JPH08129404A - Controller and control method for object to be controlled - Google Patents

Abstract

PURPOSE: To improve the reliability for control of such a system to be controlled that is indefinite as to which monitoring information is used as a controlling variable to decide operation fee by acquiring automatically the logical structure of the rule of thumb of an operator through observation of the examples of controls carried out by the operator. CONSTITUTION: An example recording means 2 stores the example data which are shown in a pair of the monitor information on a object to be controlled and the manipulated variable set by the instruction of an operator together with the time series sequence information for each example. A state estimation means 3 extracts the monitor information that has a prescribed approximation degree to other monitor information among those recorded example data and calculates a conviction degree of coincidence between the state information on the extracted monitor information and that on other monitor information. Then, a control path generation means 4 traces the sequence information and extracts an example candidate that can reach a target state, and a manipulated variable decision means 5 selects the state information based on the calculated assurance degree and the example candidate. Thus the controlled system is controlled.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to, for example, a control device and a control method used for plant control and the like, and in particular, obtains control knowledge possessed by a skilled operator (operator) as objective information, The present invention relates to a control device and a control method of an object to be controlled, which is capable of performing optimum control without using the rule format.

[0002]

2. Description of the Related Art Monitoring information for controlling ventilation in tunnels, furnace temperature of blast furnaces, sludge in settling basins of sewage treatment plants, shape of steel sheet in steel sheet rolling process of steel plant (steel sheet rolling shape), traffic signals, etc. As a result, a control that changes by evaluating a large number of patterns related to the controlled object, such as a pattern composed of a large number of monitoring images related to the controlled object and a large number of measured values related to the controlled object, etc. It is important to estimate the state of the target and determine the amount of manipulation appropriately. That is, data representing the states at a large number of observation points of a plant are measured by observation means provided at each observation point, collected online, and based on these data, the ever-changing plant state is evaluated and the manipulated variable It is important to make a proper decision.

However, since the state change of these controlled objects depends on a non-linear factor, it has been difficult to obtain an appropriate manipulated variable by using a conventional technique such as sequence control.

As a solution to such a problem, knowledge engineering has been conventionally applied as seen in, for example, “Technology for intelligent process control systems”, measurement and control, vol.31, no.7. According to this method, the neural network is made to learn the correspondence between the pattern and the state of the controlled object, and the relation between the state of the controlled object and an appropriate manipulated variable is based on the empirical rule of the operator. Can be described in the form of fuzzy rules, that is, expressions in language, and the amount of operation can be determined by fuzzy reasoning.

Thus, a control device capable of performing sophisticated control corresponding to a non-linear state change of the controlled object by incorporating an empirical rule concerning control of the controlled object, which is possessed by a skilled operator with respect to the controlled object. It will be realized.

[0006]

However, the conventional control device as described above has the following problems.

First, tunnel ventilation, blast furnace temperature,
In the control of sewage sludge, steel plate rolling shape, traffic signals, etc., the operator (operator) looks at the image showing the state at many observation points of the plant, evaluates the state of the plant that changes moment by moment, and manipulates the operation amount. This image is, for example,
Since the operator contains a large amount of spatially-distributed information, such as the distribution of soot concentration in the tunnel, the temperature distribution in the furnace of the blast furnace, the shape of sludge, the shape of the steel plate, the distribution of vehicle volume, etc. It is very difficult to clarify which information included in the image is used to determine the operation amount. Therefore, the control technology by the conventional expert system etc., which must clearly specify the controlled variable,
It was difficult to apply to these plant control devices.

That is, in order to describe the empirical rules of the operator in a linguistic expression represented by a rule format, it is necessary to clearly define the control amount, but the operator can store various information including images. It is extremely difficult to clarify what part and how the operation amount is determined. Therefore, it is difficult to apply the control technology by the conventional expert system or the like, which requires the operator's empirical rule to be defined and programmed in a rule format to these control devices.

Secondly, when describing the rules of thumb of an operator in a rule format, it is necessary to clarify what kind of logical judgment the operator makes. In many cases, the empirical rules possessed by the operators were merely pieces of knowledge that did not have a logical structure, and it was also difficult to extract the logical structure of the empirical rules from the operator, such as the order of application of rules.

That is, even if the knowledge from the operator who does not have a clear logical structure of the empirical rule is to be used, in order to grasp the state based on various indexes and to judge the operation contents to be performed, those indexes are used. Can not be applied to the control technology by the conventional expert system etc. which must be programmed clearly how to evaluate
It was difficult to apply to the above-mentioned control device.

Thirdly, the transition of the state of the controlled object often has dynamics, and in this case, the state of the controlled object cannot be estimated only from the instantaneous image information measured at a certain time. Therefore, it is difficult to obtain an appropriate manipulated variable. For example, even if the soot concentration in the tunnel is high, is it because the number of passing vehicles is increasing and is it dark, or is it because it is temporarily congested and dark?
It is not possible to grasp from the instantaneous monitoring information what direction the situation transitions, such as whether the soot concentration is increasing or decreasing, and what is the optimum operation amount? Is difficult to judge from instantaneous information.

Fourthly, there is no function of selecting an efficient control method and performing optimum control by searching a logical structure of an empirical rule concerning control of a controlled object obtained from an operator. In other words, if it is not possible to perform operation control while deciding how to perform the most appropriate operation considering the current state, and if a certain control operation is determined, Since only the operation amount corresponding to that is increased / decreased, it is not possible to flexibly change the best operation target and the operation amount at each time according to a situation that changes every moment. Therefore, the control cannot be performed economically and in a short time.

Therefore, the present invention has been made in order to solve the above-mentioned problems of the prior art. The first object of the present invention is, firstly, that the operator is the monitoring information. In control of a target in which it is not clear which information contained in (monitored image, combination pattern of many measured values, etc.) is used as the control amount and the operation amount is determined, describe the rule of thumb of the operator. It is an object of the present invention to provide a control device that semi-automatically controls a control target and reduces the work load on an operator by acquiring an empirical rule regarding control of the control target without performing the difficult work of performing.

A second object of the present invention is to automatically obtain the logical structure of the empirical rule such as the order of application of the empirical rule concerning the control of the operator by observing the case of the control by the operator. It is to provide a control device.

