Table 2 Comparison of dominant methodologies for agent-based IES modeling

From: A review of smart integrated energy systems towards industrial carbon neutrality: Opportunity and challenge

Method

Main Idea

Model requirement

Optimality guarantee

Limitations

Typical application

Ref.

Distributed optimization

Decomposing a global problem, iterative convergence to a system-wide optimum.

Requires an explicit mathematical model of the system.

Provable convergence, mathematical rigor.

Sensitive to communication failures, struggles with non-convexity.

Coordinated economic dispatch, voltage/frequency control.

286

Game theory

Modeling strategic interactions between rational, self-interested agents to find an equilibrium.

Less dependent on the physical model than optimization.

Guarantees convergence to an equilibrium, may not be globally optimal.

Assumes perfect rationality, hard to solve the equilibrium for many players.

P2P trading, electricity pricing, and demand response incentives.

287

Multi-agent reinforcement learning

Agents learn optimal policies through interaction with the environment.

Model-free, learns directly from data and interaction.

Generally finds near-optimal policies.

Sample inefficiency, scalability challenges.

Real-time adaptive control, resilience against extreme events.

288

Back to article page