融合注意力机制的BiLSTM集群负载预测模型

Cluster load prediction model with BiLSTM incorporating attention mechanism

服务器集群的负载预测有助于集群资源的优化配置,以提升集群资源的利用率。针对传统负载预测算法存在预测精度低的问题,本文在双向长短期记忆网络模型的基础上,提出一种融合注意力机制的负载预测模型(Att-BiLSTM)。该模型充分考虑到服务器CPU、内存、磁盘和网络等因素,利用双向长短期记忆网络前后传递信息的特点,并使用注意力机制关注负载时间序列中的重要信息,从而提高了预测精度。实验结果表明,相对于目前已经提出的负载预测模型ARIMA、CNN-LSTM和BiLSTM、Att-BiLSTM模型具有更好的预测性能。

Load forecasting for server clusters contributes to optimizing cluster resource allocation and improving the utilization of cluster resources.Addressing the issue of low prediction accuracy in traditional load forecasting algorithms,this paper proposes a load prediction model,called Att-BiLSTM,which integrates an attention mechanism into the Bidirectional Long Short-Term Memory(BiLSTM)network model.This model takes into full consideration factors such as server CPU,memory,disk,and network.It leverages the bidirectional information flow characteristics of the BiLSTM network and employs an attention mechanism to focus on crucial information within the load time series,ultimately enhancing prediction accuracy.Experimental results indicate that,in comparison to existing load prediction models,including ARIMA,CNN-LSTM,and BiLSTM,the Att-BiLSTM model demonstrates superior predictive performance.