Multi-Agent Reinforcement Learning Results for Power Router Energy Management

Experimental results from a study integrating Proximal Policy Optimization with a multi-agent framework for power router energy management. The 5.5 KB XLS file contains metrics like average round reward, strategy loss, DC bus voltage, electricity cost, and runtime from 420 training episodes. Author Junyan Lyu published the data on figshare in April 2026.

Use Cases

• Benchmarking multi-agent reinforcement learning algorithms using the reported average round reward and strategy loss stabilization metrics.
• Analyzing the impact of a double-buffer mechanism on training stability and data utilization efficiency as demonstrated in the results.
• Comparing the performance of the proposed model against baseline models using the provided DC bus voltage fluctuation range, electricity cost, and total runtime.
• Validating Generative Adversarial Imitation Learning applications in energy systems with the 500-episode strategy loss convergence data.

Strengths

• Results include specific performance metrics: average round reward stabilized at -410, DC bus voltage between 728V and 732V.
• Clear experimental scope with 420 training episodes documented.
• Published under a permissive CC-BY-4.0 license.

Limitations

• Extremely small 5.5 KB size indicates limited raw data, likely only summary results.
• No raw state-action trajectories or environment code provided, limiting reproducibility.
• Single experimental scenario results may not generalize to other power system configurations.

Provenance

Source

Junyan Lyu via figshare.

Collection Method

Generated from simulation experiments of a multi-agent reinforcement learning model for power router control.

Time Range

Covers the experimental period of the described study.

Freshness

Last updated in April 2026, indicating projected future relevance.

Geography

null