Power Router Energy Optimization Ablation Results

420 training runs of a multi-agent reinforcement learning algorithm for power router energy management are documented. The dataset contains experimental results from an ablation study comparing an improved Proximal Policy Optimization algorithm with a double-buffer mechanism against other models. It was authored by Junyan Lyu and uploaded to figshare in April 2026.

Use Cases

• Analyze the relationship between training iteration count and average round reward, which stabilized around -410 after 420 runs.
• Compare strategy loss function convergence across algorithms, noting the proposed model stabilized first after 500 iterations.
• Evaluate practical performance metrics like DC bus voltage fluctuation, recorded between 728V and 732V.
• Benchmark economic and runtime efficiency using the reported electricity cost (3846.36 yuan) and total runtime (53.32 seconds).

Strengths

• Results are derived from 420 explicit training runs, providing a concrete experimental basis.
• Reports multiple specific performance metrics: reward, loss, voltage range, cost, and runtime.

Limitations

• The dataset is very small at 5.5 KB, suggesting it contains only summary results, not the full training logs or raw interaction data.
• The scope is limited to a single ablation study, preventing generalization to other power system configurations.

Provenance

Source

figshare, authored by Junyan Lyu.

Collection Method

Results from a computational ablation experiment on a proposed reinforcement learning model.

Time Range

Experimental timeframe not specified; results published in 2026.

Freshness

Last updated in April 2026, indicating recent experimental work.

Geography

null