Abstract: This application is a transportation network speed forecasting method using deep capsule networks with nested LSTM models. The method includes the following steps: (1) This method divides the transport network into road links, calculates average speeds of each road link, maps the average speeds into a grid system, and generate traffic images representing traffic state at time intervals; (2) the method uses a CapsNet to capture the spatial relationship between road links. The learn patterns are represented in vectors; (3) The vectors of CapsNet are feed into a NLSTM model to learn temporal relationships between road links; (4) The model is trained using and training dataset, and predicts future traffic states using testing dataset. This application uses a new and advanced CapsNet neural structure, while can more efficiently deal with complex traffic networks than CNN models.

Abstract: This application is a transportation network speed forecasting method using deep capsule networks with nested LSTM models. The method includes the following steps: (1) This method divides the transport network into road links, calculates average speeds of each road link, maps the average speeds into a grid system, and generate traffic images representing traffic state at time intervals; (2) the method uses a CapsNet to capture the spatial relationship between road links. The learn patterns are represented in vectors; (3) The vectors of CapsNet are feed into a NLSTM model to learn temporal relationships between road links; (4) The model is trained using and training dataset, and predicts future traffic states using testing dataset. This application uses a new and advanced CapsNet neural structure, while can more efficiently deal with complex traffic networks than CNN models.