A Novel Task-Driven Method with Evolvable Interactive Agents Using Event Trees for Enhanced Emergency Decision Support

TL;DR

The paper addresses agile emergency decision-making in unforeseen incidents by integrating Event Tree Analysis with a zero-shot, task-driven framework called EvoTaskTree. It uses two LLM-powered agent types—task executors and task validators—to perform three interdependent tasks: initiating event subevent analysis, event tree header analysis, and decision recommendations, guided by a continually evolving record library of successes and an experience base of failures. Demonstrated in a nuclear power plant simulacrum, EvoTaskTree achieves high to perfect accuracy on unseen scenarios and outperforms baseline prompting strategies, benefiting from dense retrieval and reasoned feedback. This approach offers a practical path to rapid, objective, and scalable emergency decision support in safety-critical infrastructures.

Abstract

As climate change and other global challenges increase the likelihood of unforeseen emergencies, the limitations of human-driven strategies in critical situations become more pronounced. Inadequate pre-established emergency plans can lead operators to become overwhelmed during complex systems malfunctions. This study addresses the urgent need for agile decision-making in response to various unforeseen incidents through a novel approach, EvoTaskTree (a task-driven method with evolvable interactive agents using event trees for emergency decision support). This advanced approach integrates two types of agents powered by large language models (LLMs): task executors, responsible for executing critical procedures, and task validators, ensuring the efficacy of those actions. By leveraging insights from event tree analysis, our framework encompasses three crucial tasks: initiating event subevent analysis, event tree header event analysis, and decision recommendations. The agents learn from both successful and unsuccessful responses from these tasks. Finally, we use nuclear power plants as a demonstration of a safety-critical system. Our findings indicate that the designed agents are not only effective but also outperform existing approaches, achieving an impressive accuracy rate of up to 100 % in processing previously unencoun32 tered incident scenarios. This paper demonstrates that EvoTaskTree significantly enhances the rapid formulation of emergency decision-making.

Figures (11)

• Figure 1: An overview of NPP simulacrum. The operators are autonomous agents powered by large language models. An interesting finding is that the agents can keep improving performance over time without manually labeled data.
• Figure 2: Simplified System Diagram for an NPP.
• Figure 3: Event tree for Large LOCA.
• Figure 4: The integration of Event Tree Theory with emergency decision support.
• Figure 5: The overview of the EvoTaskTree. (a) Architecture. It is a parameter-free strategy. For each task, the task executors and task validators interact and iterate continuously to complete tasks. The three tasks are interconnected through a task flow, collectively fulfilling the decision support. (b) Agent-Agent Interaction. This diagram illustrates the methods to achieve self-evolution: 1) accumulating examples; 2) adding correct responses directly to the record library; 3) adding incorrect responses directly to the experience base; 4) utilizing both knowledge to retrieve the most similar content for reasoning during the inference process. 5) During the training process, consistently repeat steps 1, 3, and 4 until the correct response is identified; 6) During the testing process, directly apply step 4 to generate the final answer.

Theorems & Definitions (4)

• Theorem 1: Event tree analysis
• Example 1
• Proposition 2
• proof