JPH06162233A - Chaos controller and learning method thereof - Google Patents

Abstract

PURPOSE:To provide the chaos controller and learning method thereof, which can execute optimal control of a control system in which plural control elements influence complicatedly each other. CONSTITUTION:Prior to an image forming operation of a document image of an image forming part 1, a test chart is placed on contact glass, image data VD obtained by reading its image by a CCD 2 is stored in an image memory 15, and also, and when a value of recording density data PD obtained by reading the image on transfer paper on which an image is formed in accordance with this image data VD, and differential data DF outputted from a first comparator 16 for calculating a difference of the image data VD are smaller than differential data DF' stored in a differential memory 14, the differential data DF' stored in the differential memory 14 is replaced with the differential data DF outputted from a first comparator 16, and also, with respect to a coupling coefficient Wk stored in a coefficient memory 13, a learning processing operation of a chaos N/N 12 for storing the prescribed number of sets [Wk] of the coupling coefficient Wk used for determining a control value CD of each control part in the coefficient memory 13 is executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is capable of detecting a state quantity of a controlled system in which the relationship between an input signal and an output signal cannot be specified and controlling the output signal output by the controlled system to an optimum value. The present invention relates to a chaos control device and a learning method thereof.

[0002]

2. Description of the Related Art For example, in a copying machine for reading an original image and recording the image on a transferred transfer sheet, in order to faithfully reproduce the recorded original image recorded on the transfer sheet, If one parameter is changed in an attempt to adjust the charging potential, the exposure amount, the toner amount, the color conversion coefficient, and the like, the influence on other control elements, and these parameters are sequentially adjusted to optimum values. It wasn't easy. Therefore, as one of the optimum control methods for a control system in which the influence of such a variation in one parameter affects other control elements, studies have been made to incorporate a neural network into the control system.

[0003]

A control system using a neural network enables optimum control corresponding to input information of a plurality of control elements that affect each other in a complicated manner by a relatively small-scale control device. A control result above a certain level could not be obtained depending on the learning information provided when the neural network was made to learn. The present invention is intended to solve such problems in the prior art,
An object of the present invention is to provide a chaos control device capable of optimal control of a control system in which a plurality of control elements affect intricately, and a learning method thereof.

[0004]

According to the present invention, in order to solve the above-mentioned problems, a chaos control device learns based on a preset output signal and a plurality of coupling coefficients, and according to the learning result each time it learns. A neural network that outputs a coupling coefficient, detects a state quantity of a controlled system in which a relationship between an input signal and an output signal cannot be specified, and outputs a controlled variable for controlling the output signal to the controlled system, and the neural network And a coupling coefficient storage means for storing the coupling coefficient in a rewritable state. The learning method of the chaos control device is to use a preset output signal and a plurality of coupling coefficients. Based on the learning, the coupling coefficient corresponding to the learning result is output every time learning is performed, and the relationship between the input signal and the output signal cannot be specified. A process of detecting a state quantity of a control system and storing a plurality of coupling coefficients of a neural network which outputs a control amount for controlling the output signal to the controlled system, and chaotically recalling the plurality of coupling coefficients. , A process of using an output vector in a chaotic neural network as a coupling coefficient in a neural network, a process of storing the coupling coefficient, and a process of calculating a difference between an output signal output from a controlled system and a preset output signal Based on the result of comparing the differences, the process of sequentially storing the differences each time learning is performed, the process of comparing the already-stored differences with the differences obtained by newly calculating, and the already stored This includes the process of instructing the rewriting of the coupling coefficient.

