Intelligent control of nuclear facility ventilation system using hybrid FNN-PSO model

• A novel FNN-PSO model is proposed for nuclear facility ventilation control. • FNN-PSO achieves superior training efficiency while maintaining high accuracy. • CFD identifies flow instability as the main challenge for fitting at small openings. • Experimental validation confirms the method's high accuracy and applicability. The ventilation systems of nuclear facilities (VSNF), characterized by complex pipeline networks, pose significant challenges for precise control. Enhancing their control efficiency is crucial for ensuring operational safety and reliability. This study proposes a smart control strategy for VSNF using a hybrid deep learning framework that combines a feed-forward neural network with particle swarm optimization (FNN-PSO). The computational performance of three typical deep learning frameworks and three PSO-optimized hybrid models was systematically evaluated and compared, with a thorough analysis of their error distributions. The accuracy and generalizability of the proposed strategy were experimentally validated using a multi-branch pipe network test bench. Results demonstrate that the FNN-PSO framework achieves superior training efficiency (approximately 40% faster than LSTM-PSO) while maintaining high computational accuracy, with errors below 6% within the 0.4–1.0 valve opening range and below 15% in the challenging 0.2–0.4 range. This combination of rapid training and high accuracy makes FNN-PSO particularly suitable for VSNF applications, where control models require periodic updates to adapt to system changes such as filter performance degradation. This work provides an efficient and reliable solution for the intelligent control of VSNF, advancing the practical application of artificial intelligence in nuclear safety.