loading page

A DBN-BILSTM Short-Term Traffic Flow Prediction Model Based on Variational Mode Decomposition
  • Guowen Dai
Guowen Dai
Central South University School of Traffic and Transportation Engineering

Corresponding Author:[email protected]

Author Profile

Abstract

In the intelligent traffic management system, it is necessary not only to grasp the real-time traffic flow status, but also to understand the future traffic change and development trend, and the future traffic data can be obtained through short-term traffic flow forecasting. In this paper, a short-term traffic flow forecasting model based on DBN-BILSTM combination model based on variational mode decomposition (VMD) is proposed. The prediction method proposed in this paper is reflected in two aspects: data decomposition and model optimization: 1. In terms of data decomposition, this paper uses variational mode decomposition (VMD) to decompose the traffic flow time series data into multiple modal components (IMF), so that the relatively stationary input data can be obtained; 2. In terms of model, this paper combines the advantages of deep belief network (DBN) and two-way short-term memory network (BILSTM) to propose DBN-BILSTM short-term traffic flow forecasting model. Variational Modal Decomposition (VMD) decomposes the original traffic flow sequence, and then pre-trains it through a deep belief network to extract and reduce the dimension of spatial features, which greatly reduces the time required for model learning; At the same time, the bidirectional LSTM neural network is added to optimize the problem that the deep belief network is difficult to capture the long-term dependencies in the time series data: the bidirectional LSTM neural network is an extension of the traditional LSTM. It can further improve the accuracy of prediction results and make the network learn faster and more fully. Finally, experiments are carried out on the model with actual data, and through comparative analysis, it is proved that the short-term traffic flow forecasting model based on variational mode decomposition (VMD) DBN-BILSTM combination model is reliable in terms of prediction accuracy and stability.