Abstract:
With the continuous development of satellite industry, online monitoring and fault diagnosis for satellite communication system becomes more important. Due to the difficu...Show MoreMetadata
Abstract:
With the continuous development of satellite industry, online monitoring and fault diagnosis for satellite communication system becomes more important. Due to the difficulty in obtaining sufficient features of communication system, the conventional multi-classification algorithm Directed Acyclic Graph Support Vector Machine (DAG-SVM) has low diagnostic efficiency and poor coupling diagnosis performance. On the other hand, it has been proved that extending the feature space can effectively improve the classification performance. Therefore, this paper proposed a modified multi-classification algorithm called Feature-Extended Directed Acyclic Graph Least Square Twin Support Vector Machine (FEDAG-LSTSVM). The new algorithm combined all the initial features and their random combinations to establish coupling and redundancy for every feature, and then constructed the Separable Metric (SM) as classification measurement to arrange the structure sequencing of DAG-LSTSVM. To verify the utility of the algorithm, the standard monitoring signal indicators in satellite communication system were taken as experimental data. Preliminary simulation results demonstrate that the proposed algorithm improves the fault diagnosis accuracy to 89.69% but with 54.20% less computational time in 10-fold cross-validation compared with DAG-SVM, which means it can be well applied to diagnose fault for satellites communication system.
Date of Conference: 09-10 May 2019
Date Added to IEEE Xplore: 25 July 2019
ISBN Information: