Data processing: artificial intelligence – Plural processing systems
Patent
1996-01-16
1998-09-15
Davis, George B.
Data processing: artificial intelligence
Plural processing systems
706 3, 706 52, 706 16, G06F 1518
Patent
active
058094873
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
The present invention relates to an arrangement for modeling a non-linear process having at least one input variable and at least one output variable, comprising a neural network.
As is known, for example, from the book by Eberhard Schbneburg, Nikolaus published in 1990 by Markt Technik Publishers, neural networks can be used to simulate complex, non-linear functions having many input and output variables by training them on the basis of learning data. This ability can be utilized to model the static, non-linear behavior of the process on the basis of measuring data acquired from a process. In this manner, one obtains a process model which is realized by the non-linear output function of the neural network and can be used to improve the process control in a closed-loop control system. However, when acquiring measuring data from the real process, the problem occurs that generally not all possible process states can be run through. The output data and, thus, also the function learned by the neural network are, therefore, also only valid in those operating states for which ample measuring data are available. If new process states occur during the current operation, then unforeseeable interventions in the process can lead to critical or even unacceptable states when the neural network is used for the closed-loop or open-loop control of the process.
SUMMARY OF THE INVENTION
The present invention creates an arrangement which will avoid the mentioned disadvantages when a neural network is used to model a non-linear process.
The present invention provides a new arrangement for modeling a nonlinear process having at least one input variable (x1, x2) and at least one output variable (y). The arrangement for modeling the non-linear process includes a neural network (1), whose function is determined in a first part of the domain of input variables (x1, x2) by learning from measuring data, which are obtained from the process by acquiring measured values. The arrangement additionally includes a device for specifying functional (operational) values in a second part of the domain of input variables (x1, x2) in which there are no measuring data for training the neural network (1).
In an alternative embodiment of the present invention, in the second part of the domain of input variables, the device for specifying functional values includes a device for monitoring the input variables (x1, x2) when they leave the first part of the domain of input variables. In this embodiment, the device for specifying functional values additionally includes a controllable gate (9) which is controlled by the monitoring device. In response to input variables in the second part of the domain, the controllable gate switches over to the specification (setpoint entry) of functional values.
In another embodiment of the present invention, in the second part, the device for specifying functional values is a fuzzy system. Membership of the values of the input variables (x1, x2) in the first part is characterized by membership functions (14, 15) and, in response to input variables in the second part, the values of the output variables (y) are specified in the manner of fuzzy logic.
In a further embodiment of the present invention, the membership functions (14, 15) are determined by the rate of occurrence of the values of the input variables (x1, x2) when measured values are acquired.
In an additional embodiment of the present invention, in the second part of the domain of the input variables (x1, x2) and in the domain of the output variables (y), other membership functions (16, 17, 18, 19, 20, 21, 22, 23) are defined for linguistic values. In the controlling mechanism (action), the behavior of the process is simulated empirically by IF/THEN rules on the basis of the linguistic values.
An additional embodiment of the present invention provides that the device for specifying functional values in the second part is a fuzzy system. The process behavior is simulated empirically in the second part of the domain of the input va
REFERENCES:
patent: 5046019 (1991-09-01), Basehore
patent: 5255344 (1993-10-01), Takagi et al.
patent: 5465320 (1995-11-01), Enbutsu et al.
Proceedings of the 1992 American Control Conference, vol. 1, Jun. 1992, Chicago, US, pp. 475-479, M.A. Kramer et al.: Embedding Theoretical Models in Neural Networks.
Second IEEE International Conference on Fuzzy Systems, vol. 1, Apr. 1993, San Francisco, US, pp. 321-326, B. Freisleben et al.: Combining Fuzzy Logic and Neural networks to Control an Autonomous Vehicle.
Proceedings of the 1992 American Control Conference, vol. 3, Jun. 1992, Chicago, US, pp. 1917-1921, D.C. Psichogios et al.: Process Modeling Using Structured Neural Networks.
Neural Networks, vol. 6, No. 4, 1993, Elmsford, US, pp. 485-497, H. Gomi et al.: Recognition of Manipulated Objects by Motor Learning with Modular Architecture Networks.
Neuronale Netzwerke, Eberhard Schoneburg et al., published by Markt & Technik Verlag, 1990.
Dash et al, "A Fuzzy Adaptive Correction Scheme of Short Term Load Forecasting Using Fuzzy Layered Neural Network" Applications of Neural Networks to Power Systems, 1993 Forum, IEEE 1993.
Ikonomopoulos et al, "Measurement of Fuzzy Values Using Artificial Neural Networks and Fuzzy Arithmetic", IEEE International Conference on Fuzzy Systems, Mar.-Apr. 1993.
Wong FS, "Fuzzy Neural Systems for Decision Making", IEEE International Conference on Neural Network, 1991.
Davis George B.
Siemens Aktiengesellschaft
LandOfFree
Arrangement for modeling a non-linear process does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Arrangement for modeling a non-linear process, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Arrangement for modeling a non-linear process will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-104287