A neural network for traffic flow prediction with parallel processing of expanded convolutional and radial networks

Considering that traffic flow data are affected by multiple aspects of extrinsic and intrinsic factors, traditional time series analysis models may suffer from low accuracy. Temporal Convolutional Networks (TCNs) are widely used in traffic flow prediction due to their unique parallel computational extraction capability and ability to fix the sensory field. However, temporal convolutional networks are unable to cope with the initial filtering nature of the raw data, resulting in many raw data features being ignored and their own weak ability for real-time processing. To solve these problems, this study proposes a new model based on radial temporal convolutional networks (RSCN). We effectively utilize the multi-layer processing mechanism of TCN by adding a radial network and inflated convolution for parallel processing, and finally, extracting the feature data matrix using the residual layer and dimensionality reduction to obtain the output layer. Experimental results on real datasets show that RSCN is more efficient than several existing methods.