Historical database training method for neural networks

• US 5,640,493 A
• Filed: 04/17/1995
• Issued: 06/17/1997
• Est. Priority Date: 08/03/1990
• Status: Expired due to Term

• Current Assignee: Rockwell Automation Technologies Incorporated (Allen-Bradley Co., Inc.)
• Original Assignee: E. I. du Pont de Nemours and Company (Corteva, Inc.)
• Inventors: Skeirik, Richard D.
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/422,955
• Time in Patent Office: 792 Days
• Field of Search: 382/12-15, 382/155-159, 395/20-25, 395/27, 395/26, 364/420
• US Class Current: 706/25
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

An on-line training neural network for controlling a process for producing a product having at least one product property that trains by retrieving training sets from a stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network. When multiple presentations are needed to effectively train, a buffer of training sets is filled and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed. An historical database of timestamped data can be used to construct training sets when training input data has a time delay from sample time to availability for the neural network. The timestamps of the training input data are used to select input data for use in the training set. Using the historical database, the neural network can be trained retrospectively by searching the historical database and constructing training sets based on past data.

Claims

1. A method for constructing training sets for a neural network, comprising the steps of:

(1) developing a first training set for a neural network by;

(a) retrieving from an historical database first training input data having a first timestamp(s);

(b) selecting a first training input data time based on said first timestamp(s);

(c) retrieving a first input data indicated by said first training input data time; and

(2) developing a second training set for said neural network by;

(a) retrieving from said historical database second training input data having a second timestamp(s);

(b) selecting a second training input data time based on said second timestamp(s);

(c) retrieving a second input data indicated by said second training input data time.

2. The method of claim 1, further comprising a step of:

(3) searching said historical database in either a forward time direction or a backward time direction so that said second training input data is the next training input data in time to said first training input data in said forward time direction or said backward time direction, whichever is used.

3. The method of claim 1, further comprising the step of (3) training or retraining said neural network using said first training set and/or said second training set.

4. A computer neural network process control method adapted for predicting output data provided to a controller used to control a process for producing a product having at least one product property, the computer neural network process control method comprising the steps of:

(1) monitoring for the availability of new training input data by monitoring for a change in an associated timestamp of said training input data;

(2) constructing a training set by retrieving first input data corresponding to said training input data;

(3) training the neural network using said training set; and

(4) predicting the output data from second input data using the neural network.

5. The computer neural network process control method of claim 4, wherein step (2) further comprises a step of using data pointers to indicate said training input data and said first input data.

6. The computer neural network process control method of claim 4, wherein:

(a) step (1) is preceded by the steps;

(i) presenting to a user a template for a partially specified neural network; and

(ii) entering data into said template to create a complete neural network specification; and

(b) step (3) further comprises a step of using a neural network representative of said complete neural network specification.

7. The computer neural network process control method of claim 4, wherein:

(a) step (1) is preceded by;

(i) presenting to a user an interface for accepting a limited set of natural language format specifications; and

(ii) entering into said interface sufficient specifications in said substantially natural language format to completely define a neural network; and

(b) step (3) further comprises a step of using a neural network representative of said completely defined neural network.

8. The computer neural network process control method of claim 4, wherein steps (1), (2), and (3) operate in real-time.

9. A method for constructing training sets for a neural network, comprising the steps of:

(a) retrieving, from an historical database, training input data having a timestamp(s);

(b) selecting a training input data time based on said timestamp(s); and

(c) retrieving an input data indicated by said training input data time.

10. A computer neural network process control method adapted for predicting output data provided to a controller used to control a process for producing a product having at least one product property, the computer neural network process control method comprising the steps of:

(1) monitoring for the availability of new training input data;

(2) constructing a training set by retrieving first input data corresponding to said training input data including the steps of,

(a) selecting a training input data time using a timestamp(s) associated with said training input data; and

(b) retrieving input data representing measurement(s) at said training input data time, said input data comprising said first input data;

(3) training the neural network using said training set; and

(4) predicting the output data from second input data using the neural network.

11. The computer neural network process control method of claim 10, wherein step (1) comprises a step of monitoring for a change between two successive training input data values.

12. The computer neural network process control method of claim 10, wherein step (1) comprises a step of computing a difference between a most recent training input data value and a next most recent training input value;

and

wherein step (3) further comprises a step of using said difference with said first input data for said training.

13. The computer neural network process control method of claim 10, wherein step (2) further comprises a step of using data pointers to indicate said training input data and said first input data.

14. The computer neural network process control method of claim 10, wherein steps (1), (2) and (3) operate in real-time.

15. A computer neural network process control method adapted for predicting output data provided to a controller used to control a process for producing a product having at least one product property, the computer neural network process control method comprising the steps of:

(1) presenting to a user a template for a partially specified neural network;

(2) entering data into said template to create a complete neural network specification;

(3) monitoring for the availability of new training input data;

(4) constructing a training set by retrieving first input data corresponding to said training input data;

(5) training the neural network using said training set, said training step further including using a neural network representative of said complete neural network specification; and

(6) predicting the output data from second input data using the neural network.

16. The computer neural network process control method of claim 15, wherein step (3) comprises a step of monitoring for a change between two successive training input data values.

17. The computer neural network process control method of claim 15, wherein step (3) comprises a step of computing a difference between a most recent training input data value and a next most recent training input value;

and

wherein step (5) further comprises a step of using said difference with said first input data for said training.

18. The computer neural network process control method of claim 15, wherein step (4) further comprises a step of using data pointers to indicate said training input data and said first input data.

19. The computer neural network process control method of claim 15, wherein steps (3), (4) and (5) operate in real-time.

20. A computer neural network process control method adapted for predicting output data provided to a controller used to control a process for producing a product having at least one product property, the computer neural network process control method comprising the steps of:

(1) presenting to a user an interface for accepting a limited set of natural language format specifications;

(2) entering into said interface sufficient specifications in said substantially natural language format to completely define a neural network;

(3) monitoring for the availability of new training input data;

(4) constructing a training set by retrieving first input data corresponding to said training input data;

(5) training the neural network using said training set, wherein said training step further includes using a neural network representative of said completely defined neural network; and

(6) predicting the output data from second input data using the neural network.

21. The computer neural network process control method of claim 20, wherein step (3) comprises a step of monitoring for a change between two successive training input data values.

22. The computer neural network process control method of claim 20, wherein step (3) comprises a step of computing a difference between a most recent training input data value and a next most recent training input value;

and

wherein step (5) further comprises a step of using said difference with said first input data for said training.

23. The computer neural network process control method of claim 20, wherein step (4) further comprises a step of using data pointers to indicate said training input data and said first input data.

24. The computer neural network process control method of claim 20, wherein steps (3), (4) and (5) operate in real-time.