Process control method and system for performing control of a controlled system by use of a neural network

• US 5,311,421 A
• Filed: 12/10/1990
• Issued: 05/10/1994
• Est. Priority Date: 12/08/1989
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Matsumoto, Hiroshi, Saito, Tadayoshi, Miyagaki, Hisanori, Yokokawa, Nobuyuki, Kondoh, Masakazu, Nomura, Masahide, Shimoda, Makoto, Sugano, Akira
• Primary Examiner(s): Smith, Jerry
• Assistant Examiner(s): TRAMMELL, JAMES P
• Application Number:
  US07/625,096
• Time in Patent Office: 1,247 Days
• Field of Search: 364/162, 364/157, 364/148, 364/149, 364/150, 364/151, 364/158, 364/159, 364/160-164, 395/22, 395/23, 395/24, 395/21, 395/81, 395/94, 395/903, 395/904, 395/905, 395/906
• US Class Current: 700/37
• CPC Class Codes: G05B 13/0285 using neural networks and fuzzy logic
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

A method for controlling a controlled system by a controller such that a controlled variable can be brought into conformity with a desired value. With respect to at least one of input/output variables for a combined controlling-controlled system, which includes in combination the controller and the controlled system, and input/output variables for the controlled system, information containing its characteristics is taken out from the combined controlling-controlled system. The information with the characteristics contained therein is inputted to a neural network which has been caused beforehand to learn a correlation between the information containing the characteristics and control parameters. From the neural network, one or more of the control parameters, said one or more control parameters corresponding to a corresponding number of inputs to the neural network, are outputted to the controller.

Claims

1. A method for controlling a controlled system by a controller so as to bring a controlled variable into conformity with a desired value, said method comprising the steps of:

receiving information which contains characteristics of at least one of input/output variables for a combined controlling-controlled system, said combined controlling-controlled system comprising in combination a controller and a controlled system, and input/output variables for the controlled system;

inputting the information with the characteristics contained therein to a neural network for learning in advance a correlation between the information containing the characteristics and a control parameter and determining a control parameter for the controller;

tuning the control parameter based on said input/output variables for the controlled system; and

outputting a tuned control parameter to the controller.

2. The control method of claim 1, wherein the information containing the characteristics of the input/output variable is obtained by the determination of a time response on at least one characteristic of each of models of the controller and controlled system through simulation making use of the models;

and the neural network is caused to learn by obtaining a control parameter capable of providing optimal control results conforming with the characteristic of each of the models and then using the information, which contains the characteristics of the input/output variable, as learning input data and the control parameter as learning teacher data.

3. The control method of claim 1, wherein the information containing the characteristics of the input/output variable is obtained from time responses of good characteristics out of time responses from the controlled system and is then stored;

and the neural network is caused to learn by storing control parameters for the controller, said control parameters corresponding to the time responses of the good characteristics, and using the information as learning input data and the parameters as learning teacher data.

4. A control system having a controller for controlling a controlled system such that a controlled variable is brought into conformity with a desired value, comprising:

a tuning system for the controller, wherein said tuning system comprises;

a neural network having a plurality of mutually-connectable units and capable of obtaining an output signal corresponding to an input signal in accordance with the state of connection among the units,

a means for receiving information which contains characteristics of at least one of input/output variables for a combined controlling-controlled system, comprising in combination the controller and the controlled system, and input/output variables for the controlled system, and

a means for inputting the information with the characteristics contained therein to the neural network to output a control parameter for the controller; and

a parameter tuning system for tuning the control parameter based on said input/output variables and outputting a tuned control parameter to the controller.

5. The control system of claim 4, wherein the neural network has beforehand stored the state of connection among the units so that the control parameter can be obtained as an output signal upon when the information with the characteristics contained therein is inputted.