[0005]

In the chaos control device, the neural network learns in the learning mode based on the preset output signal and the plurality of coupling coefficients, and outputs the coupling coefficient according to the learning result every learning. In the control mode, the state quantity of the controlled system in which the relationship between the input signal and the output signal cannot be specified is detected, and the controlled quantity for controlling the output signal is output to the controlled system. The chaotic neural network generates the coupling coefficient of the neural network. The coupling coefficient storage means stores the coupling coefficient in a rewritable state.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing an outline of a control system according to an embodiment in which the present invention is applied to a digital copying machine, and FIG. 2 is a schematic diagram showing a configuration of various sensors and a control section of an image forming section of the digital copying machine. is there. As shown in FIG. 2, even in the present embodiment, each control unit and various sensors for controlling the electrostatic photography process of the image forming unit 1 included in the general digital copying machine are provided, and the original image is faithfully reproduced. It is controlled so that it can be reproduced. The outline of the known image forming process will be described. The original M pressed against the contact glass 22 by the pressure plate 23 is illuminated by the illumination lamp 20 that scans along the contact glass 22, and the reflected light is solid-state imaged by the mirror group 21. Guided to the device (CCD) 2, C
It is photoelectrically converted by the CD 2 and converted into an electric signal, and further converted into image digital data by the A / D converter 3. Various types of image processing are performed on the image digital data by the image processing unit 4, and the image digital data is sent to the exposure control unit 41. The exposure control unit 41 controls the light emission of a laser diode (LD) (not shown) according to the sent image data, and the laser light emitted from the LD is charged to a high voltage in advance by the charger 24 and is rotated on the photosensitive drum 7. A latent image corresponding to the original image is formed on. The latent image formed on the photosensitive drum 7 is eventually developed by the developing device 6 into a toner image, conveyed to a transfer position, where it is fed from the paper feed unit 5 and corona charged by the transfer separation unit 8. It is transferred onto the transfer paper. The transfer sheet on which the image has been transferred is separated from the photosensitive drum 7 by the separation and charging of the transfer separation section 8 and is heated and pressed by the fixing section 9 so that the toner image is fixed and the copy is recorded as an image outside the apparatus. Is discharged to. The image data amount detector 30 detects the data amount of the image data output from the A / D converter 3, and the image detector 31 reads the image density of the transfer paper on which the toner image is formed to obtain the recording density data PD. Send out. The temperature sensor 32 and the humidity sensor 33 respectively detect the temperature and humidity around the photosensitive drum 7 and send the information to the CPU 10. Similarly, the surface potential meter 34 detects the potential of the surface of the photosensitive drum 7, and the toner concentration detector 35 detects the concentration of the toner stored in the developing device 6, and the potential data and the concentration data are respectively sent to the CPU 10. Send out. The CPU 10 determines the optimum value of the surface potential of the photosensitive drum 7, the exposure amount of the LD, and the density of the toner in the developing device 6 according to the detection information, and charges the charge via the charging control unit 40 according to these values. The developing device 24 controls the exposure amount of the LD through the exposure control unit 41, the developing bias voltage of the developing device 6 through the developing bias control unit 42, and the toner amount in the developing device 6 through the toner replenishment control unit 43. In addition, the transfer voltage of the transfer separation unit 8 is controlled via the transfer voltage control unit 44.

