| Popis: |
The prediction and control of short-term traffic flow have become one of the key fields of urban traffic research. However, there are some issues to be addressed, such as the low accuracy of current urban road short-term traffic flow prediction and the high traffic congestion rate resulted from most urban road traffic signal control methods. In this paper, we propose a short-term traffic flow prediction method based on graph convolution neural network and Seq2Seq (Sequence to Sequence) model, which can excavate the spatial and temporal relationships among road traffic flows and jointly perform multi-step predictions to realize a more accurate prediction of short-term traffic flows. In addition, a regional traffic signal control method based on the back-pressure algorithm is proposed, which is based on single-point traffic signal control to establish a distributed regional traffic signal coordination control model, thus effectively realizing dynamic traffic control with a lower congestion rate. Compared to other related methods, we compare and simulate the current short-term traffic flow prediction and urban road traffic signal control methods via actual experiments to demonstrate the practicability of our method. Based on the two method, we conduct two experiment using real world data and compare our method with the current traffic flow prediction and traffic signal control method. After that, we get three conclusions. First, our short-term traffic flow prediction method is clearly outperformed than the LSTM algorithm and ARIMA algorithm in terms of root mean square error, average absolute error and average absolute percentage error. Second, compared with the timing traffic signal control method, our regional traffic signal control method greatly reduces the average waiting time, average travel time and average number of waiting vehicles. Besides, compared to the single-point traffic signal control method, the experiment presents that our method makes a significant improvement in reducing the traffic congestion rate. |
