Traffic prediction with deep learning for vehicular communication network

Abstract

Nowadays, the construction of Internet of vehicles provides a lot of convenience for urban public safety and transportation. Among them, the prediction of traffic flow based on traffic road condition prediction is very important. Due to the interference of many external factors, such as unexpected events and weather, this problem is difficult. In recent years, due to its strong learning ability, deep learning has made many breakthroughs in various image classification and data analysis and processing problems. Among them, the traffic flow prediction method based on deep learning models can provide a set of rich tools, which can make good use of these data to build the Internet of vehicles. In this paper, we compare the deep learning model with the traditional model, and further attempt the preprocessing methods such as Wavelet and Singular Spectrum Analysis (SSA) to improve the performance of the deep learning model. Based on a real International Mobile Equipment Identity (IMEI) data set, we evaluated LSTM, GRU, ARIMA, SVR and ANN models. Considering the problem of whether the deep learning method is suitable for real-time traffic prediction, we counted and compared the training preprocessing time. We found that compared with LSTM and GRU, ARIMA is still competitive and Wavelet or SSA is not helpful as we expected.