The neural network element (N / N) 11 for giving control data of the control system of the image forming unit 1 is a CPU 10
When a command signal for starting a copying operation is received from the control unit, the control data of each control unit of the image forming unit 1 is received based on the control information stored by learning by receiving the detection data DD sent from the various sensors described above. The control value CD of is output. FIG. 3 is a schematic diagram conceptually showing the function of the N / N 11. As shown in FIG. 3, the N / N 11 receives various detection data DD output from the image data amount detector 30, the temperature sensor 32, the humidity sensor 33, the toner concentration detector 35, and the surface potential meter 34, The detected data DD are input to the neuron elements N _ij arranged in four stages, data conversion is performed according to the coupling coefficient W _k of each neuron element N _ij , and the charging controller 40, the exposure controller 41, the developing bias controller 42. The control data CD of each control unit of the toner replenishment control unit 43 and the transfer voltage control unit 44 is output. The coupling coefficient W _k is stored in the coefficient memory 13, and the chaotic neural network element (chaos N /
N) 12 reads the coupling coefficient W _k from the coefficient memory 13 and outputs to the N / N 11 a set {W _k } of coupling coefficients W _k suitable for the various detected data DD that has been input. N / N1
1 outputs the control value CD of each control data of each control unit of the image forming unit 1 based on the set {W _k } of the coupling coefficient W _k output from the chaos N / N 12, and each of the image forming units 1 The control unit controls each controlled object according to the given control value CD to execute the electrostatographic process. The recording density data PD obtained by reading the image on the transfer paper with the image detector 31 is used as the first comparator 16
The image on the transfer paper is read in advance and compared with the image data VD corresponding to one original document stored in the image memory 15, and the difference is calculated and output as difference data DF. The difference data DF is the second comparator 1
7, the difference data DF ′ previously stored in the difference memory 14 is compared, and the comparison output is the CPU 1
It is output to 0. When the difference data DF output from the first comparator 16 is smaller than the difference data DF ′ stored in the difference memory 14, the CPU 10 outputs the difference data DF ′ stored in the difference memory 14 from the first comparator 16. The combined coefficient W _k stored in the coefficient memory 13 is replaced with the output difference data DF, and the image detector 31
Replaces the set {W _k } of the coupling coefficient W _k used for determining the control value CD of each control unit of the image forming unit 1 when the recording image of the transfer paper is formed by reading the recording density data PD.
On the other hand, when the difference data DF output from the first comparator 16 is equal to or greater than the difference data DF ′ stored in the difference memory 14, the difference data D stored in the difference memory 14
F'and the coupling coefficient W stored in the coefficient memory 13
_The stored data is retained without updating _k . Before the image forming operation of the document image by the image forming unit 1, the learning processing operation of chaos N / N 12 is performed.
First, a test chart is placed on the contact glass 22, and the image is read by the CCD 2 to obtain image data VD.
Is stored in the image memory 15 and the image data VD
The difference data DF between the value of the recording density data PD and the image data VD obtained by reading the image of the transfer paper image-formed according to
The constant value set {W _k } of the coupling coefficient W _k used for determining the control value CD of each control unit when is less than a predetermined value is stored in the coefficient memory 13. Then, each time the above-described learning processing operation is repeated, a more appropriate set {W _k } of coupling coefficients W _k is stored in the coefficient memory 13. FIG. 4 is a block diagram showing the actual hardware configuration of the control system of the image forming unit 1. N / N11, chaos N / N12, setting unit 1
9 and coefficient memory 13, difference memory 14, storage device 18 including image memory 15 is connected to CPU 1 via a data bus.
0 is connected for communication. The setting unit 19 changes or sets the storage contents stored in the storage device 18 by operating a key on an operation panel (not shown).

Here, N / N11 and chaos N / N1
2 will be described in more detail. FIG. 5 is a conceptual diagram showing a generalized relationship between N / N 11 and chaos N / N 12.
The controlled object 1 ′ receives the input signal In and outputs the output signal Ou.
Output t. State quantity o from controlled object 1'to N / N 11
1 is input and the controlled variable i is input from the N / N 11 to the controlled object 1 ′.
1 is output. First, a set of constants {W of the coupling coefficient W _k of N / N11 obtained experimentally by the learning processing operation {W
_k } is given and stored in the memory. The chaos N / N 12 reads out the coupling coefficient W _k from the memory and recalls the new coupling coefficient W _k at any time according to chaotic behavior. Output signal Ou
The difference d (t) between t and the output signal Out ′ one stage before is 1
Compared with the difference d (t-1) before the stage, d (t) <d
At (t-1), the coupling coefficient _Wk is updated.

