基于循环神经网络的系统负载预测
首发时间:2018-01-22
摘要:在应用性能管理系统中,系统未来的负载情况对运维调度有重要的指导意义。在云计算环境下,弹性伸缩计算能力为调整系统规模提供了可能,根据系统将来的负载情况可以提前做出相应的调整:可以在负载加重前扩展好集群,保证服务质量;在负载降低之后若预测一定时间内没有负载加重的情况,则可以及时缩减集群规模,降低企业运营成本。在金融领域,ARIMA模型是常用的时序预测模型,但其应用需要人工介入分析时序的平稳性,调参过程过于复杂。近年来神经网络技术的发展带动了人工更智能技术的发展,本论文设计并测试了ANN、RNN、GRU、LSTM等神经网络的负载预测的效果。实验结果表明LSTM网络预测精准且表现稳定,是系统负载预测的理想模型。
关键词: 计算机应用技术 多元时间序列 时间序列预测 循环神经网络
For information in English, please click here
System load forecasting based on recurrent neural network
Abstract:In the application performance management system, the future load of the system has an important guiding significance for operation and maintenance scheduling.In a cloud computing environment, the flexible scalability computing power provides a way to change the scale of computing clusters. It is of great importance toadjust the scale of the cluster according to the system load in the future.Expand the cluster before the load is aggravated to ensure the quality of service. Reducing compute clusters When forecasting system load is low for some time to come. Predicting future load can improve service quality and reduce business operating costs. In financial field, ARIMA model is a commonly used time series forecasting model, but its application requires the stability of human interventional analysis of timing, the process of adjusting parameters is too complicated.In recent years,the development of neural network technology has led to the boom of artificial intelligence technology. This paper designs and tests the load forecasting effects of ANN, RNN, GRU, LSTM and other neural networks.The experimental results show that LSTM network is accurate and stable, which is an ideal model for system load forecasting.
Keywords: Computer Application Technology Multivariate time series Time Series Prediction Recurrent neural network
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于循环神经网络的系统负载预测
评论
全部评论0/1000