A researcher from the Siberian State University of Telecommunications and Informatics (SibGUTI) has developed an algorithm based on machine learning that allows for quick and accurate analysis of time series. This method is proposed to use weather forecasting, exchange rates, market conditions and other processes.

A time series is any sequence of events distributed over time, explains SibGUT. A highly effective method of forecasting time series was previously proposed by the mathematician Borys Yakovych Ryabko, but its implementation requires significant efforts and complex mathematical calculations. The new algorithm makes it possible to effectively apply this method, ensuring maximum accuracy and speed of analysis.

“This algorithm makes it possible to implement the method of forecasting time series in such a way that its labor intensity reaches the maximum possible efficiency. It can work in online mode, read large data at once, process and immediately issue forecasts at high speed,” said the author of the algorithm, associate professor of the department of applied mathematics and cybernetics of SibGUTI, Anton Rakitskyi.

Currently, a team of scientists is developing a library of algorithms and methods for time series analysis.

“This is a set of various methods, effectively implemented and optimized, which allow you to predict timelines. We already have methods for reading, for preprocessing and preparing data, for switching between alphabets and other things,” Rokytsky clarifies.

The library will be able to be used by software developers to analyze specific time series and solve the tasks of various companies or departments.

Next year, SibGUTI plans to start work on improving the forecasting of gas consumption for the company “Gazprom Mizhregiongaz Novosibirsk”, the university’s industrial partner. The scientists’ task is to reduce the forecasting error from 10% to 3%. The team plans to create a model that will provide accurate forecasts of gas consumption for different categories of users.