FIG. 6 shows N / N11 and chaos N / N12.
It is a schematic diagram showing the model structure of. That is, N / N11
Has a layered model structure as shown in (b), and chaos N / N 12 has a meshed model structure as shown in (a). FIG. 7 shows N / N 11 and chaos N / N.
12 is a schematic diagram showing a functional model of a neuron element _nij forming 12; FIG. The neuron element n _ij multiplies the input information x _i by the coupling coefficient w _k and weights it, and sums them.
= Σx _i w _i The converted output of the response function Y = f (X) is output. As the response function Y, for example, a sigmoid function represented by Y = 1 / [1+ EXP (−X)] as shown in FIG. 8 can be used. The N / N 11 can also be configured by software. By the learning processing operation described above, the input signal In and the output signal O of the controlled object 1 '
The values of the coupling coefficient W _k such that ut match are determined. Even if the response function Y = f (X) has not been clarified, the output signal Out that accurately responds to the new input signal In is output through the learning processing operation. As a learning method, the output side coupling coefficient W
_A backpropagation method that sequentially corrects from _k can be used. The learning processing operation may be performed not only when the apparatus is started up but also intermittently when the apparatus is in operation. FIG. 9 shows an example of input / output data when the conversion process is performed after the learning process in the present embodiment. In this example, when the input data shown in (1) to (3) is input,
When the input data shown in (4) is input by performing the learning process of storing the coupling coefficient W _k in the coefficient memory 13 such that the output data shown in (1) to (3) is output, The output data shown in () will be output.
In the chaos N / N 12, when the known pattern data is stored, the sum of the correlation matrix that outputs the desired output pattern data is set as the coupling coefficient w _k . N / N11
In the control system using only the output signal Out gradually converges to a constant value by the learning processing operation, whereas in the control system using the chaotic N / N 12, the output signal Out converges to a constant value by the learning processing operation. There is nothing to do. Therefore, in the present embodiment, even when the optimum value is not near the convergent value, the optimum value can be reached soon. That is, the chaos N / N 12 outputs the different chaotically varying coupling coefficients w _k infinitely one after another, depending on the stored pattern data.

[0010]

