RESEARCH AND FORECASTING OF TIME SERIES USING PARALLEL COMPUTING TECHNOLOGIES

This study addresses the challenge of efficient time series processing for forecasting purposes using distributed computing technologies in a cloud environment. The focus is placed on adapting modern approaches to time series analysis for handling large data volumes and integrating them with cloud computing infrastructure. Particular attention is given to processing ultra-long time series, characterized by low signal-to-noise ratios, complex structures, and long-term trends. A wide range of forecasting methods is analyzed, including classical statistical models such as autoregressive integrated moving average (ARIMA) and modern machine learning approaches, particularly long short-term memory neural networks. The advantages of parallel computing in significantly accelerating the processing of large data volumes are demonstrated. Specifically, the study confirms the effectiveness of the proposed approach using Amazon Web Services cloud infrastructure, enabling resource optimization and improving forecasting accuracy. A software package based on Apache Spark technologies was developed for time series analysis in distributed environments. Performance testing of the software demonstrated its practical applicability for solving forecasting and anomaly detection tasks in large time series. The application of the adapted autoregressive integrated moving average model, combined with parallel computing, is substantiated as an effective method for time series forecasting. The challenges associated with implementing parallel computing for time series forecasting are explored, including the need for algorithm optimization and ensuring scalability of solutions within a cloud environment. The study outlines prospects for further software enhancements, such as integrating adaptive algorithms and expanding their application to fields like cybersecurity, financial analytics, infrastructure monitoring, and forecasting in economics and industry. The results of extensive computational experiments confirm the effectiveness of the developed algorithms in improving forecast accuracy and reducing data processing time. These findings lay the foundation for future research aimed at creating comprehensive time series analysis systems that account for the specific needs of various industries.