服务器性能预测的递归神经元网络方法

正确有效地预测服务器性能负载，是计算机系统性能管理系统的一个重要环节。通常，传统的预测方法有最小二乘、二次指数平滑法等，但这些模型往往不能很好地捕捉服务器性能负载数据的时序关系。利用基于局部回归的递归神经网络（RNN），采用改进的RPROP学习算法进行服务器性能负载的预测。并与传统的二次指数平滑法相比较，实验结果证明，RNN得到的预测结果要比二次指数平滑法高出5个百分点以上，并且有较强的预测能力，可以预测较长周期的数据。

Server performance prediction using recurrent neural network

It's vital to predict effectively the performance ofservers in performancemanagement of computersystems.Usually,lestmean square error algorithm,Quadric exponential smoothing and so on are used in such predictions.But these models are insensitive to the temporal relationship of the data set of server performance.A prediction model was provided using partial recurrent neural network.And an improved learning algorithm,RPROP(resilient propagation) was applied to the prediction model.Experiments show that recurrent neural networkmodel is more accurate and effectivethan quadric exponential smoothing models.Accuracy is improved byup to 5 percent at least,and the prediction capability is much better than traditional ones.