As described above, according to the first aspect of the invention, learning is performed on the basis of a preset output signal and a plurality of coupling coefficients, and each time learning is performed, a coupling coefficient corresponding to the learning result is output. In addition, the state quantity of the controlled system in which the relationship between the input signal and the output signal cannot be specified is detected, and the neural network for outputting the controlled quantity for controlling the output signal to the controlled system and the coupling coefficient of the neural network are Since it has a chaotic neural network to generate and stores the coupling coefficient in a rewritable state,
For example, post-processing such as comparison with a result learned later can be performed, and the learning function of the neural network can be complemented. According to the invention described in claim 2, since the chaotic neural network stores the coupling coefficient that causes the chaotic behavior in advance, the coupling coefficient of the neural network that causes the controlled system to output an accurate output signal is generated. be able to. According to the third aspect of the present invention, the difference between the output signal output by the controlled system and the output signal already output by the controlled system is calculated, and each time the learning is performed, the difference is sequentially stored, and the stored difference is newly stored. Since the calculated differences are compared and the rewriting instruction signal for instructing the rewriting of the stored coupling coefficient is output based on the comparison result, it is possible to control the neural network without adding a complicated control function to the controlled system. The optimum coupling coefficient can be set flexibly. According to the invention described in claim 4, since the start of operation of the controlled system can be detected and the end of learning can be instructed to the neural network, it is possible to omit unnecessary learning processing operations, so that the apparatus is efficient. It can be operated well. According to the fifth aspect of the present invention, after the learning end instructing means gives an instruction to end the learning, the learning is performed based on the preset output signal and the plurality of coupling coefficients, and each time the learning is performed, the learning result is responded to. In addition to outputting the coupling coefficient, the state quantity of the controlled system is detected only by the neural network, and the control quantity for controlling the output signal is output to the controlled system. The versatility of is improved. According to the sixth aspect of the present invention, it is determined whether the controlled system is operating, and when it is determined that the controlled system is operating, the neural network is instructed to start learning every predetermined time. It is possible to follow changes with time and deterioration with time. According to the invention of claim 7, the difference already stored and the newly calculated difference are compared, and as a result, if the already stored difference is smaller than the newly obtained difference, If the stored coupling coefficient is used as a coupling coefficient in the neural network without being rewritten and the already stored difference is not smaller than the newly obtained difference,
Since the stored coupling coefficient is rewritten to the difference obtained by a new calculation to instruct to use it as the coupling coefficient in the neural network, the controlled system can be used without using complicated software logic. Even if is changed, the control can be flexibly performed, and further, the coupling coefficient of the neural network can be rewritten to make it optimal. According to the invention described in claim 8, the image data obtained by reading the original is used as an input signal, and the image data obtained by reading the recorded image recorded on the transfer paper is used as an output signal, and the state of the electrostatic photography process is determined. The control system for controlling the electrostatographic process of the image forming apparatus equipped with the photoconductor, which uses the detection value of the detecting means for detection as the state quantity, is the controlled system, so that the optimum coupling coefficient for the electrostatographic process control is obtained. You can ask.
According to the ninth aspect of the invention, the neural network is formed by forming a plurality of neurons in a mesh shape while being related by the coupling coefficient, and the chaotic neural network is provided with a plurality of neurons to which a chaotically varying function is given. The image forming apparatus forms a latent image of an original image by exposing a pre-charged photoconductor on the basis of image data obtained by reading an original and converting it into image data while being related to each other by a coupling coefficient. Then, the latent image formed on the photoconductor is developed with toner and converted into a visible image, the developed image is transferred to the fed transfer paper, and the amount of black pixels in the image data obtained by reading the original is determined. The surface potential of the photoconductor and the humidity and temperature in the vicinity thereof are detected, and the toner concentration of the developing unit is detected, and the charge amount of the photoconductor, the exposure amount of the exposing unit, the photoconductor and the developing unit are detected. The bias voltage is controlled to read the visible image transferred to the transfer paper and convert it into image data. Furthermore, when the newly calculated difference is smaller than the previously stored difference, the coupling coefficient of the already stored Since the rewriting is instructed, the electrostatic photography process of the image forming apparatus can be accurately controlled. According to the invention as set forth in claim 10, learning is performed based on a preset output signal and a plurality of coupling coefficients, and each time learning is performed, a coupling coefficient corresponding to the learning result is output, and the state quantity of the controlled system is calculated. The plurality of coupling coefficients of the neural network that detects and outputs the control amount for controlling the output signal to the controlled system is stored, and the plurality of coupling coefficients are chaotically recalled, and the output vector of the chaotic neural network is stored. Is used as the coupling coefficient in the neural network, the coupling coefficient is stored, the difference between the output signal output by the controlled system and the preset output signal is calculated, and the difference is sequentially stored each time learning is performed, and the difference is stored in advance. Based on the result of comparing the difference obtained by the new calculation with the difference obtained, the process of instructing the rewriting of the coupling coefficient already stored is included. Even the control system is changed without the use of miscellaneous software logic can be flexibly controlled. According to the invention of claim 11,
After the learning of the neural network is judged to be completed, and after the learning is completed, if the difference stored is smaller than the difference newly calculated, the already stored coupling coefficient is changed. Since each process used as the coupling coefficient in the neural network is included without rewriting, it is possible to flexibly control even if the controlled system changes without using complicated software logic. Can be rewritten to be optimal. According to the twelfth aspect of the present invention, learning is performed based on a preset output signal and a plurality of coupling coefficients, a coupling coefficient corresponding to the learning result is output each time the learning is performed, and the state quantity of the controlled system is calculated. Detect and
A neural network that outputs a control amount that controls the output signal to the controlled system, a chaotic neural network that stores in advance a coupling coefficient that causes a chaotic behavior of the neural network, and generates a coupling coefficient of the neural network; In a chaotic controller that stores coefficients in a rewritable state, a plurality of coupling coefficients of a neural network are stored, the plurality of coupling coefficients are chaotically recalled, and an output vector of the chaotic neural network is It is used as a coupling coefficient, the coupling coefficient is stored, the difference between the output signal output by the controlled system and the preset output signal is calculated, and the difference is sequentially stored each time learning is performed, and the difference already stored Is already stored based on the result of comparing the differences obtained by Since it includes each process for instructing the rewriting of the coupling coefficient, it is possible to flexibly control even if the controlled system changes without using complicated software logic. Can be something.

[Brief description of drawings]

FIG. 1 is a block diagram showing an outline of a control system of a digital copying machine according to an embodiment of the present invention.

FIG. 2 is a schematic diagram showing a configuration of various sensors and a control unit of an image forming unit of a digital copying machine.

FIG. 3 is a schematic diagram conceptually showing the function of a neural network.

FIG. 4 is a block diagram showing an actual hardware configuration of a control system of an image forming unit.

FIG. 5 is a conceptual diagram showing a generalized relationship between a neural network and a chaotic neural network.

