Identification and diagnosis of flow instability in parallel channel systems using machine learning

Density wave oscillations in parallel thermal-fluid systems threaten operational safety due to uneven flow and instability. Predicting these oscillations is a critical challenge for both forced and natural circulation systems. Here, we demonstrate an intelligent framework that accurately identifies such hazardous states. We combine numerical modeling of open and closed circulation systems with a hybrid method integrating Genetic Algorithm (GA) optimization and K-Nearest Neighbors (KNN) classification. For open systems, we map oscillation patterns across a wide operational space. For closed systems, we efficiently pinpoint DWO-sensitive regions. Our results show that the flow state recognition method, validated against manual and signal analysis, achieves an overall accuracy of 99.19%. Furthermore, GA-optimization of the KNN model achieved 96.4-100% accuracy with superior blind-test generalization, while spatial non-uniformity of heat load distribution critically governs instability patterns, informing differentiated safety strategies. This confirms the framework's effectiveness and reliability for practical engineering applications.