Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Models based Multi-Agent Framework for Objective Oriented Control Design in Power Electronics

Chenggang Cui, Jiaming Liu, Junkang Feng, Peifeng Hui, Amer M. Y. M. Ghias, Chuanlin Zhang

TL;DR

This work introduces an LLM-based multi-agent framework for objective-oriented control design in power electronics, addressing challenges from parameter uncertainties and costly design cycles. By decomposing the process into specialized agents (objective design, model design, control algorithm design, parameter optimization, and verification) coordinated by a Manager, the approach translates high-level objectives into implementable controllers via Modelica models, PID/MPC options, PSO-based parameter tuning, and Gym-based verification. The framework's initial implementation demonstrates end-to-end automation and collaboration among agents, validated on a DC-DC boost converter where prompts yield controller designs, parameters, and performance indicators. Overall, the method promises greater flexibility, reduced design effort, and autonomous, constraint-compliant controller development for power electronics in modern grids.

Abstract

Power electronics, a critical component in modern power systems, face several challenges in control design, including model uncertainties, and lengthy and costly design cycles. This paper is aiming to propose a Large Language Models (LLMs) based multi-agent framework for objective-oriented control design in power electronics. The framework leverages the reasoning capabilities of LLMs and a multi-agent workflow to develop an efficient and autonomous controller design process. The LLM agent is able to understand and respond to high-level instructions in natural language, adapting its behavior based on the task's specific requirements and constraints from a practical implementation point of view. This novel and efficient approach promises a more flexible and adaptable controller design process in power electronics that will largely facilitate the practitioners.

Large Language Models based Multi-Agent Framework for Objective Oriented Control Design in Power Electronics

TL;DR

This work introduces an LLM-based multi-agent framework for objective-oriented control design in power electronics, addressing challenges from parameter uncertainties and costly design cycles. By decomposing the process into specialized agents (objective design, model design, control algorithm design, parameter optimization, and verification) coordinated by a Manager, the approach translates high-level objectives into implementable controllers via Modelica models, PID/MPC options, PSO-based parameter tuning, and Gym-based verification. The framework's initial implementation demonstrates end-to-end automation and collaboration among agents, validated on a DC-DC boost converter where prompts yield controller designs, parameters, and performance indicators. Overall, the method promises greater flexibility, reduced design effort, and autonomous, constraint-compliant controller development for power electronics in modern grids.

Abstract

Power electronics, a critical component in modern power systems, face several challenges in control design, including model uncertainties, and lengthy and costly design cycles. This paper is aiming to propose a Large Language Models (LLMs) based multi-agent framework for objective-oriented control design in power electronics. The framework leverages the reasoning capabilities of LLMs and a multi-agent workflow to develop an efficient and autonomous controller design process. The LLM agent is able to understand and respond to high-level instructions in natural language, adapting its behavior based on the task's specific requirements and constraints from a practical implementation point of view. This novel and efficient approach promises a more flexible and adaptable controller design process in power electronics that will largely facilitate the practitioners.
Paper Structure (39 sections, 6 figures)

This paper contains 39 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Problem description and PRELIMINARIES
  3. Objective-Oriented Control Design in Power Electronics
  4. LLM-based Agent Frameworks
  5. Objective Oriented Controller Design
  6. Multi-Agent Framework Architecture
  7. Task Translation
  8. Manager
  9. Agent Collaboration
  10. Simulation and Verification
  11. Evaluation
  12. Agent Role Design
  13. Manager Agent
  14. Objective Design Agent
  15. Model Design Agent
  16. ...and 24 more sections

Figures (6)

  • Figure 1: Workflow of Objective Oriented Controller Design.
  • Figure 2: Flow chart of Objective Oriented Controller Design.
  • Figure 3: Initial implementation of the multi-agent framework
  • Figure 4: Designing objective oriented controller prompts.
  • Figure 5: Web interfaces of the multi-agent framework for task-oriented controller design
  • ...and 1 more figures