Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Behavioral Generative Agents for Power Dispatch and Auction

Shaoze Li, Justin S. Kim, Cong Chen

TL;DR

Positive initial evidence that generative agents can relax the rigidity of traditional mathematical models for human decision-making in power dispatch and auction settings is presented and LLM-powered agents provide a flexible and expressive testbed for modeling human decision-making in power system applications.

Abstract

This paper presents positive initial evidence that generative agents can relax the rigidity of traditional mathematical models for human decision-making in power dispatch and auction settings. We design two proof-of-concept energy experiments with generative agents powered by a large language model (LLM). First, we construct a home battery management testbed with stochastic electricity prices and blackout interventions, and benchmark LLM decisions against dynamic programming. By incorporating an in-context learning (ICL) module, we show that behavioral patterns discovered by a stronger reasoning model can be transferred to a smaller LLM via example-based prompting, leading agents to prioritize post-blackout energy reserves over short-term profit. Second, we study LLM agents in simultaneous ascending auctions (SAA) for power network access, comparing their behavior with an optimization benchmark, the straightforward bidding strategy. By designing ICL prompts with rule-based, myopic, and strategic objectives, we find that structured prompting combined with ICL enables LLM agents to both reproduce economically rational strategies and exhibit systematic behavioral deviations. Overall, these results suggest that LLM-powered agents provide a flexible and expressive testbed for modeling human decision-making in power system applications.

Behavioral Generative Agents for Power Dispatch and Auction

TL;DR

Positive initial evidence that generative agents can relax the rigidity of traditional mathematical models for human decision-making in power dispatch and auction settings is presented and LLM-powered agents provide a flexible and expressive testbed for modeling human decision-making in power system applications.

Abstract

This paper presents positive initial evidence that generative agents can relax the rigidity of traditional mathematical models for human decision-making in power dispatch and auction settings. We design two proof-of-concept energy experiments with generative agents powered by a large language model (LLM). First, we construct a home battery management testbed with stochastic electricity prices and blackout interventions, and benchmark LLM decisions against dynamic programming. By incorporating an in-context learning (ICL) module, we show that behavioral patterns discovered by a stronger reasoning model can be transferred to a smaller LLM via example-based prompting, leading agents to prioritize post-blackout energy reserves over short-term profit. Second, we study LLM agents in simultaneous ascending auctions (SAA) for power network access, comparing their behavior with an optimization benchmark, the straightforward bidding strategy. By designing ICL prompts with rule-based, myopic, and strategic objectives, we find that structured prompting combined with ICL enables LLM agents to both reproduce economically rational strategies and exhibit systematic behavioral deviations. Overall, these results suggest that LLM-powered agents provide a flexible and expressive testbed for modeling human decision-making in power system applications.
Paper Structure (21 sections, 1 theorem, 5 equations, 5 figures, 1 table)

This paper contains 21 sections, 1 theorem, 5 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Generative agents for auctions and strategic operations
  3. Generative agents for power dispatch and auctions
  4. Dispatch Behavior for Battery Operations
  5. Agent decision dynamics with blackout intervention
  6. In-context learning for behavior discovery
  7. Dynamic program policies without blackout intervention
  8. Experimental Settings
  9. Experimental Results
  10. Bidding Behavior for Power Auctions
  11. Power Network Access Right Auction Mechanism
  12. Auction Rules and Price Updates
  13. Straightforward Bidding Strategy
  14. Generative Agent Testbed
  15. In-Context Learning (ICL)
  16. ...and 6 more sections

Key Result

Proposition 1

Problem (p1) is equivalent to the following problem:

Figures (5)

  • Figure 1: LLM prompt design using a Thought-Action-Reflection-Journal (TARJ), augmented by in-context learning (ICL) examples.
  • Figure 2: Comparing SoC resulted from LLM decision during blackout. Left: without IC examples. Right: with ICL-blackout examples.
  • Figure 3: Comparing bidding process/behavior of generative agents and Straightforward Bidding Strategy
  • Figure 4: Average accumulated reward with blackout intervention. Left: without IC examples. Right: with ICL-blackout examples.
  • Figure 5: o1-preview results for the average SoC and accumulated reward over the 20-day experiment. Chen25behavioral

Theorems & Definitions (3)

  • Definition 1
  • Proposition 1
  • Proof 1