多核相关向量机在电站运行数据监测系统中的应用

Multi-kernel Relevance Vector Machine in Operating Data Monitoring System of Power Plant

针对相关向量机(RVM)的不足,提出一种改进的多核相关向量机方法,结合高斯核函数和多项式核函数形成混合核函数,用差分进化粒子群算法优化参数,与传统的SVM和RVM相比,该方法具有更高的数据预测精度。将该方法应用于电站运行数据监测系统,对运行数据建立回归模型,采用测量检验法检验数据是否超限、存在异常,并对异常数据进行重构。通过某600MW机组数据的实例分析,验证了方法的有效性。

To resolve the disadvantages of relevance vector machine （RVM）,an improved multi-kernel relevance vector machine is proposed.The mixture of kernels is formed of Gaussian kernel and polynomial kernel. Parameters are optimized by using differential evolution particle swarm algorithm.Comparing with traditional SVM and RVM,this method has a higher forecasting precision.This method used in operating data monitoring system of power plant is applied to structure regression model of operating data.Measurement Test is used to check whether the data has entered an overrun condition.