6. The control system of claim 4, further comprising:

a tuning rule learning system for causing the neural network, which is adapted to tune the control parameter of the controller, to learn a tuning rule, wherein said tuning rule learning system comprises;

a means for obtaining the information, which contains the characteristics of the input/output variable, by developing models of the controller and controlled system and then determining a time response on at least one characteristic of each of the models;

a means for determining a control parameter capable of providing optimal control results conforming with the characteristic of each of the models; and

a means for causing the neural network to learn by using the information with the characteristics of the input/output variable and the control parameter as learning input data and learning teacher data, respectively.

7. The control system of claim 6, wherein the parameter tuning system performs on-line tuning for the controller and the tuning rule learning system performs on-line setting of the tuning rule for the parameter tuning system.

8. The control system of claim 4, wherein the information-receiving means has a function such that the information-receiving means can receive and store information which contains characteristics on an actual input/output variable of at least one of the combined controlling-controlled system and the controlled system.

9. The control system of claim 8, further comprising:

a means for storing, as learning teacher data, control parameters corresponding to time responses of good characteristics received upon tuning of the controller; and

a means for causing the neural network to learn by using the learning teacher data and data out of the information with the characteristics on the actual input/output variable, the latter data corresponding to the learning teacher data.

10. The control system of claim 4, wherein the controlled system is a plant having a boiler for burning fuel to generate steam;

the controller controls the feed amount of the fuel to the boiler to generate steam of a desired temperature, and the controller performs control computation in accordance with a preset control parameter with respect to the difference between the desired temperature of the steam and an actual steam temperature (controlled variable), thereby outputting a fuel feed quantity demand (manipulated variable);

the control parameter tuning system estimates the control parameter by using information, which contains characteristics of the steam temperature and fuel feed quantity demand, and sets the thus-estimated control parameter in the controller; and

the boiler has a means for controlling the feed quantity of the fuel on the basis of the fuel feed quantity demand.

11. The control system of claim 10, further comprising a tuning rule learning system for determining a turning rule for the control parameter tuning system, said learning system having a neural network.

12. The control system having a controller for controlling a controlled system such that a controlled variable is brought into conformity with a desired value, comprising:

a tuning system for the controller, wherein said tuning system comprises;

a neural network having a plurality of mutually-connectable units and capable of obtaining an output signal corresponding to an input signal in accordance with the state of connection among the units,

means for receiving information which contains characteristics of at least one of input/output variables for a combined controlling-controlled system, said combined controlling-controlled system comprising in combination a controller and a controlled system, and input/output variables for the controlled system, and

means for inputting the information with the characteristics contained therein to the neural network to output a control parameter for the controller;

a tuning rule learning system for causing the neural network, which is adapted to tune the control parameter of the controller, to learn a tuning rule, wherein said tuning rule learning system comprises;

a second neural network different form the first-mentioned neural network,

means for developing models of the controller and controlled system,

means for determining a time response on at least one characteristic of each of the models to obtain the information with the characteristics of the input/output variable contained therein,

means for determining a control parameter capable of providing optimal control results conforming with the characteristics of each of the models, and

means for causing the second neural network to learn by using the information with the characteristics of the input/output variable and the control parameter as learning input data and learning teacher data, respectively; and

wherein the results of the learning by the second neural network are transferred to the first neural network and are used therein.

13. A method for controlling a controlled system by a controller so as to bring a controlled variable into conformity with a desired value, comprising the steps of:

receiving information which contains characteristics of at least one of input/output variables for the controlled system;

inputting the information to a neural network;

identifying a model of the controlled system by using a learning function of the neural network; and

determining a manipulated variable by using the model, whereby the controlled system is controlled;

wherein the neural network is caused to learn by using time-series signals of an output variable and time-series signals of an input variable as the information with the characteristics of the input/output variable of the controlled system, the time-series signals of the output variable as learning input data and the time-series signals of the input variable as learning teacher data, whereby a reverse system model of the controlled system is identified;

wherein an output variable of a reference model of a control system, which comprises in combination the controlled system and a controller for controlling the controlled system is inputted to the neural network with the reverse system model identified therein and the controller is tuned to bring time-series signals of an output variable of the controller into conformity with the time-series signals of the input variable of the controlled system;

wherein an additional neural network different from the neural network with the reverse system model identified therein is used as the controller.

