【摘要】 将最小二乘支持向量机方法引入火电厂DCS的测量数据时序异常检测领域,该方法很好地建立了火电厂DCS的测量数据时序预测模型,具有预测真实值能力强、全局优化及泛化性好等优点。将该方法应用于某600 MW超临界火电机组DCS测量数据中,经过训练后的LS-SVM模型对再热蒸汽温度数据的检验样本进行不良值检测与真实值预测,均方根误差和平均相对误差分别为0.067%和0.050%,均方根误差是BP网络模型、RBF网络模型的8.756%和8.272%,平均相对误差是BP网络模型、RBF网络模型的7.541%和7.236%。应用结果表明,最小二乘支持向量机方法优于多层BP与RBF神经网络法,能很好地满足异常检测与真实值预测要求。更多还原

【Abstract】 A new algorithm for outliers detecting in the bad measured data in the distributed control system(DCS) in power plant based on the least squares support vector machine(LS-SVM) method is presented in this paper.The method establishes a model to reflect the true value of the measured data in DCS in power plant.It has the advantages of high forecasting accuracy,global optimal property,and more generalized performance.Applied to a 600 MW super critical coal-burning utility boiler,the LS-SVM model which had been trained detects the bad measured data in the test samples in the reheat steam temperature data set,forecasted the true value,and got the mean square root error and the mean relative error of 0.067% and 0.050%,the mean square root error are 8.756% and 8.272% of back propagation (BP) method and radial basis function(RBF) neural network,and the mean relative error are 7.541% and 7.236% of BP and RBF method.These results show that LSSVM method is more accurate than BP and RBF neural network,and can satisfy the outliers detecting and the true value forecasting demand well.更多还原