A third object of the present invention is to provide a pattern (monitoring information including monitoring images, etc.) consisting of a combination of measured values related to the controlled object, and the state of the controlled object determined by the operator by looking at the pattern. When the transition of the state of the controlled object has dynamics by recording the correspondence of the monitoring information and the temporal change of the pattern that is the monitoring information and the operation amount determined by the operator looking at the pattern. Also,
It is an object of the present invention to provide a control device that can generate a hypothesis of a state of a controlled object from instantaneous monitoring information measured at a certain time point and determine an appropriate manipulated variable from the hypothesis of the generated state.

A fourth object of the present invention is to select a safe and efficient control method by searching a logical structure of an empirical rule concerning the control of an object obtained from an operator, and control the control accordingly. It is to provide a control device for performing.

[0017]

In order to achieve the above object, the present invention is configured as follows. That is, according to the present invention, firstly, a monitoring information collecting unit that collects monitoring information about a control target, monitoring information collected by the monitoring information collecting unit, status information of a control target for the monitoring information, and the status information. On the other hand, a case recording means for recording case data represented by a set with an operation amount based on an instruction given by the operator together with time-series order information for each case, and the case recording means. Of the case data, the monitoring information having a predetermined degree of approximation to the other monitoring information collected by the monitoring information collecting means is extracted, and the status information for the other monitoring information and the extracted monitoring are extracted. It is extracted by the state estimating means with reference to the state estimating means for obtaining a certainty factor of coincidence between the information and the state information and the case data recorded in the case recording means. Based on the state information for the visual information and the information about the target state of the controlled object that is separately provided, the control path unit that traces the order information and extracts a case candidate that can reach the target state, and the state estimation unit. An operation amount that selects desired state information based on the determined certainty factor and the case candidates extracted by the control path generation means, and gives an operation amount for the case data for the selected state information to the control target And a determining means.

Secondly, the control device of the first structure is further provided with a time measuring means for giving time information, and the case recording means is provided with monitoring information (measurement value pattern).
And case data represented by a set of a state of the controlled object supplied from the monitoring information collecting means, an operation amount for the controlled object based on an instruction given to the state information by an operator, and the time information. Is added to each case in time series, the control path generating means refers to the case data recorded in the case recording means, and the status information for the monitoring information extracted by the status estimating means is added. And a function of extracting the case candidate that can reach the target state by tracking the time information based on the information about the target state of the controlled object which is separately provided. .

Thirdly, in the control device having the first or second configuration, further, the record maintenance for taking in the case data recorded in the case recording means and for registering, correcting and deleting them. It is characterized by the addition of means.

Fourthly, a collecting step for collecting monitoring information about the controlled object, the monitoring information collected by this collecting step, status information of the controlled object for the monitoring information, and an operator for the status information. A case recording step of recording case data represented by a set with an operation amount based on the instruction of each case together with time-series order information for each case,
Of the case data recorded in this case recording step, the monitoring information having a predetermined degree of approximation to the other monitoring information collected in the collecting step is extracted, and the status information for the other monitoring information is extracted. The state estimation step of obtaining a certainty factor of a match between the state information for the extracted monitoring information and the case data recorded in the case recording step, and the monitoring information extracted in the state estimation step On the basis of the state information for the target state and the information about the target state of the controlled object, which is obtained by the control path extraction step of extracting the case candidates that can reach the target state by tracking the order information, and the state estimation step. Based on the certainty factor and the case candidates extracted by the control path extracting step, desired state information is selected, and the selected state information is selected. More configuration and the operation amount determination step of providing a manipulated variable to the controlled object with respect to case data for state information.

[0021]

The control device having the first configuration is for achieving the above-mentioned first to fourth objects, and the monitoring information collecting means includes the monitoring information on the control target (pattern of measured values on the control target, for example, monitoring). (Pattern information in which physical information detected from each part of an image or a control target is arranged) is captured.

The case recording means includes the monitoring information collected by the monitoring information collecting means, status information of a control target for the monitoring information, and an operation amount based on an operator's instruction for the status information (an operator based on the monitoring information. Case data represented by a set of (manipulator's operation amount instructed to the controlled object) is recorded together with time-series order information for each case. Then, the state estimating means extracts, from the case data recorded in the case recording means, the monitoring information having a predetermined degree of approximation to the other monitoring information collected by the monitoring information collecting means. Then, the certainty factor of coincidence between the status information for the other monitoring information and the status information for the extracted monitoring information is obtained.

On the other hand, the control path generating means refers to the case data recorded in the case recording means, and relates to the status information for the monitoring information extracted by the status estimating means and the target status of the control target separately given. Based on the information, the order information is tracked to extract case candidates that can reach the target state.

The manipulated variable determining means selects desired state information based on the certainty factor obtained by the state estimating means and the case candidates extracted by the control path generating means, and the selected state information is selected. The operation amount for the case data is given to the controlled object.

Therefore, according to this configuration, which information included in the monitoring information (measurement value pattern regarding the controlled object) is used as the controlled variable by the operator (operator) to determine the manipulated variable. In the control of an unclear object, the controlled object is controlled automatically or semi-automatically by acquiring the rule of thumb regarding the control of the controlled object without performing the difficult work of describing the rule of thumb of the operator in a rule format. can do. Furthermore, the logical structure of the empirical rules such as the order of application of the empirical rules concerning the control of the operator can be automatically acquired by observing the case of the control by the operator.

Further, in the control device of the second configuration,
The monitoring information (measurement value pattern regarding the control target) is captured by the monitoring information collecting unit. It is also presented to the operator who operates the controlled object. The monitoring information collecting means is
The state of the controlled object estimated by the operator by looking at the monitoring information and the operation amount in the present state determined by the operator are fetched. This manipulated variable is added to the controlled object. Subsequently, the case recording means fetches the monitoring information, the state and the operation amount decided by the operator, and further fetches the current time from the time measuring means. Then, the time, the time elapsed from one step before to the present, the monitoring information, the state, the operation amount, and the corresponding relationship between them are recorded in order. By repeating the above operation, the empirical knowledge of the operator concerning the control of the target that changes with time is suitable for the time, elapsed time, various states of the control target, monitoring information in that state, and that state. It is acquired in order as case data indicating the correspondence with the operation amount. The control device automatically controls the control target as described below, using the empirical knowledge of the operator acquired and accumulated in this way. The monitoring information of the observed value regarding the controlled object is input to the state estimating means. Next, this state estimation means selects, from the past monitoring information stored in the case recording means, one that is close to the currently observed monitoring information and determines the state recorded in correspondence with the selected monitoring information. It is the current hypothesis of the controlled object. At the same time, the degree of similarity between the currently monitored information and the selected monitoring information is taken as the certainty factor of the state hypothesis.
Subsequently, the control path searching means takes in the record stored in the case recording means, searches for a record corresponding to the current state hypothesis of the control target supplied by the state estimating means, and controls to reach the target state of the control target. Search for route candidates. Next, the manipulated variable determining unit takes in the control route candidates supplied by the control route searching unit and the certainty factor supplied by the state estimating unit, and calculates the manipulated variable for the controlled object.

