Table 6 Multi-dimensional comparison of methods.
From: An integrating RAG-LLM and deep Q-network framework for intelligent fish control systems
Feature / Method | Human | RAG-LLM | DQN | RAG-LLM + DQN |
|---|---|---|---|---|
Fish Growth | Baseline | \(-2.0\%\) vs. expert | \(\sim 1\%\) below | \(+1.8\%\) above |
Automation | None | Medium (rules) | High (policy) | Very High (ensemble) |
Error Rate | Subjective | \(\sim 8\%\) | \(\sim 5\%\) | \(<2\%\) |
Explainability | High (manual) | High (text) | Low (black-box) | High (LLM-aided) |
Stability | N/A | Medium | Medium | High |
Intervention Rate | Continuous | \(\sim 10\)–\(12\%\) | \(\sim 5\)–\(7\%\) | \(\sim 5\)–\(8\%\) |
Feedback Adaptation | Medium | Low | High | High |
Scalability | Low | Medium | High | Very High |
