Patent
1991-12-23
1998-06-30
Davis, George B.
395 24, 395 22, 395903, 395906, G06F 1518, G06F 944
Patent
active
057746338
DESCRIPTION:
BRIEF SUMMARY
TECHNICAL FIELD
The present invention relates generally to a method and a system for supporting an operation of a process wherein at least one phenomenon which varies with time is dealt with.
BACKGROUND ART
Conventional methods of operation (or managing) various processes wherein at least one phenomenon which varies with time is dealt with--such as water treatment processes; river information processing processes: meteorological information processing processes; thermal, nuclear and hydraulic power generation processes; cogeneration processes; chemical processes; biological processes, security/exchange information processing processes, and bank management information processing processes--are practiced using formula models which describe these processes.
It is however impossible to convert a process into a formula model unless casualities or causal relationships among a group of variables describing the process have been made clear. On the other hand, when a logical model such as the "if then" rules is employed without using a formula model, application of such a logical model is infeasible unless a causal relationship between causes and the corresponding results have been ascertained. Needless to say, even in the case of a fuzzy method which makes combined use of a formula model and a logical model, its application is impossible unless both the models are described. A judgment and/or an operation (management) has therefore been carried out in the light of precedence or past experiences in such cases, In unusual cases or the like where neither cause nor result is known, an operator has conducted the operation on the basis of the past phenomenological history or his memory. Accordingly, it has been difficult to conduct a good operation all the time.
Further, described generally, these methods have not yet permitted any automated modification of the model structure or elements (rule, etc.). It has hence been difficult to flexibly cope with an actual phenomenon which varies in time.
SUMMARY OF THE INVENTION
An object of the invention is therefore to provide a method and a system for supporting an operation of a process, which can support the operation of the process in a steady state or a non-steady or abnormal state by making effective use of a past history which has not heretofore been used effectively.
Another object of the invention is to provide a method for automatically extracting knowledge such as a causal relationship between a value of an input variable and its corresponding output variable from a learned neural circuit model.
A process operation supporting method according to the invention is a method for supporting an operation of a process, which includes determination of a value of a control variable for a target, to be controlled, in accordance with values of time-dependent input variables so as to bring the target closer to a desired state. The method comprises the steps of providing a neuron circuit model of a hierarchical structure constructed of an input layer, at least one hidden layer and an output layer; causing the neuron circuit model to learn, out of information on a past operation history of the process, a typical pattern of values of input variables at different points in time as input signals and a value of: the control variable, said control value corresponding to the typical pattern, as teacher signal; and inputting, as the values of the input variables, an unlearned pattern to the thus-learned neuron circuit model to determine its corresponding value of the control variable.
The process operation supporting method according to the invention is, in another aspect, a method for supporting an operation of a process, which includes determination of a value of a control variable for at least one target, to be controlled, in accordance with values of time-dependent input variables such that the target can be brought into a desired state. The method comprises the steps of providing a neuron circuit model of a hierarchical structure constructed of an input layer, at least one hid
REFERENCES:
patent: 5023045 (1991-06-01), Watanabe et al.
patent: 5092343 (1992-03-01), Spitzer et al.
patent: 5093792 (1992-03-01), Taki et al.
patent: 5109475 (1992-04-01), Kosak et al.
patent: 5129039 (1992-07-01), Hiraiwa
patent: 5195169 (1993-03-01), Kamiya et al.
Williams et al, "A Class of Gradient-Estimating Algorithms for Reinforcement Learning in Neural Networks" IEEE 1st Inter-Conf-on Neural Networks, Jun. 1987.
Gherrity et al, "A Learning Algorithm for Analog, Fully Recurrent Neural Networks" Jun. 18, 1980.
Rohwer et al, "Training Time-Dependence in Neural Networks" IEEE First Inter. Conf. on Neural Networks, Jun. 1987 Neural Networks Primer AI Expert, Feb. 1989.
Wasserman et al, "Neural Networks, Part 2:" 1988 IEEE Expert.
Widrow et al, "Layered Neural Nets for Pattern Recognition" IEEE Trans. on Acoustics, Speech and Signal Processing, vol. 36, No. 7, Jul. 1988.
"Learning Internal Representations in the Coulomb Energy Network", Scofield, Jul. 1988, pp. 271-276.
Baba Kenji
Enbutsu Ichiro
Hara Naoki
Maruhashi Fumio
Matsumoto Hiroshi
Davis George B.
Hitachi , Ltd.
LandOfFree
Supporting neural network method for process operation does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Supporting neural network method for process operation, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Supporting neural network method for process operation will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1869185