A multi-agent reinforcement learning approach to heterogeneous resource allocation in lifelong education

Lifelong education faces challenges in resource allocation due to heterogeneity, dynamism, and scalability.This study proposes a distributed allocation model using multi-agent reinforcement learning (MARL), where learners and providers act as autonomous agents.Employing a centralised training with decentralised execution (CTDE) paradigm, the model applies the multi-agent deep deterministic policy gradient algorithm for collaborative learning.A composite reward function integrates user quality of experience (QoE), system cost, and fairness.A tripartite stochastic game model theoretically confirms the existence of a Nash equilibrium.Simulations show the model outperforms baseline algorithms, achieving superior overall system utility (37.1% higher), average quality of experience (41.8% higher), resource utilisation (76.3%), and fairness (Jain's index = 0.89), with strong convergence and adaptability.This provides an efficient, scalable solution for heterogeneous resource management in lifelong education.