Therefore, in addition to the effect of the device having the first configuration, the same is achieved when the controlled object is in the same state due to the difference in time and the difference in the time elapsed from the step one step before to the present time. Even if the controlled object has complex dynamics that transit to different states even if the operation amount is added, a hypothesis of the controlled object state is generated from the instantaneous monitoring information measured at a certain time, and the generated state is generated. From this hypothesis, an appropriate amount of operation can be determined.

Further, in the control device of the third configuration,
The record maintenance means fetches the state of the controlled object and the operation amount and the correspondence relationship with respect to the controlled object recorded in the case recording means, and newly registers and deletes these records. In addition to these operations, operations similar to those of the control device having the first or second configuration are further performed.

Therefore, if the operator's empirical rule stored in the case recording means has an error, the error can be removed.

In the fourth case, the monitoring information about the controlled object is collected in the collecting step, and in the case recording step, the collected monitoring information, the status information of the controlled object for the monitoring information, and the status information. On the other hand, the case data expressed as a set with the operation amount based on the instruction given by the operator is recorded together with the time-series order information for each case. Then, the state estimating step extracts, from the case data recorded in the case recording step, the monitoring information having a predetermined degree of approximation to the other monitoring information collected in the collecting step, and The degree of certainty of coincidence between the status information for the monitoring information and the status information for the extracted monitoring information is obtained.

On the other hand, in the control path extracting step, the case data recorded in the case recording step is referred to, the state information for the monitoring information extracted in the state estimating step, and the target state of the control target separately given. Based on the information regarding, the case information that can reach the target state is extracted by tracing the order information, and the operation amount determination step is extracted by the certainty factor obtained in the state estimation step and the control path extraction step. The desired state information is selected based on the selected case candidates, and the operation amount for the case data for the selected state information is given to the control target.

Therefore, according to this method, which information contained in the monitoring information (measured value pattern regarding the controlled object) is used as the controlled variable by the operator (operator) to determine the manipulated variable. In the control of an unclear object, the controlled object is controlled automatically or semi-automatically by acquiring the rule of thumb regarding the control of the controlled object without performing the difficult work of describing the rule of thumb of the operator in a rule format. can do. Furthermore, the logical structure of the empirical rules such as the order of application of the empirical rules concerning the control of the operator can be automatically acquired by observing the case of the control by the operator.

[0033]

Embodiments of the present invention will be described below with reference to the drawings.

In the plant operation control, the present invention extracts and learns control knowledge of a trained operator, which cannot be verbalized, and based on the learned knowledge, automatically makes the same judgment and operation as a trained person does. The present invention provides a plant control device capable of monitoring and controlling a plant by performing the automatic control. Therefore, the present invention is, for example, empirical knowledge that is difficult for a trained operator to verbalize and explain, such as tunnel ventilation control, blast furnace temperature control, sewage sludge control, steel plate rolling shape control, traffic signal area control, etc. It can be used for automatic control of a plant controlled by, and enables control without waste. Hereinafter, the present invention will be described with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing the basic configuration of the present invention. In the figure, 1 is an input / output means, and this input / output means 1 is a measured value for a controlled object such as a plant, for example, a tunnel ventilation, a furnace temperature control of a blast furnace, a steel plate rolling shape control, a traffic signal and the like. Pattern (including monitoring information such as a monitoring image) is displayed and displayed on a display, and the state S of the controlled object and the operation amount U for the controlled object are acquired. For example, an image display (display) ) And an input operation device such as a keyboard, a man-machine interface having a man-machine interface, a personal computer, and an intelligent terminal such as a workstation.

Reference numeral 2 is a case recording means, and the case recording means 2 controls the monitor image (measurement value pattern, monitor image, etc.) given from the input / output means 1 and the input / output means 1. This is for taking in and recording the state S of the target and the manipulated variable U for the controlled target, and recording the corresponding relationship between them.

Reference numeral 3 denotes a state estimating means, which takes in the pattern (monitoring information) and the pattern (monitoring information) recorded in the case recording means 2 and the state S of the controlled object. , Hypothesis S of the controlled state
x and its confidence factor b {Sx} are generated.

Reference numeral 4 denotes a control route search generating means, which controls the state S of the controlled object recorded in the case recording means 2, the operation amount U for the controlled object, and the corresponding relationship, and The state hypothesis Sx of the controlled object and the target state T of the controlled object supplied from the state estimation means 3 are taken in and a candidate for the control path is generated.

Reference numeral 5 denotes an operation amount determining means, which operates from the control route search generating means 4 and the certainty factor b {Sx} of the state hypothesis Sx supplied from the state estimating means 3. It is for determining the operation amount U for the control target by taking in the supplied control path candidate.

The operation of the present apparatus having such a configuration will be described with reference to the flowchart. In this device, the operation mode includes a system learning stage and an automatic driving stage in which the vehicle automatically operates based on the learning result.

[First step of introduction (learning mode operation)] The learning mode operation step is a step of constructing a system so as to collect empirical knowledge about control of an operator and utilize it for automatic operation. Implement at the stage of introduction. This mode setting is performed by the operator performing an input operation using the input / output means 1.

The operation in the learning mode operation stage will be described first. The operator inputs the mode from the input / output means 1 to set the system to the mode of the driving operation at the learning stage. Then, the system operates as shown in the flowchart of FIG. That is, in FIG. 2, the control target outputs the monitoring information referred to by the operator as one of the measurement information when controlling the control target, for example, the “measurement value pattern of the control target”. Is taken into the system (step St1 in FIG. 2).

Here, the pattern of the measured value as the monitoring information specifically differs depending on the individual control target. For example, in the case of tunnel ventilation control, a monitoring camera (at the measurement point in the tunnel) (TV camera), which is an image (monitoring image) showing the distribution of soot concentration in the tunnel captured by the monitoring camera. In addition, the pattern of measurement values in blast furnace temperature control when the controlled object is a blast furnace is the two-dimensional output from multiple temperature sensors installed at various locations in the furnace to measure the temperature in the furnace of the blast furnace. It is a pattern image created by arranging in a plane, and in the case of a sewage treatment plant, it is an image captured by a surveillance camera installed in the sedimentation pond to capture the shape of the sewage sludge in the chemical process of sludge precipitation. In the case of control, it is a pattern created by arranging the outputs of sensors that measure the thickness of the steel sheet at various points in the rolling plant in a two-dimensional plane, or in the traffic signal system control, it is a monitor installed at various points on the road. For example, it is an image that is captured from the camera and shows the traffic situation of the vehicle. Of course, the present invention is not limited to this, and any monitor image (pattern) that can grasp the situation may be used.