FIG. 6 is a schematic diagram showing model structures of a neural network and a chaotic neural network.

FIG. 7 is a schematic diagram showing a functional model of a neuron element.

FIG. 8 is a graph showing a sigmoid function.

FIG. 9 is a table showing an example of input / output data when performing learning processing and conversion processing.

[Explanation of symbols]

 2 CCD 6 developing device 7 photoconductor drum 8 transfer separation unit 10 CPU 11 neural network element (N / N) 12 chaotic neural network (chaos N / N) 13 coefficient memory 14 difference memory 15 image memory 16 first comparator 17th 2 Comparator 24 Charger 30 Image Data Amount Detector 31 Image Detector 32 Temperature Sensor 33 Humidity Sensor 34 Surface Potential Meter 35 Toner Density Detector 40 Charge Control Section 41 Exposure Control Section 42 Development Bias Control Section 43 Toner Supply Control Section 44 Transfer voltage controller

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Akira Kojima 1-3-6 Nakamagome, Ota-ku, Tokyo Ricoh Co., Ltd. (72) Konosuke Maruyama 1-3-6 Nakamagome, Ota-ku, Tokyo Shares Inside Ricoh Company (72) Inventor Hironori Suzuki 1-3-6 Nakamagome, Ota-ku, Tokyo Inside Ricoh Company, Ltd.

Claims (12)

