Research on adaptive Fault Prediction algorithm for super-computer

Research on adaptive Fault Prediction algorithm for super-computer

Chongqing University Lin

In this paper, using the classification prediction idea in data mining, the time axis is divided into a certain size time window, the feature extraction is carried out for the time window, and the fault prediction is based on the time window. In this paper, the AdaBoost algorithm is used in the training and learning process of SVM classifier, the classifier's core parameters are dynamically adjusted according to the training set, and the classifier is improved by adaptive learning, and an adaptive fault Prediction model is established ADABOOSTSVM. This paper takes the system running log of super Computer bluegene/l215 Day as the experimental data set, and after preprocessing, carries on the comparison experiment of the prediction model on the dataset. The experimental results show that the ADABOOSTSVM model of this paper has better classification and prediction performance than the fault prediction model based on the interval (time inclusive failure TBF), KNN, Ripper and SVM. Especially in the recall rate of the important index in the fault prediction, the recall rate of the adaptive fault Prediction model is higher than that of other forecast models 10%-20%. ADABOOSTSVM

Research on adaptive Fault Prediction algorithm for super-computer