Method for the computer-supported generation of prognoses for operative systems and a system for the generation of prognoses for operative systems

• US 7,480,535 B2
• Filed: 09/29/2003
• Issued: 01/20/2009
• Est. Priority Date: 09/30/2002
• Status: Active Grant

• Current Assignee: Gerhard Kranner
• Original Assignee: Gerhard Kranner
• Inventors: Kranner, Gerhard
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US10/529,610
• Publication Number:
  US 20060009864A1
• Time in Patent Office: 1,940 Days
• Field of Search: 700/276, 621/26
• US Class Current: 700/28
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 23/0248 Causal models, e.g. fault tree; digraphs; qualitative physics
• CPC Class Codes: G06F 18/2137 based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps

Abstract

The invention relates to a method for the computer-supported generation of prognoses for operative systems (37), in particular for control processes and similar, based on multi-dimensional data sets describing a system, product and/or process condition, using the SOM method in which an ordered grid of nodes (1), representing the data distribution is determined and an internal scaling (σ_j) of variables (x_j) with regard to non-linearity in the data, based on the non-linear influence of each variable on the prognosis variable, is carried out. Local receptive regions corresponding to the nodes (1) are determined, on the basis of which local linear regressions are calculated. Using the number of local prognosis models obtained thus, optimized prognosis values for the control of the operative system (37) are calculated, whereby for each new data set the relevant sufficient nodes are determined and the local prognosis model applied to this data set.

Claims

1. A computerized method for generating prognoses for operative systems on a basis of multidimensional data records describing a state of a system, product and/or process, by utilizing the self-organizing maps method, which comprises:

determining an ordered grid of nodes representing a data distribution;

for taking into account nonlinearities in the data, effecting an internal scaling of variables based on a nonlinear influence of each variable on a prediction variable;

determining local receptive regions assigned to the nodes and calculating local linear regressions on a basis of the local receptive regions;

calculating optimized prediction values for controlling the operative system with the aid of a set of local prediction models thus obtained, by determining a respectively adequate node for each new data record and applying a local prediction model to the data record; and

at least one of;

outputting at least one of the local prediction models or optimized prediction values for display on a visualization unit; and

controlling the operative system based on the calculated optimized prediction values.

2. The method according to claim 1, wherein the operative system is a set of control processes.

3. The method according to claim 1, wherein the data sets are descriptive of a state selected from the group consisting of system states, product states, and process states.

4. The method according to claim 1, which comprises forming for each variable a measure for an order of the variable in the SOM representation and forming a measure for a contribution of the variable to the explained variance, determining new internal scalings from the measures on a basis that an estimated change in the explained variance is maximized by varying the internal scalings, to thereby order the variables in a resulting SOM representation in accordance with respective contributions thereof to the explained variance and to more accurately resolve the existing nonlinearities.

5. The method according to claim 1, wherein the step of determining local receptive regions assigned to the nodes comprises respectively selecting a magnitude thereof to be so large that the explained variance of the local regression is at a maximum, while simultaneously safeguarding a significance and a stability in the region of the node.

6. The method according to claim 5, wherein the step of determining local receptive regions assigned to the nodes comprises selecting in each case a receptive region, which is the smallest necessary receptive region, for a significance of the regression, and is the largest possible receptive region for maximizing an accuracy of the prognosis.

7. The method according to claim 1, which comprises performing the internal scaling iteratively.

8. The method according to claim 1, which comprises preliminarily subjecting supplied data in advance to compensating scaling to at least partially compensate for possible correlations between variables.

9. The method according to claim 8, wherein the compensating scaling comprises rescaling individual data sets, with values of a respective variable of all the data sets being standardized, thereafter transforming the data into a principal component space and calculating the compensating scalings of the individual variables on a basis that a distance measure in the original variable space differs minimally from a distance measure in the standardized principal component space.

10. The method according to claim 9, which comprises multiplicatively combining the compensating scaling with the internal scaling that takes account of the nonlinearities in the data, for forming a combined variable scaling on which a thusly modified SOM representation is based.