[Claims] 1. A controlled device in which learning is performed based on a preset output signal and a plurality of coupling coefficients, a coupling coefficient corresponding to the learning result is output each time learning is performed, and a relationship between an input signal and an output signal cannot be specified. A neural network that detects a state quantity of the system and outputs a controlled variable that controls the output signal to the controlled system, a chaotic neural network that generates a coupling coefficient of the neural network, and a coupling coefficient that can be rewritten. A chaos control device having a coupling coefficient storage means for storing. 2. The chaotic control apparatus according to claim 1, wherein the chaotic neural network stores in advance a coupling coefficient that causes chaotic behavior. 3. Output signal storage means for storing an output signal output from the controlled system, and difference calculation means for calculating a difference between the output signal output from the controlled system and the output signal read from the output signal storage means. A difference storage means for sequentially storing a difference each time learning is performed, a difference stored in the difference storage means and a difference calculated by the difference calculation means are compared, and stored in the coupling coefficient storage means based on the comparison result. The chaos control device according to claim 1, further comprising a first rewriting instruction means for outputting a rewriting instruction signal for instructing rewriting of the coupling coefficient. 4. The start of operation of the controlled system is detected, and
4. The chaos control device according to claim 3, further comprising learning end instruction means for instructing the neural network to end learning. 5. After learning is instructed by the learning end instruction means, learning is performed based on a preset output signal and a plurality of coupling coefficients, and each time learning is performed, a coupling coefficient corresponding to the learning result is output. At the same time, a control instruction means for detecting the state quantity of the controlled system in which the relationship between the input signal and the output signal cannot be specified only by the neural network and for instructing the controlled system to output the controlled quantity for controlling the output signal is provided. The chaos control device according to claim 4, which is provided. 6. The chaos control according to claim 3, further comprising a learning start instruction means for instructing the neural network to start learning every predetermined time when it is determined whether the controlled system is operating or not. apparatus. 7. The difference stored in the difference storage means is compared with the difference calculated by the difference calculation means, and as a result, the difference already stored in the difference storage means is smaller than the difference newly calculated. In this case, the previously stored coupling coefficient is used as the coupling coefficient in the neural network without being rewritten, and the difference already stored in the difference storage means is not smaller than the newly obtained difference. In this case, the second rewriting instruction means for instructing to rewrite the coupling coefficient already stored in the difference storage means to the difference obtained by newly calculating and to use as the coupling coefficient in the neural network is provided. The chaos control device according to claim 3, which is provided. 8. The input signal is image data obtained by reading a document, the output signal is image data obtained by reading a recorded image recorded on a transfer sheet, and the controlled system is an image provided with a photoconductor. The chaos control device according to claim 3 or 7, which is a control unit that controls the electrostatographic process of the forming apparatus, and the state quantity is a detection value of a detection unit that detects the state of the electrostatographic process. 9. A neural network is formed by meshing a plurality of neurons in association with a coupling coefficient, and in a chaotic neural network, a plurality of neurons to which a chaotically varying function is given are associated with a coupling coefficient. The image forming apparatus is connected to each other while being connected to each other, and the image forming apparatus reads an original and converts it into image data, and a latent image of the original image by exposing the photoconductor based on the image data obtained by the image reading means. To form a latent image formed on the photoconductor by toner to develop the latent image formed on the photoconductor and convert the latent image into a visible image. A transfer unit that transfers the visible image, an image data amount detection unit that detects the amount of black pixels of the image data obtained by the image reading unit, and a humidity near the photoconductor. Humidity detecting means for detecting the temperature, temperature detecting means for detecting the temperature in the vicinity of the photoconductor, toner concentration detecting means for detecting the toner concentration of the developing means, and surface potential detecting means for detecting the surface potential of the photoconductor. A charging control means for controlling the amount of charge by the charging means, an exposure amount control means for controlling the exposure amount of the exposure means, and a developing bias voltage control for controlling the bias voltage between the photoconductor and the developing means. Means and a recorded image detecting means for reading the visible image transferred onto the transfer paper by the transferring means and converting it into image data, and the first rewriting instruction means further calculates a difference from the difference stored in the difference storing means. 9. The chaos control device according to claim 8, wherein when the difference calculated by the means is small, a rewriting instruction signal for instructing rewriting of the coupling coefficient stored in the coupling coefficient storage means is output. . 10. A controlled device which performs learning based on a preset output signal and a plurality of coupling coefficients, outputs a coupling coefficient corresponding to the learning result each time learning is performed, and cannot specify a relationship between an input signal and an output signal. A process of detecting a state quantity of the system and storing a plurality of coupling coefficients of a neural network for outputting a control amount for controlling the output signal to the controlled system in a coupling coefficient storage means, and chasing the plurality of coupling coefficients chaotically. Recalling, the process of using the output vector of the chaotic neural network as the coupling coefficient in the neural network, the process of storing the coupling coefficient, the output signal of the controlled system and the preset output signal. A step of calculating the difference and storing the difference in the difference storage means, a step of sequentially storing the difference each time learning is performed, and a step of being already stored in the difference storage means A chaos control device including a step of comparing a difference with a difference newly obtained by calculation, and a step of instructing rewriting of a coupling coefficient already stored in the coupling coefficient storage means based on a result of comparing the difference. Learning method. 11. A process of determining the end of learning of a neural network and a difference obtained after the learning of the neural network is compared, and as a result, a difference already stored in the difference storage means is newly obtained as a difference. 11. The learning method for a chaos control device according to claim 10, further comprising the step of using as a coupling coefficient in the neural network without rewriting the coupling coefficient already stored in the coupling coefficient storage means when the value is smaller. 12. A controlled device in which learning is performed based on a preset output signal and a plurality of coupling coefficients, a coupling coefficient corresponding to the learning result is output each time learning is performed, and a relationship between an input signal and an output signal cannot be specified. A neural network that detects a state quantity of the system and outputs a controlled variable that controls the output signal to the controlled system, and a coupling coefficient that causes chaotic behavior of the neural network are stored in advance, and the neural network is coupled. A chaotic neural network for generating a coefficient, and a learning method of a chaotic controller comprising at least a coupling coefficient storage means for storing a coupling coefficient in a rewritable state, the process comprising storing a plurality of coupling coefficients of the neural network. , Multiple coupling coefficients are chaotically recalled, and the output vector of the chaotic neural network is The process of using the cutout as a coupling coefficient in the neural network, the step of storing the coupling coefficient, the step of calculating the difference between the output signal output by the controlled system and the preset output signal, and the sequence of learning each time. A step of storing the difference, a step of comparing the difference already stored in the difference storage means with a difference obtained by a new calculation, and a step of storing the difference in the coupling coefficient storage means based on the result of the difference comparison. Method for a chaos control device including a process of instructing rewriting of a coupling coefficient being performed.