14. A method for predictively controlling a controlled system to bring a controlled variable into conformity with a desired value, said method comprising the steps of:

receiving input/output variables of said controlled system as input signals for a neural network, to make the neural network learn the output variable of the controlled system as teacher data, so that a model is identified which is to output an estimated value of the output variable of the controlled system in the near future when the signals of the input/output variables of the controlled system are inputted;

inputting the input/output variables of the controlled system at the present time into said model, to make said model predict and output the estimated value of the output variable of the controlled system in the near future;

determining one or more manipulated variables by tuning said estimated value based on the input/output variables of the controlled system and said desired value; and

outputting said one or more manipulated variables to the controlled system.

15. A method according to claim 14, wherein the input/output variables of the controlled system include the manipulated variables and the controlled variable.

16. A method according to claim 14, wherein operation data of the controlled system is stored and the manipulated variables and the controlled variable contained in said operation data stored in the past are inputted into the neural network for learning as the input/output variables and the output variable, thereby identifying said model.

17. A method according to claim 14, wherein operation data of the controlled system at present time is received and the manipulated variables and the controlled variable contained in said operation data are inputted to the neural network model for learning as the input/output variables and the output variable, thereby identifying said model.

18. A method according to claim 14, wherein a signal indicating a load level of the controlled system is used as the input/output variable of the controlled system.

19. A method according to claim 14, wherein time-series signals are used as the input/output variables of the controlled system.

20. A method for predictively controlling a controlled system having nonlinear characteristics, to bring a controlled variable into conformity with a desired value, said method comprising the steps of:

receiving input/output variables of the controlled system as input signals for a neural network, to make the neural network learn the output variable of the controlled system as teacher data, so that a nonlinear model is identified which is to output the output variable of the controlled system in the near future wherein signals of the input/output variables of the controlled system are inputted;

developing a linear model using said nonlinear model; and

inputting the input/output variables of the controlled system at the present time into said linear model, to make said model predict and output an estimated value of the output variable of the controlled system in the near future, and determining one or more manipulated variables based upon said estimated value and said desired value thereby controlling the controller system.

21. A method according to claim 20, wherein said linear model is developed such that the input variable to the nonlinear model is time-changed by signals corresponding to a disturbance with respect to the controlled system, so that said linear model is identified at a load level based on the time change of the output variable and the time change of the input variable at that time.

22. A control system having a predictive control system for predictively controlling a controlled system by determining one or more manipulated variables based on a deviation between a controlled variable and a desired variable, said control system comprising:

a model identification system having a neural network, which receives the manipulated variables and the controlled variable of the controlled system as input signals for a neural network, to make the neural network learn the controlled variable of the controlled system as teacher data, so that a model is identified which is to output an estimated value of the controlled variable of the controlled system in the near future, when signals of the input/output variable of the controlled system are inputted; and

a controlled variable predictive system which inputs the manipulated variables and the controlled variable of the controlled system at the present time, into the neural network which identifies the model, to predict the controlled variable of the controlled system in the near future to output the estimated value of the controlled variable.

23. A control system having a predictive control system for predictively controlling a controlled system having nonlinear characteristics, by determining one or more manipulated variables based on a deviation between a controlled variable and desired variable, said control system comprising:

a nonlinear model identification system having a neural network, which receives the input/output variable of the controlled system as input signal for a neural network, to make the neural network learn the output variable of the controlled system as teacher data, so that a nonlinear model is identified which is to output the output variable of the controlled system when signals of the input/output variables of the controlled system are inputted;

a linear model identification system for developing a linear model using said nonlinear model; and

a controlled variable predictive system which receives the manipulated variables and the controlled variable of the controlled system at the present time using the neural network which identifies the model, to predict the controlled variable of the controlled system in the near future to output an estimated value of the controlled variable.