Recurrent neural network-based fuzzy logic system and method

• US 5,828,812 A
• Filed: 05/02/1995
• Issued: 10/27/1998
• Est. Priority Date: 03/24/1993
• Status: Expired due to Term

• Current Assignee: National Semiconductor Corporation (Texas Instruments, Inc.)
• Original Assignee: National Semiconductor Corporation (Texas Instruments, Inc.)
• Inventors: Neely, William Shields, Khan, Emdadur Rahman
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Rhodes, Jason W.
• Application Number:
  US08/432,801
• Time in Patent Office: 1,274 Days
• Field of Search: 395/3, 395/61, 395/900, 395/21, 395/24, 395/22, 395/23
• US Class Current: 706/2
• CPC Class Codes: G06N 3/043 based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
• CPC Class Codes: G06N 7/046 Implementation by means of a neural network neural networks using fuzzy logic G06N3/043
• CPC Class Codes: Y10S 706/90 Fuzzy logic

Abstract

A recurrent, neural network-based fuzzy logic system includes in a rule base layer and a membership function layer neurons which each have a recurrent architecture with an output-to-input feedback path including a time delay element and a neural weight. Further included is a recurrent, neural network-based fuzzy logic rule generator wherein a neural network receives and fuzzifies input data and provides data corresponding to fuzzy logic membership functions and recurrent fuzzy logic rules.

Claims

1. A recurrent neural network-based fuzzy logic apparatus comprising:

first neural means for receiving and processing first and second pluralities of input signals and in accordance therewith providing first and second pluralities of neural signals, wherein said first plurality of input signals originates externally to said first neural means and said first plurality of neural signals corresponds to a first plurality of fuzzy logic rule antecedents;

first time-delay means for intra-neurally receiving and time-delaying said second plurality of neural signals and in accordance therewith providing a first plurality of intra-neural recurrent signals as said second plurality of input signals;

second neural means for receiving and processing said first plurality of neural signals from said first neural means and a plurality of antecedent signals and in accordance therewith providing a plurality of consequent signals, wherein said plurality of antecedent signals corresponds to a second plurality of fuzzy logic rule antecedents and said plurality of consequent signals corresponds to a plurality of fuzzy logic rule consequents; and

second time-delay means for intra-neurally receiving and time-delaying a portion of said plurality of consequent signals and in accordance therewith providing a second plurality of intra-neural recurrent signals as said plurality of antecedent signals.

2. The apparatus of claim 1, wherein said first neural means comprises a plurality of neurons for fuzzifying selected ones of said first and second pluralities of input signals and in accordance therewith providing said first plurality of neural signals.

3. The apparatus of claim 1, wherein said first neural means comprises a plurality of product-based neurons.

4. The apparatus of claim 1, wherein said first neural means comprises a plurality of sum-based neurons.

5. The apparatus of claim 1, wherein said first time-delay means comprises a plurality of time-delay elements, each for delaying by at least one unit time delay and selectively weighting a respective one of said second plurality of neural signals and in accordance therewith providing a respective one of said second plurality of input signals.

6. The apparatus of claim 1, wherein said second neural means comprises a plurality of product-based neurons.

7. The apparatus of claim 1, wherein said second neural means comprises a plurality of sum-based neurons.

8. The apparatus of claim 1, wherein said second time-delay means comprises a plurality of time-delay elements, each for delaying by at least one unit time delay and selectively weighting a respective one of said portion of said plurality of consequent signals and in accordance therewith providing a respective one of said plurality of antecedent signals.

9. A neural network apparatus for generating signals corresponding to a plurality of recurrent fuzzy logic rules, comprising:

a first plurality of neurons for receiving and selectively fuzzifying a plurality of input signals and in accordance therewith providing a plurality of fuzzified signals;

a second plurality of neurons, coupled to said first plurality of neurons, for receiving said plurality of fuzzified signals therefrom and a first plurality of intra-neural recurrent signals and in accordance therewith providing a plurality of membership signals corresponding to a plurality of fuzzy logic membership functions;

a first plurality of time-delay elements, coupled to said second plurality of neurons, for intra-neurally receiving and time-delaying a portion of said plurality of membership signals and in accordance therewith providing said first plurality of intra-neural recurrent signals; and

a third plurality of neurons, coupled to said second plurality of neurons, for receiving said plurality of membership signals therefrom and a second plurality of intra-neural recurrent signals and in accordance therewith providing a plurality of rule signals corresponding to a plurality of recurrent fuzzy logic rules.

