Adaptive Traffic Signal Control Model on Intersections Based on Deep Reinforcement Learning

Table 5

Performance in multi-intersection case. Travel time: the lower the better; other measures: the higher the better.

SL no.	Method	Reward	Average travel time (s)	Average speed (m/s)
1	DQN (ours)	2.54	438.26	2.49
	DQN (base)	2.42	486.95	2.21
	Q-learning	1.49	752.17	1.28
	LQF	2.37	496.80	1.93
	Webster	2.05	528.13	1.88
2	DQN (ours)	2.74	418.11	2.57
	DQN (base)	2.51	498.69	2.16
	Q-learning	1.68	701.82	1.38
	LQF	−0.02	816.29	1.18
	Webster	1.71	644.27	1.57
3	DQN (ours)	2.78	391.01	2.64
	DQN (base)	2.24	573.16	1.83
	Q-learning	1.71	681.12	1.44
	LQF	1.28	827.84	1.15
	Webster	1.75	588.22	1.71