Safety Control Optimization Model Based on Neural Architecture Search
Jiapeng Chen, Xiaoming Li, Gengdong Huang, Xiaojie Lin, Peixin Wu · 2025
With the increasing diversity and complexity of security incidents, traditional safety control methods are limited in their adaptability and responsiveness in dynamic and complex environments. To address this, this paper proposes a safety control optimization model based on Neural Architecture Search (NAS), which enhances the model's flexibility and emergency response capabilities through automated search for the optimal network architecture. The model can automatically adjust control strategies based on different environmental and task requirements, enabling rapid response to sudden security incidents. In this work, a customized NAS search space is constructed, incorporating traditional neural network layers, time-series analysis modules, and event prediction modules to meet the multi-dimensional requirements of safety control tasks. By integrating reinforcement learning and Bayesian optimization, the selection of network architectures and the tuning of hyperparameters are optimized, improving the model's search efficiency and accuracy. Experimental results show that the NAS-optimized safety control model demonstrates higher adaptability and decision-making accuracy in multiple test scenarios compared to traditional methods, effectively addressing complex and dynamic safety control tasks.