Scholarlink

<p>Cloud computing has transformed the IT landscape, however its success relies on accurate workload forecasting and efficient resource management. By leveraging historical data to predict future workloads, cloud providers can optimize resource allocation, mitigate SLA violations, and deliver more efficient cloud services. This paper presents CBA-HDL (Convolutional Bidirectional Attention Hybrid Deep Learning), a Hybrid Deep Learning Model that combines 1D Convolutional Neural Networks, Bidirectional Gated Recurrent Units, and an attention mechanism to effectively capture local patterns, long-term dependencies, and key input features, thereby improving workload forecasting accuracy in cloud data centers. This approach facilitates multistep-ahead predictions for CPU and memory utilization, thereby improving accuracy and efficiency of cloud operations. Additionally, the paper discusses data preprocessing techniques such as Savitzky-Golay filtering and min-max normalization, the application of Apache Spark, and Bayesian optimization for hyperparameter tuning. Finally, the CBA-HDL prediction model demonstrated improved performance, with a 40.61% reduction in training time and evaluation metrics of MAE 6.15%, MSE 5.93%, and MAPE 22.14%, outperforming comparable models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CBA-HDL: A hybrid deep learning model for scalable and efficient cloud workload forecasting with enhanced accuracy

  • 期刊论文

  • 发表日期：

    2025年9月19日

查看全文
  • Talin Azarian,
  • Meisam Yadollahzadeh-Tabari,
  • Yaser Bozorgi-Rad

摘要

Cloud computing has transformed the IT landscape, however its success relies on accurate workload forecasting and efficient resource management. By leveraging historical data to predict future workloads, cloud providers can optimize resource allocation, mitigate SLA violations, and deliver more efficient cloud services. This paper presents CBA-HDL (Convolutional Bidirectional Attention Hybrid Deep Learning), a Hybrid Deep Learning Model that combines 1D Convolutional Neural Networks, Bidirectional Gated Recurrent Units, and an attention mechanism to effectively capture local patterns, long-term dependencies, and key input features, thereby improving workload forecasting accuracy in cloud data centers. This approach facilitates multistep-ahead predictions for CPU and memory utilization, thereby improving accuracy and efficiency of cloud operations. Additionally, the paper discusses data preprocessing techniques such as Savitzky-Golay filtering and min-max normalization, the application of Apache Spark, and Bayesian optimization for hyperparameter tuning. Finally, the CBA-HDL prediction model demonstrated improved performance, with a 40.61% reduction in training time and evaluation metrics of MAE 6.15%, MSE 5.93%, and MAPE 22.14%, outperforming comparable models.