Predictive anomaly detection for marine diesel engine based on echo state network and autoencoder

Abstract

Marine diesel engine with high thermal efficiency and good economy has become the main power of ships. Anomaly detection is an important method to improve the operation reliability of marine diesel engine. Most anomaly detection research focuses on failures that have occurred, and few studies consider anomaly prediction. A predictive anomaly detection method based on echo state network (ESN) and deep autoencoder is proposed. Historical sample data is collected and used to train the prediction network ESN and the anomaly detection network deep auto-encoder. After training, the prediction network ESN is used to predict the sensor data sequence in the future. And the predicted sequence is input into the anomaly detection network deep auto-encoder to obtain the predictive anomaly detection result. The relative error and root mean square error of the proposed method are at least 0.089 and 1.002 lower than other methods, respectively. Compared with other anomaly detection methods, the proposed autoencoder method obtains the best precision, accurate, recall indicators. Experiments show that it is feasible to establish a predictive anomaly detection method. More experiments under different conditions need to be studied, and higher performance algorithms need to be developed in the future.

Keywords

• Anomaly detection
• Prediction
• Marine diesel engine
• Echo state network
• Deep autoencoder