相关向量机及其在故障诊断与预测中的应用

Relevance Vector Machine and Its Applications in Fault Diagnosis and Prognosis

摘要:

相关向量机(RVM)是一种基于稀疏 Bayesian学习理论的新型机器学习方法,具有概率式输出、稀疏性强、参数设置简单、核函数选择灵活等优点,克服了人工神经网络(ANN)和支持向量机(SVM)等典型机器学习方法的诸多固有缺陷。文章从模型选择与优化、模型计算效率和模型鲁棒性改进 3个方面综述了 RVM的理论研究进展；总结了 RVM在故障诊断与预测中的应用研究现状；分析指出了当前研究中存在的问题,并讨论了基于 RVM的故 障诊断与预测技术的研究方向。

Abstract:

Relevance vector machine(RVM) is a new machine learning method based on sparse Bayesian learn. ing theory, which has probabilistic outputs, high sparsity, simple parameter tuning and flexible selection of ker. nel function. RVM has overcome many inherent defects of typical machine learning methods, such as ANN and SVM. The research progress of relevance vector machine(RVM) was summarized in model selection and optimi. zation, model computational efficiency and model robustness improvement. The research status of applications of RVM in fault diagnosis and prognosis was introduced. The existing problems in the current research were ana. lyzed and the development trends of fault diagnosis and prognosis based on RVM were discussed.