Sensor deviation anomaly trend prediction based on phase space reconstruction and Support Vector Regression

Abstract

Ship combat systems are developing in the direction of integration and intelligence, and the integration also brings problems such as difficulty in predicting abnormal sensor trends in combat systems. The failure characteristics of sensors in combat systems are highly secretive, and it is difficult to detect and deal with system equipment failures by relying on empirical qualitative or single-indicator testing methods. In order to safeguard the ship’s combat capability and make reasonable maintenance decisions, it is important to study more accurate prediction methods. This article proposes a method for predicting sensor deviation anomaly trends based on phase space reconstruction and Support Vector Regression(SVR), taking sensors in naval combat systems as typical research objects. Firstly, a phase space reconstruction is performed on the sensor deviation time series data to increase the accuracy of subsequent prediction; then a SVR prediction model is built, and the model is trained and optimised using the reconstructed data; finally, the constructed prediction model is compared with the regular SVR model and BP neural network model to verify that the constructed model has higher prediction accuracy.