A Semi-“Smart Predict, Then Optimize” Method for Traffic Signal Control

An efficient intersection signal scheme is of vital significance to urban traffic operation. At present, multiperiod fixed-timing control is still the traffic signal control method adopted by many urban intersections. For the optimization problem of signal scheme selection at signalized intersections, we proposed three two-step prediction optimization methods that match the traffic arrival in different periods with the corresponding optimal signal scheme, aimed at reducing the total delay of vehicles at signalized intersections. The first method predicts the traffic flow in each entrance direction of the intersection by minimizing the mean square error (MSE), then obtains the total vehicle delay of each scheme in the signal scheme set through the Highway Capacity Manual 2010 delay formula, and finally substitutes it into the signal optimization model to obtain the optimal scheme combination. The second method directly predicts the total vehicle delay at intersections of all signal schemes by minimizing the MSE and then substitutes it into the signal optimization model to obtain the optimal scheme combination. The third method directly predicts the total vehicle delay at intersections of all signal schemes by minimizing both the MSE and the prediction error between every two schemes and then substitutes it into the signal optimization model to obtain the optimal scheme combination. Verification based on actual intersection shows that better optimization results can be obtained by integrating optimization objectives into the prediction process. Besides, some practical insights can be drawn through applicability analysis and sensitivity analysis. First, the proposed model is more suitable for intersections where traffic arrivals vary greatly. Second, with the increase of scheme differences within a certain range, the advantages of the proposed method become more obvious. Finally, it is necessary to balance the flexibility of traffic control with control effectiveness in practical applications.