10. The apparatus of claim 9, further comprising a second plurality of time-delay elements, coupled to said third plurality of neurons, for receiving and time-delaying a portion of said plurality of rule signals and in accordance therewith providing said second plurality of recurrent signals.

11. The apparatus of claim 9, wherein said third plurality of neurons is further for providing a defuzzified output signal in accordance with said plurality of membership signals and said second plurality of recurrent signals.

12. The apparatus of claim 11, further comprising a comparator, coupled to said third plurality of neurons, for receiving and comparing a reference signal and said defuzzified output signal and in accordance therewith providing an error signal corresponding to a difference therebetween.

13. The apparatus of claim 9, further comprising an output neuron layer, coupled to said third plurality of neurons, for receiving and summing a plurality of signals therefrom and in accordance therewith providing a defuzzified output signal.

14. The apparatus of claim 13, further comprising a comparator, coupled to said output neuron layer, for receiving and comparing a reference signal and said defuzzified output signal and in accordance therewith providing an error signal corresponding to a difference therebetween.

15. A neural network apparatus for generating a plurality of recurrent fuzzy logic rules, comprising:

first neural means for receiving and selectively fuzzifying a plurality of input data and in accordance therewith providing a plurality of fuzzified data;

second neural means for receiving said plurality of fuzzified data from said first neural means and a first plurality of intra-neural recurrent data and in accordance therewith providing a plurality of membership data corresponding to a plurality of fuzzy logic membership functions;

first time-delay means for intra-neurally receiving and time-delaying a portion of said plurality of membership data and in accordance therewith providing said first plurality of intra-neural recurrent data; and

third neural means for receiving said plurality of membership data from said second neural means and a second plurality of intra-neural recurrent data and in accordance therewith providing a plurality of rule data corresponding to a plurality of recurrent fuzzy logic rules.

16. The apparatus of claim 15, further comprising second time-delay means for receiving and time-delaying a portion of said plurality of rule data and in accordance therewith providing said second plurality of recurrent data.

17. The apparatus of claim 15, wherein said third neural means is further for providing defuzzified output data in accordance with said plurality of membership data and said second plurality of recurrent data.

18. The apparatus of claim 17, further comprising comparison means for receiving and comparing reference data and said defuzzified output data and in accordance therewith providing error data corresponding to a difference therebetween.

19. The apparatus of claim 15, further comprising output neural means for receiving and summing a plurality of data from said third neural means and in accordance therewith providing defuzzified output data.

20. The apparatus of claim 19, further comprising comparison means for receiving and comparing reference data and said defuzzified output data and in accordance therewith providing error data corresponding to a difference therebetween.

21. A method of signal processing in accordance with recurrent neural network-based fuzzy logic, said method comprising the steps of:

receiving and processing first and second pluralities of input signals with a first artificial neural apparatus and in accordance therewith providing first and second pluralities of neural signals, wherein said first plurality of input signals originates externally to said first artificial neural apparatus and said first plurality of neural signals corresponds to a first plurality of fuzzy logic rule antecedents;

intra-neurally time-delaying said second plurality of neural signals and in accordance therewith providing a first plurality of intra-neural recurrent signals as said second plurality of input signals;

processing said first plurality of neural signals from said first artificial neural apparatus and a plurality of antecedent signals with a second artificial neural apparatus and in accordance therewith providing a plurality of consequent signals, wherein said plurality of antecedent signals corresponds to a second plurality of fuzzy logic rule antecedents and said plurality of consequent signals corresponds to a plurality of fuzzy logic rule consequents; and

intra-neurally time-delaying a portion of said plurality of consequent signals and in accordance therewith providing a second plurality of intra-neural recurrent signals as said plurality of antecedent signals.

