A Hybrid CNN-RNN-MRAC Deep Learning Framework for Adaptive Control of Intelligent Oxygen Generation Systems

Abstract

This paper presents a hybrid intelligent control framework for oxygen generation systems that integrates a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), and Model Reference Adaptive Control (MRAC). The CNN extracts spatial correlations from multi-sensor inputs, the RNN, implemented with LSTM or GRU units, captures temporal dynamics in flow rate, pressure, and environmental variables, and the MRAC module enables real-time parameter adaptation under time-varying operating conditions. The proposed architecture is implemented and validated on a commercial molecular-sieve oxygen concentrator based on Pressure Swing Adsorption (PSA) technology. Experimental results show clear improvements over conventional PID controllers and single-branch deep-learning approaches: oxygen flow control accuracy within ±1.3%, oxygen concentration control accuracy within ±1.5%, and an average pressure fluctuation rate of 9.75%. These results indicate that the system can provide efficient and adaptive oxygen therapy, which is important for medical, aerospace, and home-care applications.