Регуляризація контекстних даних при керуванні автономними системами електроживлення
Вантажиться...
Дата
2016
ORCID
DOI
10.20998/2413-4295.2016.12.31
Науковий ступінь
Рівень дисертації
Шифр та назва спеціальності
Рада захисту
Установа захисту
Науковий керівник
Члени комітету
Видавець
НТУ "ХПІ"
Анотація
Для покращення якості прийняття рішень з керування автономними системами електроживленням створено і протестовано алгоритм регуляризації контекстних даних, що дозволило зменшити помилку прогнозу контекстних часових рядів з (5-6) % до (1,5-2) % та зменшити обсяг операцій при формуванні правил керування напівпровідниковими перетворювачами електроенергії в мережі. Контекстні дані формуються з часових рядів (ЧР), значення яких фіксуються давачами через задані проміжки часу.
In this work we present a regularization of context data of autonomous power supply systems. The autonomous power supply systems is a context awareness framework that aims to provide a comprehensive solution to reason about the context from the level of sensor data to the high-level situation awareness (actuator or devices). The paper describes these challenges and presents data management solutions as a module of context data analysis for the energy control system. These solutions include sensor data acquisition and time series forecasting, ontology model and context prediction model for analytical query processing past and future context data. Context prediction requires the consideration of the preliminary time series processing consists in the detection of the series values anomalous values and series smoothing. The randomness of the commutation, though, leads to the disturbances in power consumption characteristics. Keeping a record of time points and the value of the disturbances complicates the forecasting process and can lead to erroneous results. Filtration or smoothing of context time series is the necessary preliminary prediction stage for obtaining trends. Thus, the first step of the module of context data analysis is the filtration and the second step is the prediction. There are three distinct groups of smoothing: Averaging Methods – moving average, weighted moving average; Exponential Smoothing Methods – simple, weighted, exponential, double; Kalman filter. And three group of prediction: Interpolation – linear, polynomial, spline; Extrapolation – linear, polynomial, French curve, conic; Linear prediction. If the prediction value falls outside the confidence range of prediction errors, the task of regularizing sample n of the prediction method is performed. By sample regularizing we understand sample value alteration up to the value which provides the transition of prediction value to the area of confidence range. The proposed approach of regularization (adaptation) of time series for forecasting method allows reducing forecasting error from 6-5% to 2-1.5%, as the test results showed.
In this work we present a regularization of context data of autonomous power supply systems. The autonomous power supply systems is a context awareness framework that aims to provide a comprehensive solution to reason about the context from the level of sensor data to the high-level situation awareness (actuator or devices). The paper describes these challenges and presents data management solutions as a module of context data analysis for the energy control system. These solutions include sensor data acquisition and time series forecasting, ontology model and context prediction model for analytical query processing past and future context data. Context prediction requires the consideration of the preliminary time series processing consists in the detection of the series values anomalous values and series smoothing. The randomness of the commutation, though, leads to the disturbances in power consumption characteristics. Keeping a record of time points and the value of the disturbances complicates the forecasting process and can lead to erroneous results. Filtration or smoothing of context time series is the necessary preliminary prediction stage for obtaining trends. Thus, the first step of the module of context data analysis is the filtration and the second step is the prediction. There are three distinct groups of smoothing: Averaging Methods – moving average, weighted moving average; Exponential Smoothing Methods – simple, weighted, exponential, double; Kalman filter. And three group of prediction: Interpolation – linear, polynomial, spline; Extrapolation – linear, polynomial, French curve, conic; Linear prediction. If the prediction value falls outside the confidence range of prediction errors, the task of regularizing sample n of the prediction method is performed. By sample regularizing we understand sample value alteration up to the value which provides the transition of prediction value to the area of confidence range. The proposed approach of regularization (adaptation) of time series for forecasting method allows reducing forecasting error from 6-5% to 2-1.5%, as the test results showed.
Опис
Ключові слова
часові ряди, помилка прогнозу, часові ряди, ЧР, root-mean-square error, times series
Бібліографічний опис
Киселева А. Г. Регуляризація контекстних даних при керуванні автономними системами електроживлення / А. Г. Киселёва, Г. Д. Киселёв // Вісник Нац. техн. ун-ту "ХПІ" : зб. наук. пр. Темат. вип. : Нові рішення в сучасних технологіях = Bulletin of National Technical University "KhPI" : coll. of sci. papers. Ser. : New solutions in modern technologies. – Харків : НТУ "ХПІ", 2016. – № 12 (1184). – С. 125-130.