PATHOPHYSIOLOGY CORRECTION WİTH A PI CONTROLLER AND A Q-LEARNING-BASED REINFORCEMENT LEARNING ALGORITHM FOR IMPROVED LUNG VENTILATION

BİLSEL INTERNATIONAL TURABDİN SCIENTIFIC RESEARCHES AND INNOVATION CONGRESS, 23 - 25 Ocak 2024, ss.679

Yayın Türü: Bildiri / Özet Bildiri
Sayfa Sayıları: ss.679
Yıldız Teknik Üniversitesi Adresli: Evet

Özet

This study describes a methodology for correcting specific physiological irregularities within a lung ventilation model, with a focus on addressing tachycardia and increased respiratory resistance in conditions similar to emphysema. Khoo's linear respiratory mechanics model was used to create a complete representation of both central and peripheral airways. As an input, the framework uses airway opening pressure, and as an output, it quantifies the volumetric flow rate of air into the alveoli. Pathophysiological states with increased volumetric flow rates and respiratory resistance were used to simulate the abnormal conditions. A multi-step approach was used to address this imbalance. To begin, a low-pass filter was used to reduce the high-frequency respiratory patterns associated with arrhythmia. Following that, a Proportional-Integral (PI) controller was added to the system to regulate the volumetric flow rate and return it to a more natural state. The PI controller's parameters were determined using a Q-learning-based reinforcement-learning algorithm based on the Bellman equation. This iterative process resulted in optimized controller parameters that were more effective in correcting the simulated pathophysiological conditions. Analyses of controller input-output responses and Integral Absolute Errors for each approach demonstrated the efficacy of this methodology. The study presents a satisfactory lung ventilation model, demonstrating its ability to effectively correct selected physiological disorders within the simulated environment.