350 rub
Journal Radioengineering №5 for 2009 г.
Article in number:
Using of Time Series and Neural Networks Models for Forecasting Degradation Parameters Integral Circuits
Keywords:
integrated circuits
electrical parameters
degradation
prognosis
parametric fault
model
time series
Box
Jenkins
neural networks
Authors:
A.V. Strogonov
Abstract:
By the example of forecasting process degradation TTL IC such as 106LB1 release of january 1980 y. at tests for durability in 130 th.h. On extreme values of parameter the opportunity of use time-series models (Box-Jenkins time-series models or autoregressive integrated moving-average, ARIMA-models) and neural networks (NN) is shown. Time-series degradation parameter IC at tests for durability contains the data with misses.
It is offered to use some methods of reception of missing values - approximation by cubic splines, a method of interpolation and forecasts of linear regress, filling with average value of lines.
Graphic verification of forecasts of models of time-series is lead. The basic idea of graphic verification of forecasts consists that all over again on the basis of some rational criterion it is necessary to divide available data file about degradation of controllable parameters IC, on subsets, and then to use one or several parts of these data for construction of "forecasting" model, the rests to use for "check" ("examination") of this model.
Forecasting is conducted down to approach of parametrical failure. For parametrical failure achievement of the top 90 % by border of a confidential interval of time-series model failure a level established in specifications, equal 0.35 V is accepted.
Check of reliability of long-term forecasts was carried out by construction statistically possible trajectories of time-series model a method of Monte Carlo by generation of values white noise with the normal law of distribution.
Short-term forecasts ARMA-and ARIMA-models of time-series and the multiple layers of neurons network, a network with radial basic elements, generalized regression a network being a version of NN with radial basis elements, a linear network are constructed by the example of process degradation of the worst values of parameter TTL IC such as 133LA8 at tests for durability in 150 th.h
Pages: 4-9
References
- Бокс Дж., Дженкинс Г. Анализ временных рядов. Прогноз и управление. М.: Мир. 1974.
- Боровиков В.П., Ивченко Г.И. Прогнозирование в системе STATISTICA в среде Windows. Основы теории и интенсивная практика на компьютере: Учеб. пособие. М.: Финансы и статистика. 1999.
- Песаран М., Слейтер Л. Динамическая регрессия: Теория и алгоритмы: Пер. с англ.; Предисл. Э.Б. Ершова. М.: Финансы и статистика. 1984.
- Льюнг Л. Идентификация систем. Теория для пользователя: Пер. с анг. / под ред. Я.З. Цыпкина. М.: Наука. Гл. ред. физ.-мат. лит., 1991.
- Справочник по прикладной статистике. В 2-х т. Т.2: Пер. с англ. / под ред. Э. Лойда, У. Ледермана, С.А. Айвазяна, Ю.Н. Тюрина. М.: Финансы и статистика, 1990. 526 с.
- Кей С.М., Марпл С.Л. Современные методы спектрального анализа: Обзор // ТИИЭР. 1981. Т.69. №11. С. 5-43.
- Тюрин Ю.Н., Макаров А.А. Статистический анализ на компьютере / под ред. В.Э. Фигурнова. М.: ИНФРА-М. 1998.
- Медведев Г.А., Морозов В.А. Практикум на ЭВМ по анализу временных рядов. Минск: Электронная книга БГУ. 2003.
- Грешилов А.А., Стакун В.А., Стакун А.А. Математические методы построения прогнозов. М.: Радио и связь. 1997.