Federated Reinforcement Learning-Based Dynamic Resource Allocation and Task Scheduling in Edge for IoT Applications

. 2025 Mar 30;25(7):2197. doi: 10.3390/s25072197

Algorithm 2 Task Offloading Algorithm

Input: Workload matrix

W (t)

representing the resource requirements of tasks over time and Edge server information matrix EdgeServerInfo

Output: Predicted data, Energy consumption matrix, Delay matrix, Cost matrix

1:
Initialize $Counter = 0$
2:
Load a pre-trained model for predicting task resource requirements
3:
for each edge server i do
4:
Calculate $P_{i} = Predict (D_{i})$
5:
end for
6:
for each edge server i do
7:
Calculate $F_{i} = Combine (P_{i}, E_{i}, Δ_{i}, C_{i})$
8:
end for
9:
Sort edge servers based on the decision factor: $SortedServers = Sort (F_{i})$
10:
for each service do
11:
if $F_{i} < Threshold$ then
12:
Process the service locally
13:
else
14:
Migrate the service to the selected edge server
15:
end if
16:
end for
17:
for each edge server i in $SortedServers$ do
18:
if the edge server has enough capacity to host the current service and the service is not already hosted then
19:
$Provision (i)$
20:
Increment $Counter$
21:
Exit the loop to ensure the service is migrated only once
22:
end if
23:
end for
24:
if $Counter = N$ then
25:
Halt the process
26:
end if