A Deep Learning Based Traffic State Estimation Method for Mixed Traffic Flow Environment

Traffic state estimation plays a fundamental role in traffic control and management. In the connected vehicles (CVs) environment, more traffic-related data perceived interacted by CVs can be used to estimate state. However, when there is low penetration rate of CVs, collected from would inadequate. Meanwhile, representativeness positively correlated with rate. This article presents method based on deep learning algorithm under dynamic environment. Specifically, we design K-Nearest Neighbor (KNN) filling model integrating acceleration solve problem insufficient data. fuse time feature speed modification mine distribution features KNN. addition, reduce error caused rate, Long Short-Term Memory (LSTM) model, which uses estimated Macroscopic Fundamental Diagram (MFD) as one input factors. Finally, use concept operational efficiency for reference, dividing into three categories according speed: free flow, optimal congestion. SUMO simulate cases different rates evaluate our scheme. The results suggest that significantly improve accuracy rate; also better performance than other comparison models both rates.