Power Router Optimization Model Results and Metrics

420 training runs were conducted to develop a multi-agent reinforcement learning model for collaborative energy optimization in power routers. The model, proposed by Junyan Lyu, was last updated in April 2026 and is documented in a 9.5 KB XLS file. It demonstrates performance metrics including a stable average reward and reduced operational costs.

Use Cases

• Analyze the relationship between training iteration count and average round reward, which stabilized at approximately -410 after 420 runs.
• Compare strategy loss function convergence across models, noting it stabilized first after 500 iterations in the proposed model.
• Evaluate DC bus voltage stability in practical scenarios using the recorded fluctuation range between 728V and 732V.
• Benchmark operational economics by comparing the recorded electricity cost of 3846.36 yuan and total runtime of 53.32 seconds against other models.

Strengths

• Model performance is validated with specific numerical results: a DC bus voltage range of 728V-732V and a cost of 3846.36 yuan.
• Training process is detailed with concrete metrics: 420 training runs leading to a stable average reward of about -410.

Limitations

• The dataset is extremely small at 9.5 KB, indicating it likely contains only summary results, not the full training or simulation data.
• Without access to the raw XLS file structure, the specific features, columns, and sample size used for model validation are unknown.

Provenance

Source

Junyan Lyu via figshare.

Collection Method

Results from a computational study proposing a reinforcement learning model for power router optimization.

Time Range

null

Freshness

Last updated in April 2026, indicating recent model documentation.

Geography

null