改进RVM的装备退化状态预测方法

Prediction Method of Equipment Degradation State Based on Improved RVM

为提高相关向量机(relevance vector machine,RVM)模型的预测精度,提出了一种改进RVM的装备退化状态预测方法。首先,通过构建一种方差高斯核函数(variance Gauss kernel function,VGKF)来提高核函数的全局性能和泛化能力;然后通过借鉴混沌序列局域预测法中邻近点个数的选取方法,利用H-Q准则(Hannan-Quinn rule)对训练空间预测嵌入维数进行优化,避免了主观选取的盲目性;最后提出一种线性搜索算法,优化确定核参数值。通过末制导雷达装备测试参数的预测实例,验证了改进RVM的有效性和优越性。

To improve the prediction accuracy of the relevance vector machine model,an improved method for equipment condition prediction is proposed. An improved kernel function of variance Gauss kernel is constructed to improve the global performance and generalization ability of the kernel function.By using the method of selecting the number of adjacent points in the chaotic sequence local prediction method,the H-Q criterion is used to optimize the embedding dimension of the training space to avoid the blindness of subjective selection; a linear search algorithm is proposed to optimize the kernel parameter values. Through the prediction example of terminal guidance radar equipment test parameters,the effectiveness and superiority of the improved RVM are verified.