Research On Data-driven Residual Life Prediction Method Of Inertial Navigation System

Inertial navigation system is a key component of the aircraft,and its degradation assessment and residual life prediction are of great significance to ensure the flight safety of the aircraft.The inertial navigation data collected by the sensor periodically is the basic data for the health prediction of the inertial navigation system,and it is also a special multidimensional time series data.Based on machine learning theory,the evolution law and internal correlation relationship of the health status of the inertial navigation system are excavated from the large-scale inertial navigation data,and an effective health management model of the inertial navigation system is established,which provides auxiliary decision-making for the state evaluation and maintenance of the inertial navigation system,which has important practical significance for ensuring flight safety,and also has important research value in the field of data mining technology.This paper focuses on the data-driven inertial navigation system residual life prediction method,first of all,the anomaly data detection method in the inertial navigation system is studied,and then on the basis of anomalous data cleaning,the residual life prediction method of the inertial navigation system based on the improved LSTM-DBN is studied,and finally the anomaly data detection method and the residual life prediction method are integrated into it,and the inertial navigation health management system with residual life prediction and health monitoring functions is constructed.The main tasks of this article are as follows:Firstly,aiming at the problem that anomalous data affects the prediction result,this paper proposes an anomalous data detection method based on discriminant autoencoder.In order to improve the reconstruction efficiency of generative adversarial networks,this paper proposes a new divergence function to avoid the problem of gradients approaching zero in conduction and the constraints of Lipschitz conditions.In order to make the threshold setting more accurate,this paper designs a point-to-point autoencoder threshold setting method.On the inertial navigation data set of the 2016-2017 Antarctic Circumnavigation,the inertial anomaly data detection method proposed in this paper was tested,and the method proposed in this paper achieved the best results compared with the other three anomaly detection methods.Secondly,an improved LSTM-DBN method is proposed for the remaining life prediction of inertial navigation system.This method can train different parameters according to the different states of the system,and use mutual information entropy to measure correlation,which effectively solves the problem that forgotten information needs to be quantitatively interpreted.The method also uses the particle swarm algorithm to optimize the system parameters,which further improves the accuracy of prediction.On the inertial navigation dataset of the 2016-2017 Antarctic Circumnavigation,the method was tested.In 1000 experiments,the residual life prediction method of inertial navigation system based on the improved LSTM-DBN has about 1.3 misjudgments,while the LSTM-DBN-based method has about 3.7 misjudgments.The effectiveness of the proposed method is proved by experiments.Finally,an inertial navigation health management system is constructed.The anomaly data detection algorithm based on discriminant autoencoder proposed in this paper is applied to the system to detect anomalies in the data and realize the cleaning of the data,so as to improve the accuracy of subsequent data processing.The remaining life prediction method of LSTM-DBN-based inertial navigation system proposed in this paper is applied to the system to achieve the prediction of inertial navigation life.The system has data selection,data statistics,life management and life prediction functions to achieve the management and monitoring of the health status of the inertial navigation system.