Service Pricing and Networked Microgrid Trading for Heterogeneous Demand: A Causal Diffusion and Multi-Agent Reinforcement Learn

Dynamic pricing is an important means to maximize revenue and balance customer satisfaction in service marketing and distributed energy markets. Because demand functions are unknown and exhibit time-varying heterogeneity, traditional pricing methods based on parametric models have difficulty adapting to complex environments. In recent years, reinforcement learning, causal inference and diffusion models have emerged in pricing problems, but existing studies mostly focus on a single method or static scenarios, ignoring the challenges posed by heterogeneous demand and causal structures. This paper proposes a service pricing framework that integrates a causal diffusion model, nonparametric Bayesian techniques and multi-agent reinforcement learning. A conditional diffusion model is used to learn the demand distribution under price interventions, a Dirichlet process realizes customer clustering, and multi-agent reinforcement learning optimizes dynamic pricing strategies. To address the complexity of distributed energy trading, the framework is further extended to a networked microgrid environment; by combining local optimization, peer-to-peer transactions and a system marginal pricing mechanism, dynamic bidding within and outside microgrid clusters is coordinated. Simulation results show that this method outperforms traditional approaches in revenue, fairness and stability. In microgrid simulations, the framework reduces electricity sales by about 43.9\\%, decreases the cost of electricity purchased from the grid by about 7.5\\%, and reduces the loss of electricity sales revenue by about 44.6\\%, thereby improving energy self-sufficiency and market efficiency. Finally, this paper discusses research limitations and future directions.