Developing a Deep Learning-Based Multimodal Intelligent Cloud Computing Resource Load Prediction System

Abstract

This study aims to predict the dynamic changes in critical cloud computing resource indicators, namely Central Processing Unit (CPU), Random Access Memory (RAM), hard disk (Disk), and network. Its primary objective is to optimize resource allocation strategies in advance to enhance overall system performance. The research employs various deep learning algorithms, including Simple Recurrent Neural Network (SRNN), Bidirectional Simple Recurrent Neural Network (BiSRNN), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Through experimentation with different algorithm combinations, the study identifies optimal models for each specific resource indicator. Results indicate that combining CNN, LSTM, and GRU yields the most effective predictions for CPU load, while CNN and LSTM together are optimal for RAM load prediction. For disk load prediction, GRU alone proves optimal, and BiSRNN emerges as the optimal choice for network load prediction. The training results of these models demonstrate R-squared values (R²) exceeding 0.98, highlighting their high accuracy in predicting future resource dynamics. This precision facilitates timely and efficient resource allocation, thereby enhancing system responsiveness. The study's multimodal precise prediction capability supports prompt and effective resource allocation, further enhancing system responsiveness. Ultimately, this approach significantly contributes to sustainable digital advancement for enterprises by ensuring efficient resource allocation and consistent optimization of system performance. The study underscores the importance of integrating advanced deep learning techniques in managing cloud computing resources, thereby supporting the robust and sustainable growth of digital infrastructure.