Traffic Status Prediction of Arterial Roads Based on the Deep Recurrent Q-Learning

Table 3

Gradient parameter value table.

States	Group	Variable parameters					Fixed parameter
		Learning rate	Reward decay	Greedy	Replacement	Memory size	Batch size
Weakened states	1	0.03	0.6	0.6	200	300	32
	2	0.03	0.6	0.6	100	400	32
	3	0.03	0.6	0.3	200	400	32
	4	0.03	0.3	0.6	200	400	32
	5	0.01	0.6	0.6	200	400	32
Initial state	6	0.03	0.6	0.6	200	400	32
Strengthened states	7	0.05	0.6	0.6	200	400	32
	8	0.03	0.9	0.6	200	400	32
	9	0.03	0.6	0.9	200	400	32
	10	0.03	0.6	0.6	300	400	32
	11	0.03	0.6	0.6	200	500	32