Such pattern information fetched from the controlled object is input to the input / output unit 1 and the case recording unit 2. The input / output unit 1 is, for example, an intelligent terminal such as a personal computer used when driving an object such as an operating console, and has a display and a keyboard, and outputs the pattern information as described above. It can be captured and output to the display, and the operation amount U can be given to the controlled object (step St2 in FIG. 2).

The operator looks at the pattern information of the controlled object displayed on the input / output means 1 and inputs the manipulated variable U for the controlled object if necessary, and gives it to the controlled object as a command. This manipulated variable U is given to the above-mentioned controlled object from the input / output means 1, and the controlled object carries out an operation corresponding to the manipulated variable U (step St3 in FIG. 2).

The operation amount U given to the controlled object by the operator by operating the input / output means 1 is the state S exhibited by the controlled object at the time when the pattern observed on the controlled object occurs. On the other hand, it is the content of the treatment itself for transitioning this to the target state T.

Therefore, in the learning stage, the output of the input / output means 1 is taken into the case recording means 2 together with the pattern information (step St4 in FIG. 2). The information is collected by the case recording means 2 in the learning stage. The state S of the control target when this pattern is observed on the control target, and the order relationship between the record one step before and the current record are also recorded as information. To do. The state S of the control target is label information, and the input / output unit 1 gives this label.

In this way, in the case recording means 2 in the learning stage, the pattern information, the state S of the controlled object when this pattern is observed on the controlled object, and the state of the controlled object transit to the target state T. Information on the operation performed by the operator to perform the operation (information on the appropriate operation amount U performed by the operator to transition the current state of the control target to the target state T) is collected.

The information on the state S of the controlled object and the appropriate manipulated variable U collected is the information resulting from the empirical knowledge accumulated by the skilled operator engaged in the control of the object. Therefore, although it is not a theoretical analysis of empirical knowledge, it is useful as information containing empirical knowledge.

In this way, in the mode of the learning stage, the case recording means 2 takes in a pattern of measured values from the outside and also takes in the state S of the controlled object and the appropriate manipulated variable U from the input / output means 1 and records them. Further, the order relationship between the recording one step before and the current recording is also recorded as information (step St4 in FIG. 2).

By repeating this, the empirical knowledge about the control of the operator is accumulated in the case recording means 2. That is, it is determined in step St5 whether or not the learning mode has been released. If not, that is, if the case recording is to be continued, the process returns to step St1 and the above-described processing is repeated to collect the case. Go on.

Here, the state S to be controlled is label information, which is arbitrary code information given in correspondence with the pattern of the measured value. The label is automatically generated and given in order, but the character string of the label name given at the beginning may be given by the operator by operating the input / output means 1.

When the operator determines that the case has been sufficiently sampled, at that stage the operator operates the input / output means 1 to end the learning mode and sets the system to the automatic operation mode.

As a result, the system starts the automatic operation reflecting the empirical knowledge of the skilled operator accumulated in the case recording means 2. Next, this will be explained.

[Flow of Processing During Control] Next, the operation when the present invention controls the controlled object will be described based on the control cases sampled in the learning mode operation. The data and the flow of processing in this mode will be described with reference to FIG. It is to be noted that FIG. 4 is a partially enlarged view of a portion from Start to Step St13 of FIG. 3 and a portion from Symbols A and B to Step St22, and a portion from Step St14 to Step St16 which is a subsequent stage of Step St13 of FIG. The part from Step St20 to End which is the latter stage of Step St13 and Step St22 and the step St23 to End which is the latter stage of Step St22.
5 and an enlarged view of a part up to step St24.
Then, FIG. 6 shows a partially enlarged view of a portion from step St17, which is the latter stage of step St16 in FIG. 3, to the symbols A and B, and a portion from step St25, which is the latter stage of step St24, to the symbol C.

As shown in FIG. 3, first, a pattern of measured values of the controlled object at the present moment is fetched (step St11 in FIG. 3). The fetched current pattern is input to the input / output unit 1 and the state estimation unit 3. Then, the state estimating means 3 searches for past patterns close to the fetched present pattern from the cases recorded in the case recording means 2, and selects the closest one pattern or a plurality of patterns in the order of closeness. Yes (step S in FIG. 3)
t12).

Here, the closeness of the patterns is specifically different depending on each control target, but the following is an example.

For example, it is assumed that the pattern of measured values relating to the controlled object is an image vector from the camera showing the distribution of soot concentration in the tunnel in the tunnel ventilation control.

[0059]

[Equation 1] Therefore, since the closeness of the pattern can be expressed by the Euclidean distance, it is said that this is used.

If a pattern sufficiently close to the current pattern is not stored in the case recording means 2, it means that the accumulated cases are not sufficient. In this case, since it is not possible to automatically control according to the case, the manipulated variable determining means 5 gives an output of which the manipulated variable cannot be determined to the input / output means 1, and the input / output means 1 receives this output and is manually operated by the operator. The operation mode is entered, and a screen prompting a manual input operation by the operator is displayed on the display (step St13 in FIG. 3,
St20).

In this way, when sufficient cases for determining the manipulated variable are not accumulated in the case recording means 2, the manipulated variable determining means 5 causes the operator to execute the subsequent operations via the input / output means 1. Request control. After this, the mode changes to the control case recording mode described above (process flow at the time of control case recording). Since this mode is a manual mode, the operator determines the operation amount U required to make a transition to the target state by looking at the displayed pattern, and operates the input / output means 1 to set the operation amount U to the control target. Give them control. In this state, the learning mode also functions at the same time, the information of the state S and the operation amount U are generated and given to the case recording means 2,
The case recording means 2 collects this as information including the above pattern and the order relationship between the record one step before and the current record. This learning mode continues until the operator sets the system to automatic mode, during which
The operator determines the manipulated variable.

The description of the operation when the mode is not switched to the manual mode, which is the learning mode operation, will be continued.

