基于组合核函数的相关向量机的变压器故障预测研究

The Transformer Fault Prediction Research of the Relevance Vector Machine Based on Combined Kernel Function

阐述了相关向量机是一种在贝叶斯框架下进行训练,根据主动相关决策机理来移除不相关点,从而获得稀疏化模型的机器学习方法。相关向量机与传统的支持向量机相比具有核函数不需要满足马瑟条件,学习算法简单,运算速度快,预测精度高等优点。提出了一种基于相关向量机回归预测算法与粒子群算法相结合的电力变压器故障预测新方法,该方法利用具有组合核函数的相关向量机算法对变压器油中溶解气体数据进行训练,由此得到相关向量和权值之后再进行变压器故障的回归预测。经实例验证取得了良好的预测结果。

A relevance vector machine （RVM） is a machine learning technique to obtain the sparse model by removing the irrelevant points using active relevance decision mechanism under the trainings of Bayesian framework. Compared with support vector machine （SVM）, RVM not only doesn＇t need to meet the Mecer conditions, but also is easier to learn, faster to calculate and more precise in prediction. This article puts forward a new power transformer fault prediction method by combining RVM regression forecast method and particle swarm optimization. This method trains the oil-dissolved gas data in the transformer using the RVM method based on combined kernel function, and then makes regression forecast of the transformer fault from the obtained relevance vector and weight. This method has produced good prediction results through practice.