error of short-time traffic flow and has better prediction performance by comparing with the prediction results of the support vector machine (SVR) model optimized by genetic algorithm (GA) and the LSTM model.

Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call