Sensors: Research on Real-time Dynamic Control of Traffic Signal Lights
Traffic signal lights can manage traffic flow at intersections through periodic color changes to reduce traffic congestion. This paper briefly introduced the intersection model under signal light control and the sensor-based signal light dynamic control strategy. The strategy predicted the traffic flow with long short-term memory (LSTM) and adjusted the green duration of different lanes according to the predicted traffic flow to obtain the signal light control scheme. The prediction performance of back-propagation neural network (BPNN) and LSTM on traffic flow was compared in simulation experiments. Moreover, the traditional fixed time control strategy, the BPNN-based dynamic control strategy, and the LSTM-based dynamic control strategy were compared. The results showed that the LSTM algorithm predicted the traffic flow more accurately than the BPNN algorithm; the sensor and LSTM-based signal light dynamic control strategy achieved more passing vehicles and higher vehicle passing efficiency in a single signal cycle compared with the other two control strategies.
Keywords: sensor, signal light, dynamic control, long short-term memory