If you need to construct the mathematical model of a plant or process which is explored, investigated, predicted as well as wanted to be perfected or optimized,
and if you have in this case a data sample of observations on behavior of input (independent) and output (dependent) variables,


Even if you:

  • have a short sample of noisy data,
  • are not sure that all relevant variables was measured,
  • have no information on the internal structure of the plant,
  • do not know in what class of structures the model should be constructed,
  • are not so familiar with methods of modelling -


And there is no definitive meaning to what a subject field your problem belongs: to economy or technology, finances or ecology, business or agriculture in any model constructing problem for detection of relationships, prediction and control


Problem-Oriented Program Sysytem For Structure Modelling and Forecasting of Complex Processes from Observation Data

ASTRID is an interactive software system for building mathematical models of complex plants and processes from experimental data under conditions of uncertainty. The built models may be used for detection of relationships, modelling, prediction, simulation, optimization and control.

Originality of the software.
The system is the first full realization of the ideas and algorithms of the Group Method of Data Handling (GMDH) in the form of dialogue program package accessible for users of different skill level. The GMDH-type algorithms differ from the other algorithms (for example of regression type) for identification and building models from experimental data by an active using principles of an automatic generation of model structures variants, non-final decisions, sequential selection of models of an optimal complexity. The method was developed by A.G. Ivakhnenko, a corresponding member of the National Academy of Sciences of Ukraine, and his colleagues (from 1968 up to now). All the algorithms (combinatorial, combinatorial-selective and selective) and procedures realized in ASTRID are original ones.

Comparative features.
The system ASTRID has the following advantages in comparison with software tools based on regression methods or approaches of Akaike, Mallows, Vapnik and others: more wide opportunities for automatic generation of model structures (using multilayer selection procedures); using the external criteria for model selection based on partition of a data sample in two ore more subsamples (that decreases requirements to the a priori information volume); fitness to operate with short data samples; accessibility for users of different skill level; fitness to work in conditions of uncertainty when having limit and noisy data as well as incomplete information on the set of relevant variables and the structure of the internal connections of the plant or process.

Operational and technical parameters.
It enables to build models of static plants, time series as well as dynamic plants and processes in the following classes of structures: linear, polynomial, autoregressive, difference (dynamical) and others.

Application fields.
The system ASTRID may the most expediently be used for solving data analyzing problems with a high level of incompleteness of the available a priori information, that is in problems of building models for detection of relationships, prediction and control in the following fields: economic processes; industrial technology; ecological processes (for instance spread of pollution); aerospace data processing; data compression.
The system ASTRID is applicable separately as well as in computer-aided control systems or decision-making support systems of various levels.

Main  |  A.G.Ivakhnenko  |  GMDH  |  Articles  |  Conference  |  Staff  |  Software  |  Partners  | 

© 2008-2015. Department ITIM. All Rights Reserved.

Administrator Galyna Pidnebesna pidnebesna@irtc.org.ua.