In step St13, when the past pattern close to the present pattern is recorded in the case recording means 2, the state estimating means 3 then corresponds to the selected past pattern and the case recording means is recorded. The information of the state S of the controlled object recorded in No. 2 is fetched (step St14 in FIG. 3). Then, the state estimating means 3 performs the process of the next step St15.

That is, the acquired information on the state S becomes the hypothesis Sx of the state S to be controlled when the current pattern is observed. Then, the certainty factor of this hypothesis Sx is obtained. That is, the state S of the controlled object at the time when the current pattern is observed is mechanically extracted (only by external judgment) which pattern is presumed to be the situation close to the past pattern. Therefore, it is the hypothesis Sx referred to here that it is assumed that the extracted one is similar to the state S of the controlled object when the current pattern is observed. The degree of certainty is obtained as a numerical value of the degree of confidence.

Therefore, in step St15, from the above-mentioned captured past pattern and present pattern, when these patterns are observed, the certainty factor b {Sx} that the controlled objects are in the same state S can be obtained. calculate.

Here, the certainty factor b {Sx} is different depending on each control target, like the proximity of the pattern described above. For example, a camera whose pattern represents the distribution of soot concentration in the tunnel in the tunnel ventilation control. The current pattern when is an image vector from

[Equation 2] The state S of the controlled object when is observed is the pattern recorded in the past.

(Equation 3) State of the controlled object that was present when was observed

[Equation 4] The certainty factor b that can be said to be the same as

(Equation 5) Since it can be expressed as, it is said that it is used.

That is, the certainty factor of the image here means that the closer the image as the phenomenon that has just occurred is to the image recorded as the past case, the more the phenomenon that has just occurred becomes the past case. It is an indicator that it can be said that the situation is close to the phenomenon that occurred in the controlled object when the image was obtained.Therefore, the more similar the situations in which both cases occurred, the more confident the image of that case is. The degree increases.

Next, the control path generating means 4 fetches from outside the target state T of the current control target state S, that is, the target state T at which the control target should settle.
The state hypothesis Sx and its certainty factor b {Sx} supplied from the state estimating means 3 are fetched (step St1 in FIG. 3).
6). When the process of step St16 ends, the process proceeds to step St17.

Here, the target state T is set as, for example, a reference value in a setting means (not shown), and the target state T is given to the control path generation means 4 to be taken into the control path generation means 4 and set as the target state T. . Then, the confidence factor b {S
The control route candidates for reaching the target state T are obtained starting from the state of the image (pattern) corresponding to the hypothesis Sx having the highest x} (generation of control route candidates). The process of obtaining the control route candidates from past cases is the process after step St17.

The generation of the control path candidate may have occurred for the first time when the current state of the controlled object appears in the automatic operation, or has already appeared one or more times and the control path candidate having the origin as the starting point has already been obtained. Correspondence depends on whether or not. In other words, the control procedure for guiding the current state to the target state is searched from the already sampled past operation examples of the skilled person, and all the possibilities are extracted.
When it is used as the starting point, it differs depending on whether it is extracted for the first time or is extracted in the past and held as information in the case recording means 2 and referred to. In other words, it depends on whether you need to search or not. Therefore, these will be described separately.

[When Generating Control Route Candidates for the First Time] In the processing of step St17, the control route generating means 4 determines whether it is the first generation of control route candidates, and when generating the control route candidates for the first time, The process moves to the route of the symbol A in FIG. Then, the control path generation means 4 searches for a case including the above-described hypothesis Sx of the state S among the cases stored in the case recording means 2 and fetches it, starting from this case and reaching the case of the target state. A chain of cases,
It is generated by tracing the information of the order relation between the cases recorded in the case recording means 2 (step St18 in FIG. 3).

Since there are generally a plurality of state hypotheses Sx, there are also a plurality of generated case chains. The chain of these cases becomes a candidate for a route that changes the state S of the controlled object by controlling the controlled object, that is, a control route candidate. The control route candidates are recorded in the route candidate set, which is an internal storage of the control route generating means 4 (step S in FIG. 3).
t19). When this process ends, the process moves to step St22.

[Second Time and Later] On the other hand, when it is judged in the step St17 of FIG. 3 that the control path candidate has been generated for the second time and thereafter, the process of the symbol B is started. Then, in step St21, the control path candidates that do not include the state hypothesis are deleted from the internal storage.

That is, when the control route candidates are generated for the second and subsequent times, among the control route candidates recorded in the above-mentioned route candidate set, those which do not include the above-described hypothesis Sx are excluded from the control route candidates. Therefore, it is deleted from the route candidate set. As a result, the control route candidates are gradually narrowed down. When the route candidate set is emptied by this erasing process, it means that there are not enough cases accumulated in the case recording means 2. At this time, since it is not possible to automatically control according to the past cases, the operator is requested to perform subsequent control via the input / output unit 1 (FIG. 3).
Steps St22 and St20). After that, the control state by the manual operation of the operator described in the section (Processing flow at the time of recording the control case) is performed, and the operation changes to the learning mode operation stage.

In step St22, if the control route candidate set is not empty, the process proceeds to step St23. That is, the process of step St23 is ended by judging whether or not the target state T is reached, and if it is reached, the process is ended.

This is because some of the above-mentioned state hypotheses Sx match the target state T, and moreover, the state hypotheses Sx
, There is a sufficiently high confidence factor b {Sx}, which means that the control has reached the target state T with a high confidence factor b {Sx}. Since it can be said that the state has reached, the control is ended (steps St22 and S23 in FIG. 3).

On the other hand, if it is determined in step St23 that the control route candidate set is not empty and the controlled object has not reached the target state T, further control is required. Therefore, the process proceeds to step St24.

This processing is performed by the manipulated variable determining means 5, and the manipulated variable determining means 5 is the state estimating means 3.
The certainty factor b {Sx} of the state hypothesis Sx supplied from the above is taken in, and the higher the certainty factor b {Sx}, the higher the certainty factor b.
Is the confidence factor b {Sx} and the current state hypothesis S of the controlled object.
It is determined using the manipulated variable U for making a transition from x to the next state. The operation amount U thus determined by the operation amount determining means 5 is given to the control target and is subjected to control corresponding to the operation amount U (step St25 in FIG. 3).

When this process ends, the process moves to the symbol C, and the processes after step St11 are performed. That is, the pattern of the measured value of the controlled object changed by adding the determined operation amount is fetched as a new current pattern and input to the input / output unit 1 and the state estimation unit 3. Then, the above-described processing based on this is repeatedly performed.

