Recurrent and convolutional neural networks for traffic management

Traffic forecasting and monitoring are critical components of intelligent transportation system design. Accurate forecasts and traffic congestion monitoring support many aspects of decision-making by commuters and transportation system officials. Recent advancements in computational technology and availability of different types of multivariate transportation datasets motivate us to utilize deep learning-based approaches to analyze such datasets and improve the robustness of traffic forecasting applications. In this chapter, we discuss traffic monitoring and forecasting based on deep learning methods. Firstly, we present the basics of recurrent neural networks (RNNs) and their usage for traffic time-series data analysis. We offer an introduction to long short-term memory (LSTM) networks and gated recurrent units (GRU), which are considered efficient tools to model time dependency in time series data. Then we address traffic congestion forecasting using RNNs and LSTM for time series data. Later on, we introduce the application of LSTM- and GRU-based models for traffic congestion detection. To this end, we combine the desirable features of deep learning models with sensitivity of double exponential smoothing charts to detect abnormal traffic congestions. In the second part of this chapter, we focus on spatio-temporal traffic flow prediction using convolution neural network (CNN). Firstly, we outlined the basics of CNN. Then we overview the process used to convert traffic networks to space-time matrices used as input for the CNN model for spatio-temporal traffic prediction.