摘要
针对目前负载预测模型预测精度低、预测模型种类较单一的问题,建立一种基于粒子群的门控循环单元(GRU)与循环神经网络(RNN)结合的组合预测模型,该模型利用粒子群优化模型的参数,通过自适应调参数的方式为GRU-RNN组合模型匹配最优的神经元个数,利用GRU预测精度高以及RNN参数少、易收敛的优点对云计算资源负载进行预测。首先对数据集作缺失值处理、使用CRITIC权重法将CPU使用率和内存使用率的负载值进行组合,再对组合值进行标准化处理,作为PSO-GRU-RNN组合预测模型的输入,实现对云计算资源的高效预测。利用集群公开数据集Cluster-trace-v2018、PSO-GRU-RNN组合预测模型对负载进行预测。实验结果表明,笔者提出的PSO-GRU-RNN模型相比于现有的组合预测模型ARIMA-LSTM、GRULSTM等,有更高的预测精度。
Aiming at the problems of low prediction accuracy and single type of load forecasting models,we pro⁃posed a combined forecasting model based on particle swarm optimization(PSO)of Gate Recurrent Unit(GRU)and Recurrent Neural Network(RNN)in this paper.The model used the parameters of the PSO model to match the optimal number of neurons for the GRU-RNN combined model through adaptive adjustment parameter.The forecasting model took the advantages of high prediction accuracy of GRU,fewer parameters and better conver⁃gence of RNN to predict the cloud computing resource load.First,the data set was processed with missing val⁃ues,and the load values of CPU utilization and memory utilization were combined using CRITIC weight method,and then the combined values were standardized as the input of the PSO-GRU-RNN combined prediction model to realize the efficient prediction of cloud computing resources.The Cluster-trace-v2018 public dataset was ap⁃plied during the forecasting process for the PSO-GRU-RNN model.The experimental results demonstrate that the proposed PSO-GRU-RNN model has higher prediction accuracy than the existing combined prediction mod⁃els such as ARIMA-LSTM and GRU-LSTM.
作者
胡应钢
郭翔
赵海燕
姜静清
HU Ying-gang;GUO Xiang;ZHAO Hai-yan;JIANG Jing-qing(College of Mathematics Science,Inner Mongolia Minzu University,Tongliao 028043,China;College of Computer Science and Technology,Inner Mongolia Minzu University,Tongliao 028043,China)
出处
《内蒙古民族大学学报（自然科学版）》
2023年第4期315-321,共7页
Journal of Inner Mongolia Minzu University：Natural Sciences
基金
国家自然科学基金项目(62162050,61662057)。
关键词
负载预测
预测模型
门控循环单元
循环神经网络
粒子群优化
Load forecasting
Forecasting model
Gated recurrent unit
Recurrent neural network
Particle swarm optimization
