A Fuzzy-based Convolutional LSTM Network Approach for Citywide Traffic Flow Prediction

Citywide traffic flow prediction is of great importance to intelligent transportation systems and smart cities. Although many deep learning methods have been applied for citywide traffic flow prediction, deep learning is a deterministic representation and sheds little light on data uncertainty. In this paper, a fuzzy-based convolutional LSTM neural network (FConvLSTM) method is proposed to improve the accuracy of citywide traffic flow prediction by taking data uncertainty into consideration. FConvLSTM is a hybrid model which combines fuzzy learning with a convolutional LSTM neural network (ConvLSTM). The impact of data uncertainty is lessened with the help of fuzzy neural networks and ConvLSTM is adopted to explore the spatio-temporal characteristics of traffic data, which can learn spatial dependencies and temporal dependencies jointly. Experimental results on a real dataset verify the outperformance of the FConvLSTM method.