Journal influence
Higher Attestation Commission (VAK) - К1 quartile
Russian Science Citation Index (RSCI)
Bookmark
Next issue
№2
Publication date:
16 June 2024
Local trends for time series pre-preparation in forecasting problems
Date of submission article: 19.03.2018
UDC: 519.246.8, 004.67, 004.67
The article was published in issue no. № 4, 2018 [ pp. 751-756 ]Abstract:The paper focuses on the studying local trends that describe intermediate movements in non-stationary time series. The first part of the article considers the possibilities of methods of identifying patterns in historical trends using piecewise linear approximation, piecewise logarithmic approximation and the method of local principal components. Local trends have been created using the segmentation method of the bottom-up time series, which allowed identifying the main directions of time series movement. The paper determines the quality criteria and the algorithm for identifying local trends using the proposed methods. There have been some experiments for each time series preprocessing method. It is assumed that the sequence of historical local trends describes the long-term relationship in a time series and might be successfully used for forecasting, for example, based on hybrid neural network methods. The second part of the paper considers the classical application of the Hough transformation for random points approximation on a plane by line segments. There is a disadvantage of this method comparing with the dynamic Hough transformation that takes into account the sample dynamics and can be used in online learning. The authors consider the forecasting algorithm with simultaneous calculation of a local trend using the dynamic Hough transformation. The algorithm is easily extended to other methods of data ap-proximation, which have been considered in the first part of the paper. Computational experiments included real data and used the proposed method. They provided forecasts. The experiments showed that the proposed method helps determining time series trends. The complex periodicity electrocardiogram data and closing prices of Gazprom shares were used for all experiments.
Аннотация:В статье основное внимание уделяется изучению локальных трендов, характеризующих промежуточные движения в нестационарных временных рядах. В первой части работы рассмотрены возможности методов выделения закономерностей в исторических тенденциях с помощью кусочно-линейной аппроксимации, кусочно-логарифмической аппроксимации и метода локальных главных компонент. Построение локальных трендов проводилось с помощью метода сегментации временного ряда «снизу вверх», который позволил выявить основные направления движения временного ряда. Определены критерии качества и алгоритм выделения локальных трендов с помощью перечисленных методов. Проведены эксперименты для каждого метода предобработки временного ряда. Предполагается, что последовательность исторических локальных трендов описывает долгосрочную взаимосвязь во временном ряду и может быть успешно использована для прогнозирования, например, на основе гибридных нейросетевых методов. Во второй части работы рассмотрено классическое применение преобразования Хафа для аппроксимации случайных точек на плоскости отрезками прямых. Показан недостаток этого метода по сравнению с динамическим преобразованием Хафа, который учитывает динамику выборки и может быть использован в онлайн-обучении. Рассмотрен алгоритм прогноза с одновременным вычислением локального тренда с помощью динамического преобразования Хафа. Алгоритм легко распространяется на остальные способы аппроксимации данных, описанные в первой части работы. Проведены вычислительные эксперименты на реальных данных с помощью предложенного метода и получены прогнозы. Эксперименты показали возможность предлагаемого метода определять тенденции во временном ряду. Для всех экспериментов использованы данные c электрокардиограммы со сложной периодичностью и цены закрытий акций Газпрома.
Authors: Puchkov E.V. (puchkoff@i-intellect.ru) - Rostov State University of Civil Engineering, Rostov-on-Don, Russia, Ph.D, Belyavsky G.I. (gbelyavski@sfedu.ru) - Scientific Reseach Institute of Mechanics and Applied Mathematics Southern Federal University, Rostov-on-Don, Russia, Ph.D | |
Keywords: forecasting, time series, hough transformation, principal components, approximation, local trend |
|
Page views: 12094 |
PDF version article Full issue in PDF (22.98Mb) |
Применение локальных трендов для предподготовки временных рядов в задачах прогнозирования
DOI: 10.15827/0236-235X.124.751-756
Date of submission article: 19.03.2018
UDC: 519.246.8, 004.67, 004.67
The article was published in issue no. № 4, 2018. [ pp. 751-756 ]
The paper focuses on the studying local trends that describe intermediate movements in non-stationary time series.
The first part of the article considers the possibilities of methods of identifying patterns in historical trends using piecewise linear approximation, piecewise logarithmic approximation and the method of local principal components. Local trends have been created using the segmentation method of the bottom-up time series, which allowed identifying the main directions of time series movement. The paper determines the quality criteria and the algorithm for identifying local trends using the proposed methods. There have been some experiments for each time series preprocessing method. It is assumed that the sequence of historical local trends describes the long-term relationship in a time series and might be successfully used for forecasting, for example, based on hybrid neural network methods.
The second part of the paper considers the classical application of the Hough transformation for random points approximation on a plane by line segments. There is a disadvantage of this method comparing with the dynamic Hough transformation that takes into account the sample dynamics and can be used in online learning. The authors consider the forecasting algorithm with simultaneous calculation of a local trend using the dynamic Hough transformation. The algorithm is easily extended to other methods of data ap-proximation, which have been considered in the first part of the paper.
Computational experiments included real data and used the proposed method. They provided forecasts. The experiments showed that the proposed method helps determining time series trends. The complex periodicity electrocardiogram data and closing prices of Gazprom shares were used for all experiments.
Puchkov E.V. (puchkoff@i-intellect.ru) - Rostov State University of Civil Engineering, Rostov-on-Don, Russia, Ph.D, Belyavsky G.I. (gbelyavski@sfedu.ru) - Scientific Reseach Institute of Mechanics and Applied Mathematics Southern Federal University, Rostov-on-Don, Russia, Ph.D
Ссылка скопирована!
Permanent link: http://swsys.ru/index.php?page=article&id=4536&lang=en |
Print version Full issue in PDF (22.98Mb) |
The article was published in issue no. № 4, 2018 [ pp. 751-756 ] |
The article was published in issue no. № 4, 2018. [ pp. 751-756 ]
Perhaps, you might be interested in the following articles of similar topics:Perhaps, you might be interested in the following articles of similar topics:
- Адекватные междисциплинарные модели в прогнозировании временных рядов статистических данных
- Интеллектуальный подход к задаче информирования и краткосрочного прогнозирования временного ряда в онлайн-режиме
- Нейронные сети и модели ARIMA для прогнозирования котировок
- Прогнозирование временного ряда инфекционной заболеваемости
- Система раннего предупреждения о нарушении показателей качества питьевой воды
Back to the list of articles