Large-Scale Multi-Agent Learning-Based Cloud–Edge Collaborative Distributed PV Data Compression and Information Aggregation for Multimodal Network in Power Systems

To ensure the large-scale integration of distributed photovoltaic (PV) into power system, it is imperative to collect and aggregate real-time operational data. However, the substantial volume of data imposes significant stress on cloud-edge collaborative framework and multimodal networking integrated with power line communication (PLC) and high-speed radio frequency (HRF) communication, resulting in large delay and poor transmission reliability. In this paper, we formulate an optimization problem to minimize the weighted sum of information aggregation delay and packet error rate by jointly optimizing data compression and routing selection. We propose a large-scale multi-agent learning-based two-stage service priority and compression speed-aware data compression and information aggregation joint optimization algorithm to address the problem. The proposed algorithm augments conventional ant colony method with Q-learning to enhance the learning capability under large-scale complex dynamic systems. Furthermore, the incorporation of service priority and data compression speed awareness leads to improved convergence speed. Finally, through simulation verification, we demonstrate that the proposed algorithm outperforms existing methods.