WorkloadGPT: A Large Language Model Approach to Real-Time Detection of Pilot Workload

Applied Sciences, Год журнала: 2024, Номер 14(18), С. 8274 - 8274

Опубликована: Сен. 13, 2024

The occurrence of flight risks and accidents is closely related to pilot workload. Effective detection workload has been a key research area in the aviation industry. However, traditional methods for detecting have several shortcomings: firstly, collection metrics via contact-based devices can interfere with pilots; secondly, real-time challenging, making it difficult capture sudden increases workload; thirdly, accuracy these models limited; fourthly, lack cross-pilot generalization. To address challenges, this study proposes large language model, WorkloadGPT, which utilizes low-interference indicators: eye movement seat pressure. Specifically, features are extracted 10 s time windows input into WorkloadGPT classification low, medium, high categories. Additionally, article presents design an appropriate text template serialize tabular feature dataset natural language, incorporating individual difference prompts during instance construction enhance Finally, LoRA algorithm was used fine-tune pre-trained model ChatGLM3-6B, resulting WorkloadGPT. During training process GAN-Ensemble employed augment experimental raw data, constructing realistic robust extended training. results show that achieved 87.3%, standard deviation only 2.1% response just 1.76 s, overall outperforming existing studies terms accuracy, performance, generalization capability, thereby providing solid foundation enhancing safety.

Язык: Английский

Evaluating and Enhancing Trustworthiness of LLMs in Perception Tasks

2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC), Год журнала: 2024, Номер unknown, С. 431 - 438

Опубликована: Сен. 24, 2024

Язык: Английский