基于SVM的集成预测方法及其在水电机组故障诊断中的应用

Support Vector Machine-based Integrated Prediction Method and Its Application to Fault Diagnosis for Hydrogenerator Units

针对水力发电系统的非线性及非平稳性特点,提出一种基于小波变换和支持向量机(SVM)的集成预测方法,用于水电机组状态趋势分析。采用小波变换将非平稳时间序列分解成若干个具有较强规律性的子序列,然后采用合适核函数的最小二乘支持向量机(LS-SVM)方法分别对这些子序列进行预测,最后综合这些子序列的预测结果作为原始序列的预测值。将该集成方法应用于某水电机组振动峰峰值的预测,结果表明该集成方法的预测性能优于单一LS-SVM方法。

Considering the nonlinearity and unsteadiness of hydropower systems, an integrated prediction model based on wavelet transform and support vector machine （SVM） is proposed for the condition trend analysis of hydrogenerator units. Firstly, the unsteady time series are decomposed through wavelet transform into several sub-series with obvious tendency characteristics. Then, the trends of the sub-series are predicted respectively with least squares support vector machine （LS- SVM）, in which the kernel functions are chosen appropriately. Finally, the prediction results are integrated as the prediction values of original time series. The integrated prediction model is applied to the peak-peak value of vibration time series of some hydrogenerator units. The results show that its prediction performance is better than that of single LS-SVM method.