A Novel Theoretical Framework for Multi-Agent Reinforcement Learning-Based Coordination of Heterogeneous Energy Resources in Nigerian Distribution-Level Microgrids

Distribution-level microgrids in Nigerian states increasingly incorporate heterogeneous energy resources including solar photovoltaic systems, diesel generators, battery storage, and demand response loads with complex stochastic generation and consumption dynamics that strain the optimization capabilities of centralized control architectures. This dissertation develops a novel theoretical framework for multi-agent reinforcement learning-based decentralized coordination of heterogeneous energy resources in Nigerian distribution microgrids, addressing the specific challenges of partial observability, non-stationary agent strategies, reward sparsity under episodic renewable outages, and communication-constrained coordination. A cooperative multi-agent Markov decision process formulation is proposed in which each distributed energy resource is represented as an autonomous learning agent with a local observation space and action set. A novel consensus-based policy gradient algorithm is derived that enables distributed agents to converge to a coordinated near-optimal global energy management policy using only local observations and limited neighbour communication, with theoretical convergence guarantees derived under mild assumptions on the network communication topology. The framework is validated in a high-fidelity simulation environment parameterized with distribution network topology, renewable generation, and load demand data from a 33 kV distribution feeder in Ibadan, Oyo State. Against a centralized model predictive control benchmark, the proposed MARL framework achieves within 4.8 percent of optimal energy cost while requiring 94 percent less communication bandwidth. Robustness to communication link failure is demonstrated through degraded-communication experiments. Keywords: multi-agent reinforcement learning, microgrid, energy management, distributed control, Nigeria