As a result, the pattern of past measured values collected from the control target and the contents of the control performed by the experienced operator on the control target are collected as a past case together with the control history. Just by searching the past case for the pattern closest to the measured value pattern presented by the current control target, the same control is performed according to the control content at that time and the history traced at that time Optimum control using the empirical knowledge of a skilled operator can be realized by generating a quantity and giving it to the controlled object.

As described above, this embodiment has the display unit, the operation unit, and the input / output unit. The monitoring information about the control target is fetched through the input / output unit and displayed on the display unit, and the current control target is displayed. Input / output means for sequentially outputting the state information for specifying the state of and the operation amount according to the instruction from the operation unit of the operator with respect to the controlled object, the monitoring information, the state information, and the operation amount as case data. A case recording unit that accumulates information and holds case data that sequentially accumulates information in a time-series correspondence relationship, monitoring information of the case data recorded in the case recording unit, and its monitoring information. Of the case data in which the status of the control target corresponds to the current monitoring information, and the current monitoring information regarding the control target is acquired. The corresponding monitoring information is extracted and temporarily defined as the same state, and the state hypothesis that the state of the monitoring information is the relevant state is set, and the certainty factor for this state hypothesis is set to the state approximation degree. Based on the state estimation means obtained based on the case estimation means and the case data group recorded in the case recording means together with the information on the correspondence, the state hypothesis of the control object supplied from the state estimation means, and the control object separately given The target state information is used to extract candidate case data that can reach the target state by history tracking from the case data, and control path generation means for generating control path candidates to be traced; and the state estimation means. An operation included in the case data having the state information corresponding to the state hypothesis having a high certainty factor by incorporating the certainty factor of the supplied state hypothesis and the candidate supplied from the control path generating means. Determining an amount as an operation amount for the control target, a configuration and a manipulated variable determining means for providing the control target.

Then, the input / output means uses the monitoring information regarding the controlled object (the pattern of measured values regarding the controlled object, for example,
Captures the monitoring image and the pattern information in which the physical information detected from each part of the controlled object is arranged), displays it on the display,
When the transporter operates the operation unit to instruct the operation amount for the control target based on the displayed monitoring information, the input / output unit outputs the instructed operation amount and the name of the current state of the control target. To do. Then, the case recording means captures and records the monitoring information, the state name of the control target and the operation amount with respect to the control target supplied from the input / output means, and records the correspondence between them. The state estimating means takes in the monitoring information of the current controlled object and the monitoring information and the status name recorded in the case recording means, and selects the current controlled object among the recorded monitoring information of the past controlled objects. The state of is extracted as a state hypothesis that is similar to the state of the monitoring information extracted by extracting a state similar to the state of the past controlled object that has been recorded, and the certainty factor of the state hypothesis is approximated to both monitoring information. Generate from degrees. The control path generation means is separately provided with the state of the controlled object recorded in the case recording means, the operation amount and the correspondence relationship with the controlled object, and the state hypothesis of the controlled object supplied from the state estimating means. The target state of the controlled object is taken in, and using these, candidate case data that can reach the target state by history tracking is extracted from these case data, and candidate control paths to be followed are generated. A case in which the manipulated variable determining unit takes in the certainty factor of the state hypothesis supplied from the state estimating unit and the control route candidate supplied from the control route searching unit and has state information corresponding to the state hypothesis having a high certainty factor. The operation amount included in the data is determined and output as the operation amount for the control target.

Therefore, according to this configuration, it is not clear which information contained in the monitoring information (measurement value pattern regarding the controlled object) is used as the controlled variable by the operator to determine the manipulated variable. In control, by acquiring the rule of thumb for the control of the target without performing the difficult task of describing the rule of thumb of the operator in a rule format,
The controlled object can be controlled semi-automatically. further,
The logical structure of empirical rules such as the order of application of empirical rules relating to control of the operator can be automatically acquired by observing the case of control by the operator.

In the first embodiment described above, the pattern of past measured values collected from the controlled object and the contents of the control performed by the skilled operator on the controlled object are recorded together with the history of the control in the past. Just by collecting as an example, the pattern closest to the pattern of the measured value presented by the current control target is searched from past cases, and the same control is performed according to the control content at that time and the history traced at that time. By generating a controlled variable so as to carry out the above and giving it to the controlled object, it is possible to realize the optimal control utilizing the empirical knowledge of a skilled operator.

Here, in the first embodiment, when collecting a case, when the history of the transitions of the control can be reproduced and a phenomenon close to the past case occurs in the controlled object, the history is recorded. However, there is a peculiar phenomenon that occurs only in a specific time zone in the form of control, and the optimum control is performed by performing similar control. If the time zone information can also be referred to, optimal control can be realized in accordance with the situation. Therefore, the second embodiment of the present invention makes it possible to properly cope with a peculiar phenomenon that occurs only in a specific time zone by collecting cases including such temporal information.
Will be described as an example.

(Second Embodiment) FIG. 7 is a block diagram showing the structure of the second embodiment. In addition to the structure of the first embodiment described above, a clock means 6 for supplying time information is added. It is configured. Further, in the second embodiment, in addition to the case recording means 2 in the configuration of the first embodiment, a function of taking in and recording the time information supplied from the time measuring means 6 is further added. In addition to the information indicating the correspondence between the state S of the control target to be supplied, the operation amount U for the control target, and the collected information in the previous step,
The time information supplied from the time measuring means 6 is also taken in,
I try to record it.

Therefore, in the second embodiment, the case recording means is indicated by reference numeral 2a as shown in the figure, and is displayed separately from that of the first embodiment. Since the time information is also taken into consideration, the state estimating means 3a is provided instead of the state estimating means 3 in the configuration of the first embodiment, and the state estimating means 3a is provided from the controlled object side. The pattern of the measured value, the pattern of the measured value recorded in the case recording means 2a, the information of the state S of the control target and the time information are taken in, and the present time and the time information of the recorded case In consideration of the correspondence with, the function of generating the hypothesis Sx of the controlled object state and the certainty factor b {Sx} thereof is provided. The other structure is basically the same as that of the first embodiment.

Next, an embodiment of the system having such a configuration will be described below. Learning mode operation is performed as the first step of system introduction. This will be described with reference to the flowchart of FIG.

[First step of introduction (learning mode operation)] The learning mode operation step is a step of constructing a system that collects empirical knowledge of control possessed by a skilled operator and makes it available for automatic operation. It is carried out at the stage when the equipment is installed. This mode is performed by the operator performing an input operation through the input / output means 1. When this mode is set and the operation is started, the procedure as shown in FIG. 8 is performed. That is, in step St31, the pattern of the above-described measured value of the control target is generated from the control target, and this pattern is taken into the present system. The captured pattern is input to the input / output unit 1 and the case recording unit 2.

