Prediction Method of Aero-Engine Performance Baseline Based on MHSA-CNN-BiLSTM

To accurately predict the aero-engine performance baseline, traditional baseline prediction methods rely solely on single data information, leading to issues such as insufficient data feature extraction and weak data feature discrimination ability in traditional models. In response, an aero-engine performance baseline prediction method based on MHSA and CNN-BiLSTM network is proposed. This method involves studying the stable cruise segment extraction method based on QAR performance data, combining correlation features and temporal features of performance data for fusion, and constructing a high-dimensional performance baseline dataset. A baseline prediction model for aero-engine performance is designed, combining the structural features of convolutional neural network (CNN) and bi-directional long short-term memory network (BiLSTM) to enhance the model's ability to extract abstract features and temporal features. Additionally, the multi-head self-attention mechanism (MHSA) is introduced to enhance the discrimination of key features in high-dimensional performance parameters and improve the accuracy of performance baseline prediction. The results are validated using actual aero-engine performance data provided by an aerospace company, demonstrating that the method can predict the baseline exhaust gas temperature deviation (ΔEGT) with a coefficient of determination up to 98.015%.