22. The method of claim 21, wherein said step of receiving and processing first and second pluralities of input signals with a first artificial neural apparatus and in accordance therewith providing first and second pluralities of neural signals comprises fuzzifying selected ones of said first and second pluralities of input signals with a plurality of artificial neurons and in accordance therewith providing said first plurality of neural signals.

23. The method of claim 21, wherein said step of receiving and processing first and second pluralities of input signals with a first artificial neural apparatus and in accordance therewith providing first and second pluralities of neural signals comprises processing said first and second pluralities of input signals with a plurality of product-based artificial neurons.

24. The method of claim 21, wherein said step of receiving and processing first and second pluralities of input signals with a first artificial neural apparatus and in accordance therewith providing first and second pluralities of neural signals comprises processing said first and second pluralities of input signals with a plurality of sum-based artificial neurons.

25. The method of claim 21, wherein said step of time-delaying said second plurality of neural signals and in accordance therewith providing said second plurality of input signals comprises delaying by at least one unit time delay and selectively weighting individual ones of said second plurality of neural signals.

26. The method of claim 21, wherein said step of processing said first plurality of neural signals and a plurality of antecedent signals with a second artificial neural apparatus and in accordance therewith providing a plurality of output signals and a plurality of consequent signals comprises processing said first plurality of neural signals and a plurality of antecedent signals with a plurality of product-based artificial neurons.

27. The method of claim 21, wherein said step of processing said first plurality of neural signals and a plurality of antecedent signals with a second artificial neural apparatus and in accordance therewith providing a plurality of output signals and a plurality of consequent signals comprises processing said first plurality of neural signals and a plurality of antecedent signals with a plurality of sum-based artificial neurons.

28. The method of claim 21, wherein said step of time-delaying a portion of said plurality of consequent signals and in accordance therewith providing said plurality of antecedent signals comprises delaying by at least one unit time delay and selectively weighting individual ones of said plurality of consequent signals.

29. A method of generating a plurality of recurrent fuzzy logic rules, comprising:

receiving and selectively fuzzifying a plurality of input data with a first artificial neural apparatus and in accordance therewith providing a plurality of fuzzified data;

receiving said plurality of fuzzified data from said first artificial neural apparatus and a first plurality of intra-neural recurrent data with a second artificial neural apparatus and in accordance therewith providing a plurality of membership data corresponding to a plurality of fuzzy logic membership functions;

intra-neurally time-delaying a portion of said plurality of membership data and in accordance therewith providing said first plurality of intra-neural recurrent data; and

receiving said plurality of membership data from said second artificial neural apparatus and a second plurality of intra-neural recurrent data with a third artificial neural apparatus and in accordance therewith providing a plurality of rule data corresponding to a plurality of recurrent fuzzy logic rules.

30. The method of claim 29, wherein said step of receiving said plurality of membership data and a second plurality of recurrent data with a third artificial neural apparatus and in accordance therewith providing a plurality of rule data corresponding to a plurality of recurrent fuzzy logic rules further comprises time-delaying a portion of said plurality of rule data and in accordance therewith providing said second plurality of recurrent data.

31. The method of claim 29, wherein said step of receiving said plurality of membership data and a second plurality of recurrent data with a third artificial neural apparatus and in accordance therewith providing a plurality of rule data corresponding to a plurality of recurrent fuzzy logic rules further comprises providing defuzzified output data in accordance with said plurality of membership data and said second plurality of recurrent data.

32. The method of claim 31, further comprising the step of receiving and comparing reference data and said defuzzified output data and in accordance therewith providing error data corresponding to a difference therebetween.

33. The method of claim 29, further comprising the step of receiving and summing a plurality of data from said third artificial neural apparatus and in accordance therewith providing defuzzified output data.

34. The method of claim 33, further comprising the step of receiving and comparing reference data and said defuzzified output data and in accordance therewith providing error data corresponding to a difference therebetween.