Aviation Gasoline Quality Detection Using RAP Feature Selection and IMPA-XGBoost Optimization

Abstract

Quality detection of aviation gasoline is critical to flight safety. Gasoline is a critical fuel for aircraft engines, and any quality problem may affect the performance of the engine and the safety of the aircraft. Regarding the issue of aviation gasoline detection, a RAP method combining Pearson correlation coefficient method and Relief-F algorithm is studied for gasoline mass spectrometry feature selection. Improved marine predator algorithm (MPA) introduces Logistic chaotic mapping and adaptive t-distribution operator. It is used to optimize the XGBoost model to construct an aviation gasoline quality detection model for gasoline quality detection and model classification. Among them, the RAP method is chosen because it effectively removes redundant features from mass spectrometry data while preserving the correlation between features. The use of IMPA to optimize XGBoost is because the traditional MPA is easy to fall into local optimum. Whereas the improved IMPA can find the optimal hyperparameter combination of XGBoost more effectively by enhancing the population diversity and optimizing the search strategy, thus improving the model detection performance. The results showed that the area under the receiver operating characteristic curve (AUC) of the proposed model was 0.8892, which was significantly higher than the AUC values of the particle swarm optimization-XGBoost (PSO-XGBoost) model (0.8384) and the sparrow search algorithm-XGBoost (SSA-XGBoost) model (0.8497). In the classification of gasoline models, only 4 samples were misclassified, while 122 samples were classified correctly, with an accuracy rate of 96.83%. This was a significant improvement compared to the 92.06% of the SSA XGBoost model and the 88.10% of the PSO XGBoost model. The improved marine predator and extreme gradient boosting model has shown excellent performance in gasoline quality detection. Compared to traditional chemical detection methods, such as plasma emission spectroscopy and gas chromatography with an oxygen-selective detector, the AI-based detection system proposed in this study has significant advantages in terms of detection accuracy and efficiency, and does not require expensive and complex detection equipment. This provides a strong support for AI automated system in quality detection of aviation gasoline.