Server load change prediction and stability evaluation based on time series prediction model

Abstract

The rapid development of cloud computing and big data technologies has made the dynamic fluctuation of server load a core factor affecting system performance and stability. Traditional passive load management strategies suffer from lag and are difficult to adapt to complex server operation requirements. Accurate load prediction and stability evaluation are crucial for achieving proactive resource scheduling. This paper aims to construct a complete system for time-series prediction and quantitative evaluation of server load changes, providing a scientific basis for intelligent resource management of server clusters. The study first designs a multi-dimensional server load data collection scheme, extracting core indicators such as CPU load and memory utilization and completing standardized preprocessing. Then, it compares and analyzes the adaptability of traditional time-series models and deep learning models, constructs a deep learning prediction model that integrates attention mechanisms, and optimizes the model through Bayesian optimization and STL time-series decomposition. Simultaneously, it establishes a multi-indicator stability evaluation system from dimensions such as error fluctuation, trend consistency, and peak deviation, combining short-term rolling window evaluation and long-term "densethen- sparse" monitoring methods to achieve a comprehensive judgment of load stability. Experimental results show that the optimized prediction model improves the MAE and RMSE by 36.6% and 39.7% respectively, with an R² of 0.912. Furthermore, the stability evaluation system effectively quantifies the system's operating status under different load scenarios, achieving a good balance between prediction accuracy and computational efficiency. The prediction and evaluation system constructed in this study overcomes the limitations of traditional models in handling nonlinear timeseries data, providing an effective technical tool for dynamic load balancing and resource scheduling of server clusters. It also provides new research ideas and practical references for intelligent server management in cloud computing and edge computing environments.