TABLE 2.

Parameter settings.

Notation	Description
T = 100	Total time slots
E = {60, 70, 80, 90, 100}	The total number of VMs of the edge node
$d_t\sim N({\mu }_a,{\sigma }_a^2)$	The normal distribution of the number of VMs requested by the user
$l_t\sim N({\mu }_l,{\sigma }_l^2)$	The normal distribution of the computing duration requested by the user
p e = 0.03	Standby cost of one VM in the edge node
p f = 0.2	Computing cost of one VM in the edge node
p c ∼ U (1, 5)	The uniform distribution of the computing cost of private cloud
p od = 3.0	Unit price of on-demand instance in public cloud
p upfront = 800	Customization price of reserved instances in public cloud
p re = 1.5	Unit price of reserved instance in public cloud
T r = 20	Reserved duration of reserved instance in public cloud
p t ∼ N (1.5, 1)	Normal distribution of unit price of spot instance
T m = 6	Maximum service duration of spot instance
γ = 0.99	Discount factor
τ = 0.001	Soft update parameters of target network
α 1 = 0.0001	Learning rate of Actor
α 2 = 0.00001	Learning rate of Critic
p = 100,000	Size of experience pool
m = 128	Size of the batch sample in the experience pool