USING MACHINE LEARNING METHODS FOR AUTOMATED CLOUD COMPUTING OPTIMIZATION

Abstract

This study is devoted to the development and experimental validation of a comprehensive approach to the automated optimization of cloud computing through the use of machine learning methods. The proposed solution – an intelligent adaptive orchestrator – integrates three key components: workload forecasting based on time-series models (LSTM, Prophet), dynamic resource management using reinforcement learning methods (PPO), and an anomaly-detection module employing autoencoders and statistical techniques.The aim of the article. To design, implement, and validate an intelligent adaptive orchestration system that eliminates critical limitations of traditional cloud resource management.Scientific novelty. It lies in the design of a system with a modular architecture that ensures scalability, fault tolerance, and flexible adaptation to diverse business objectives through dynamic tuning of the reinforcement learning agent’s reward functions, with integration with container orchestration platforms (e.g., Kubernetes) and support for multi-cloud deployments.The conclusions. Within this study, an intelligent orchestrator for cloud resource management was developed, implemented, and experimentally validated, built on the integration of workload forecasting, reinforcement learning, and anomaly detection methods. Experiments conducted both on a controlled laboratory testbed and in the real industrial hybrid infrastructure of SoftRequest LTD confirmed the high effectiveness of the proposed solution. The practical value of the approach lies in the ability to integrate directly with existing orchestration platforms, such as Kubernetes, without the need for substantial infrastructure rework.