11. A system for generating prognoses for operative systems on a basis of multidimensional data records describing a state of a system, product and/or process, comprising:

a database for storing the data records;

a self-organizing maps (SOM) unit for determining an ordered grid of nodes representing a data distribution;

a nonlinearity feedback unit associated with said SOM unit and configured for internal scaling of variables in order to compensate a nonlinear influence thereof on a prediction variable;

a calculation unit associated with said SOM unit and configured for determining local linear regressions on a basis of local receptive regions assigned to the nodes;

a prediction unit configured to use local prediction models thus obtained as a basis for calculating optimized prognosis values by determining a respectively adequate node for each new data record and to apply the local prediction model to the respective data record; and

at least one of;

a visualization unit for visually displaying at least one of the local prediction models or optimized prediction values; and

at least one control unit for controlling the operative system based on the calculated optimized prediction values.

12. The system according to claim 11 configured to generate prognoses for control processes.

13. The system according to claim 11, which comprises a plurality of control units, assigned to individual process states, connected to said prediction unit and configured to predict process results that would arise from current process data.

14. The system according to claim 13, which comprises process units separately connected to said control units, said process units being configured to derive control parameters on a. basis of the predicted process results and of desired values for the process respectively to be carried out in the operative system.

15. A computerized method for generating prognoses for operative systems for automatic control of at least one of marketing measures, financial transactions and industrial production on a basis of multidimensional data records describing a state of a customer, system, product and/or process, by utilizing the self-organizing maps method, which comprises:

determining an ordered grid of nodes representing a data distribution;

for taking into account nonlinearities in the data, effecting an internal scaling of variables based on a nonlinear influence of each variable on a prediction variable;

determining local receptive regions assigned to the nodes and calculating local linear regressions on a basis of the local receptive regions;

calculating optimized prediction values for controlling the operative system with the aid of a set of local prediction models thus obtained, by determining a respectively adequate node for each new data record describing marketing, financial, production or processing properties, and applying a local prediction model to the data record; and

automatically controlling the operative system based on the calculated optimized prediction values.

16. The method according to claim 15, wherein the operative system is a set of control processes.

17. The method according to claim 15, wherein the data sets are descriptive of a state selected from the group consisting of system states, product states, and process states.

18. The method according to claim 15, which comprises forming for each variable a measure for an order of the variable in the SOM representation and forming a measure for a contribution of the variable to the explained variance, determining new internal scalings from the measures on a basis that an estimated change in the explained variance is maximized by varying, the internal scalings, to thereby order the variables in a resulting SOM representation in accordance with respective contributions thereof to the explained variance and to more accurately resolve the existing nonlinearities.

19. The method according to claim 15, wherein the step of determining local receptive regions assigned to the nodes comprises respectively selecting a magnitude thereof to be so large that the explained variance of the local regression is at a maximum, while simultaneously safeguarding a significance and a stability in the region of the node.

20. The method according to claim 19, wherein the step of determining local receptive regions assigned to the nodes comprises selecting in each case a receptive region, which is the smallest necessary receptive region, for a significance of the regression, and is the largest possible receptive region for maximizing an accuracy of the prognosis.

21. The method according to claim 15, which comprises performing the internal scaling iteratively.

22. The method according to claim 15, which comprises preliminarily subjecting supplied data in advance to compensating scaling to at least partially compensate for possible correlations between variables.

23. The method according to claim 22, wherein the compensating scaling comprises rescaling individual data sets, with values of a respective variable of all the data sets being standardized, thereafter transforming the data into a principal component space and calculating the compensating scalings of the individual variables on a basis that a distance measure in the original variable space differs minimally from a distance measure in the standardized principal component space.

24. The method according to claim 23, which comprises multiplicatively combining the compensating scaling with the internal scaling that takes account of the nonlinearities in the data, for forming a combined variable scaling on which a thusly modified SOM representation is based.