Research Article
“Fed-DRL”: A Timeliness Optimization Method for Dynamic Data Acquisition System Based on Mobile Edge Computing
| (1) | Algorithm 1 Fed-DRL | | (2) | Initialization: Initialize system parameters and hyperparameters for learning. | | (3) | for = 1,2, …, 5000 do | | (4) | Reset the environment for each agent, get local state ; | | (5) | Randomly generate ; | | (6) | for each agent n in 1,2, …, N do | | (7) | if then | | (8) | Randomly choose action ; | | (9) | else | | (10) | Generate actions ; | | (11) | end if | | (12) | end for | | (13) | The resulting action interacts with environment, generate and ; | | (14) | Add of each agent into | | (15) | for each agent n in 1,2, …, N do | | (16) | Sample from ; | | (17) | Calculate using the critic network; | | (18) | Predict using the target actor network; | | (19) | Calculate using the target critic network; | | (20) | Update the actor network according to equation (20); | | (21) | Update the critic network according to equation (21); | | (22) | end for | | (23) | if mod = = 1 then | | (24) | Update the target actor network and the target critic network with following method; | | (25) | | | (26) | | | (27) | end if | | (28) | if mod = = 1 then | | (29) | Run edge-federated updating according to equations (23) and (24); | | (30) | end if | | (31) | end for |
|
