Data transmission optimization in edge computing using multi-objective reinforcement learning

Reducing network energy consumption and balancing workload are two key optimization goals for data transmission in edge computing field. However, these two goals are likely to be conflicting in some cases and fail to achieve the optimum simultaneously. In this paper, we design a new data transmission optimization algorithm using multi-objective reinforcement learning. We design the vector of rewards for the two objectives, and update Pareto approximate set by multiple state steps to approach the optimal solution. In every step, we classify the candidate links into four different levels for path selection. We aggregate network traffic to construct minimum topology subset, minimizing the number of occupied device to reduce energy consumption. We optimize the load distribution on those selected links, minimizing maximum congestion factor to balance workload. For action selection, we leverage roulette-based Chebyshev scalarization function to solve the weight selection problem for multi-objectives and enforce exploration to avoid falling into local optimum. To improve the convergence rate, we design heuristic factor to control the search of solution space and enhance the guiding effect of the existing optimal solution. Simulation result shows that the proposed algorithm achieves good performance in energy-saving and load balance at the same time.