Table 1 Related works comparison

Deep Learning

[19,20,21].

The method can learn and optimize complex strategies

The implementation of the method requires large amounts of data and computational resources and poor interpretability.

High/High

Deep Reinforcement Learning

[33, 34, 43].

The method can learn and optimize complex strategies; it is suitable for tasks that are not explicitly labeled.

The implementation of the method requires large amounts of data and computational resources and poor interpretability.

High/High

Swarm Intelligence Algorithm

[17, 29, 30].

The method can find approximate solutions to complex optimization problems; it has good robustness and adaptability.

The method may fall into a local optimum; when solving complex problems, specific parameters need to be set manually to obtain a better strategy.

Mid/Low

Mathematical Optimization

[16, 18, 44].

The method exists efficiently for convex optimization problems and provides deterministic and theoretical guarantees.

The method can be very complex when dealing with high-dimensional and non-convex optimization problems; the solution may fall into local optimal solutions for complex problems.

Low/Mid

Dynamic Game Theory

[15, 45, 46].

The algorithms model the interactions between multiple decision makers; equilibrium concepts (e.g., Nash equilibrium) are provided to predict the players’ strategic choices under certain conditions.

The method is more complicated to find certain theoretical game equilibria in certain complex situations; it is still a challenge to deal with time-varying strategies and decisions in context.

Mid/Low

Greedy Strategy

[26, 47, 48].

The method is simple and easy to implement; optimal solutions can be obtained for some problems.

The method can only obtain approximate solutions for many problems; it is easy to fall into local optimization.

Low/Low

Lyapunov Method

[28, 31, 49].

The method provides an explicit and rigorous determination of the stability of a system; it applies to both linear and nonlinear systems; it is commonly used to stabilize target queues.

The method can be challenging to find suitable Lyapunov functions, proving the existence of upper bounds on Lyapunov drift. The method may be inapplicable or very difficult for some highly nonlinear or complex systems.

Mid/Low