Next, the input / output means 1 outputs and displays the fetched pattern. Then, the operator sees the pattern information of the controlled object displayed on the input / output means 1 and operates the input / output means 1 if necessary to input the operation amount U for the controlled object. , This is given to the control target as a command (step St32 in FIG. 8,
St33). The operation amount U is an appropriate operation amount determined by a skilled operator in accordance with the state of the control target S described above.

On the other hand, the manipulated variable U output from the input / output means 1 is sent to the case recording means 2a in addition to the controlled object. Further, the time counting means 6 generates current time information and gives it to the case recording means 2a (step St34 in FIG. 8).

As a result, the case recording means 2a is provided with the state S of the controlled object, the manipulated variable U, and the current time information. The control target state S is label information,
The input / output unit 1 gives this label.

As a result, the case recording means 2a fetches the pattern of measured values from the outside, the state S of the controlled object and the appropriate manipulated variable U from the input / output means 1, and the current time from the timing means 6. The difference Δt between the captured data and the time when the information is captured one step before, which is separately obtained by the case recording unit 2a, is recorded as information.
Further, the order relationship between the recording one step before and the recording presently is also recorded (step St35 in FIG. 8).

By repeating this until the learning mode is released (step St36 in FIG. 8), the actual information of the control operation performed by the operator for the state of the controlled object can be collected. In other words, it is the result of the empirical knowledge of the control possessed by the operator and the control utilizing it, and therefore the empirical knowledge of the control possessed by the operator is the same as that accumulated in the case recording means 2.

[Flow of Processing During Control] Next, the operation when the system of the present invention automatically controls the controlled object will be described based on the control cases sampled by the learning mode operation.
The data and the flow of processing in this mode will be described with reference to FIG. In addition, from Step Start to Step St4 in FIG.
3 and symbol A to step St49 and symbol B
From Fig. 10 is an enlarged view of a part from step St52.
Then, from step St44 to step St45, which is the latter stage of step St43 in FIG. 9, and from step St50 to E, which is the latter stage of step St43 and step St52.
9 is a partial enlarged view of a part from step St53 to End which is the latter stage of step St52 in FIG. 9 and from step St53 to St54 in FIG. 11, and a step St which is the latter stage of step St45 in FIG.
46 to Step St47 and Step St54
FIG. 12 shows an enlarged view of a part up to step 55 which is the latter stage.

As shown in FIG. 9, first, a pattern of measured values of the controlled object at the present time is fetched (step St41 in FIG. 9). The fetched current pattern is input to the input / output unit 1 and the state estimating unit 3a. Then, the state estimating means 3a searches the past pattern close to the fetched present pattern from the cases recorded in the case recording means 2a, and finds the closest one pattern by one pattern.
Alternatively, a plurality of patterns are selected in the order of closeness (step St42 in FIG. 9).

Here, the closeness of the patterns is the same as that already described in the first embodiment.

If a pattern sufficiently close to the current pattern is not stored in the case recording means 2a, the accumulated cases are not sufficient. In this case, since it is not possible to automatically control according to the case, the manipulated variable determining means 5 gives an output of which the manipulated variable cannot be determined to the input / output means 1.
Upon receipt of this output, the input / output unit 1 shifts to the mode of manual operation by the operator and causes the display to display a screen prompting the manual input operation by the operator (step St4 in FIG. 9).
3, St50).

In this way, when sufficient cases for determining the manipulated variable U are not stored in the case recording means 2a, the manipulated variable determining means 5 sends the information to the operator via the input / output means 1. Request control of. After that, the mode is changed to the control case recording mode described in the above “[Flow of processing at control case recording]”. Since this mode is a manual mode, the operator determines the operation amount U required to make a transition to the target state by looking at the displayed pattern, and operates the input / output means 1 to set the operation amount U to the control target. Give them control. In this state, the learning mode also functions at the same time, and the information of the state S and the manipulated variable U are generated and given to the case recording means 2a. The case recording means 2a uses the pattern and the recording one step before and the current recording. Collect as information including the order relation of. The operation as this learning mode continues until the operator sets the system to the automatic driving mode,
During this period, the operator determines the manipulated variable.

The description of the operation when the mode is not switched to the learning mode operation and the manual operation is continued.

In step St43, when a past pattern close to the present pattern is recorded in the case recording means 2a, the state estimating means 3a then causes the case recording means to correspond to the selected past pattern. The information of the state S of the controlled object recorded in No. 2 is fetched (step St44 in FIG. 9). Then, the state estimating means 3a performs the process of the next step St45.

That is, the acquired information on the states S becomes the hypothesis Sx of the state S to be controlled when the current pattern is observed. Then, the certainty factor b {Sx} of this hypothesis Sx is obtained. That is, the state S of the controlled object at the time when the current pattern is observed is a pattern that is presumed to be in a state close to the past pattern, and is extracted only by the external judgment such as the degree of approximation of the pattern. Therefore, it is hypothesis Sx here that it is assumed that the extracted one is similar to the state S of the controlled object when the current pattern is observed, and the certainty of that assumption is It is the confidence factor b {Sx} that is obtained as a numerical value to what extent.

Therefore, at step St45, when these patterns are observed, the certainty factor b {Sx} that the controlled objects can be said to be the same state S is obtained from the previously fetched past pattern and present pattern. calculate.

Next, the control path generating means 4 fetches from outside the target state T of the current state S of the controlled object, that is, the target state T at which the controlled object should settle.
The state hypothesis Sx and its certainty factor b {Sx} supplied from the state estimating means 3a are fetched. In addition, the information of the current time output by the time measuring means 6 is recorded in the internal storage. (Step St46 of FIG. 9). Here, the target state T is set as a reference value in the setting means, for example, and the control path generation means 4 takes it in and sets it as the target state T.

When the process of step St46 is completed, the process proceeds to step St47. The process of step St47 is the generation of control route candidates. The generation of control route candidates is divided into the case where it is the first control route candidate and the case where it is not. Therefore, these will be described separately.

[When Generating Control Route Candidates for the First Time] In the processing of step St47, the control route generating means 4 determines whether it is the first generation of control route candidates, and when generating control route candidates for the first time, The process moves to the route of the symbol A in FIG. Then, the control path generation means 4 retrieves and acquires a case including the above-described hypothesis Sx of the state S from the cases stored in the case recording means 2a, and starts from this case to reach the case of the target state. A chain of cases is generated by tracing the information on the order relation between cases recorded in the case recording means 2a (step St48 in FIG. 9).

