Attention based Convolutional Network for Traffic Flow Velocity Forecasting

Abstract

Accurate and efficient traffic flow velocity prediction is very important for intelligent transportation system. But traffic flow velocity prediction faces the following challenges: the data is dynamic in time and space dimensions, and the error accumulation in long-term prediction. In this paper, we propose an Attention based Convolutional Network (ACN) for traffic flow velocity forecasting. ACN combines convolution with attention mechanism to capture the spatio-temporal features of data and uses the attention output layer to model the correlation between the generated forecast data and the historical data to grasp the important features to optimize the output. Experiments on two real data show that ACN can improve the accuracy of traffic speed prediction.