摘要
硬盘故障给数据中心的可靠性和可用性带来的影响越来越大,采用不同的机器学习方法构建基于自监控分析报告技术(self-monitoring,analysis and reporting technology,SMART)属性的硬盘故障预测模型策略的研究已经取得了一定的效果.但这些模型策略无法得到较为稳定的预测效果,并且无法选择适合于不同用户需求的最佳模型.为得到更高的准确率和较低的误报率,实现了基于Adaboost算法的BP神经网络预测模型优化方法.在此基础上,为更好地适用于实际工作场景,实现了根据遗传算法(genetic algorithm,GA),按照用户的预测效果要求,选择出最恰当的预测模型的方法,在不同的效果要求下选用不同的预测模型.
Because the impact of the drive failures is more important,hard drive failure prediction models based on the self-monitoring,analysis and reporting technology(SMART)attributes have been built using different machine learning methods,and have achieved good prediction performance.However,the prediction results of these models are not stable.And the models have no way to choose the optimal models according to the different prediction condition and customers'requirement.For the sake of higher prediction rate and lower false alarm rate,this paper proposes an optimization model using the Adaboost and backpropagation(BP)neural network algorithm based on the SMART.To apply better to the actual work,we use the genetic algorithm(GA)to choose the best combination of the different classifiers to build the prediction models,which approaches the prediction requirements of the customers.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2014年第S1期148-154,共7页
Journal of Computer Research and Development
基金
国家自然科学基金项目(61373018
11301288)
教育部新世纪优秀人才支持计划基金项目(NCET-13-0301)
中央高校基本科研业务费专项基金项目(65141021)
