CHP Engine Anomaly Detection Based on Parallel CNN-LSTM with Residual Blocks and Attention

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Combined Heat and Power (CHP) engines operate under extreme conditions, increasing the risk of anomalies and potential failure.
• Early detection of engine anomalies is critical for ensuring operational reliability and preventing costly breakdowns.

Purpose of the Study:

• To introduce a novel anomaly detection model for CHP engines using sensor data.
• To evaluate the effectiveness of a parallel Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture with residual blocks and attention mechanisms.

Main Methods:

• Development of a parallel CNN-LSTM residual blocks attention (PCLRA) model for anomaly detection.
• Extraction of spatiotemporal features using parallel CNN-LSTM networks.
• Compensation for information loss via residual blocks and an attention mechanism.

Main Results:

• The PCLRA model demonstrated superior performance compared to various hybrid models across 15 test cases.
• Achieved a macro f1 score of 0.951 ± 0.033, an anomaly f1 score of 0.903 ± 0.064, and an accuracy of 0.999 ± 0.002.
• Evaluated the contributions of residual blocks and attention mechanisms to the model's enhanced performance.

Conclusions:

• The PCLRA model represents a significant advancement in CHP engine anomaly detection.
• The proposed model is expected to enhance the energy efficiency and safety of CHP engines.
• This study is the first to apply parallel CNN-LSTM networks for anomaly detection in CHP engines.