Since there are generally a plurality of state hypotheses Sx, there are also a plurality of generated case chains. The chain of these cases becomes a candidate for a route that changes the state S of the controlled object by controlling the controlled object, that is, a control route candidate. The control route candidates are recorded in the route candidate set which is the internal storage of the control route generating means 4 (step S in FIG. 9).
t49). When this process ends, the process moves to step St22.

[Second Time and Later] On the other hand, when it is judged in the step St47 of FIG. 9 that the control path candidate has been generated for the second time and thereafter, the process of the symbol B is started. Then, in step St51, a control path candidate that does not include a state hypothesis, a difference between the current time and the time one step before,
That is, in order to exclude from the control route candidates the control time candidates that are significantly different from the elapsed time included in the case,
Delete from the route candidate set (delete from internal memory).

That is, when the control route candidates are generated for the second time and thereafter, among the control route candidates recorded in the above-mentioned route candidate set, those not including the hypothesis Sx in the above-described state and those including the hypothesis Sx are not required. In order to remove from the control route candidates those that do not meet the physical condition, they are deleted from the route candidate set.
As a result, the control route candidates are gradually narrowed down. When the route candidate set is emptied by this erasing process, it means that there are not enough cases accumulated in the case recording means 2A. At this time, the input / output unit 1 cannot be automatically controlled according to the past cases.
The operator is requested to perform subsequent control via (Step St52, St50 in FIG. 9). After that, the control is manually performed by the operator as described in the section "[Flow of processing when recording control case]", and the operation changes to the learning mode operation stage.

In step St52, if the control route candidate set is not empty, the process proceeds to step St53. That is, the process of step St53 ends by determining whether or not the target state T has been reached.

This is because some of the above-mentioned state hypotheses Sx match the target state T, and moreover, the state hypotheses Sx
Confidence b {Sx} corresponding to is sufficiently high, and if the control route candidate set is not empty, high confidence b
Since {Sx} means that the control has reached the target state T, the control is ended (step St5 in FIG. 9).
2, St53).

On the other hand, if it is determined in step St53 that the control route candidate set is not empty and the control target has not reached the target state T, further control is required. Therefore, the process proceeds to step St54.

This processing is performed by the operation amount determining means 5, and the operation amount determining means 5 is the state estimating means 3.
The certainty factor b {Sx} of the state hypothesis Sx supplied from a is taken in, and the manipulated variable U is such that the higher the certainty factor b {Sx}, the higher the probability of transition to the next state along the chain of cases. , The certainty factor b {Sx} and the manipulated variable U for making a transition from the current state hypothesis Sx to be controlled to the next state. The operation amount U thus determined by the operation amount determining means 5 is given to the control target and is subjected to control corresponding to the operation amount U (step St55 in FIG. 9).

When this process ends, the process moves to the symbol C, and the processes after step St41 are performed. That is, the pattern of the measured value of the controlled object changed by adding the determined operation amount is fetched as a new current pattern and input to the input / output unit 1 and the state estimating unit 3a. Then, the above-described processing based on this is repeatedly performed.

As a result, the pattern of the past measured values collected from the control target and the contents of the control performed by the experienced operator on the control target are obtained from the control history and time information and the previous operation. Just by collecting as a past case together with the time difference information up to this operation, the pattern closest to it is added to the pattern of the measured value presented by the current control target in the past case. , The control amount is generated so that the same control is performed according to the control content at that time and the history traced at that time, and the control amount is given to the control target, and the empirical knowledge of the skilled operator is utilized. Optimal control that follows the delicate control pattern that occurs according to the time zone can be realized.

In the above embodiment, the control device of the first configuration is further provided with a clocking means for giving time information, and the case recording means is provided with monitoring information (measurement value pattern).
Each case data represented by a set of the state of the controlled object supplied from the monitoring information collecting means, the operation amount for the controlled object based on the instruction given by the operator for the state information, and the time information is represented. A function to record in time series for each case is added, the case information recorded in the case recording means is referred to the control path generation means, and status information for the monitoring information extracted by the status estimation means, Based on the information about the target state of the controlled object which is given separately, it is configured by adding a function of tracking the time information and extracting case candidates that can reach the target state. , The monitoring information (a pattern of measured values related to the controlled object) is fetched by the input / output means. It is also presented to the operator who operates the controlled object. The operator estimates the present state of the controlled object by looking at the monitoring information and gives an instruction, and also gives an operation amount suitable for the present state. The input / output unit takes in the state of the controlled object estimated by the operator by looking at the monitoring information and the operation amount suitable for the present state determined by the operator. This manipulated variable is added to the controlled object. Then, the case recording means fetches the monitoring information, the state and the operation amount decided by the operator, and further fetches the current time from the timekeeping means.

Then, the fetched time, the time elapsed from one step before to the present, and the monitoring information
Record the state / manipulation amount and the correspondence between them. By repeating the above operation, the operator's empirical knowledge about the control of the target that changes depending on time is suitable for the time, elapsed time, various states of the control target, monitoring information in that state, and in that state. It is acquired as a correspondence with the manipulated variable.

The control device automatically controls the controlled object as described below, using the empirical knowledge of the operator acquired and accumulated in this way. The monitoring information of the observed value regarding the controlled object is input to the state estimating means. Next, this state estimation means selects, from the past monitoring information stored in the case recording means, one that is close to the currently observed monitoring information and determines the state recorded in correspondence with the selected monitoring information. It is the current hypothesis of the controlled object. At the same time, the degree of similarity between the currently monitored information and the selected monitoring information is taken as the certainty factor of the state hypothesis. Subsequently, the control route searching means fetches the record (including time information) stored in the case recording means, searches for a record corresponding to the current state hypothesis of the control target supplied by the state estimating means, and searches for the control target. Search for control route candidates that reach the target state. Next, the manipulated variable determining unit takes in the control route candidates supplied by the control route searching unit and the certainty factor supplied by the state estimating unit, and calculates the manipulated variable for the controlled object.

Therefore, in addition to the effect of the device of the first embodiment, when the controlled object is in the same state due to the difference in time and the difference in the time elapsed from one step before to the present time. Even if a control target has complicated dynamics that transit to different states even if the same operation amount is added, a hypothesis of the state of the control target is generated from the instantaneous monitoring information measured at a certain time. From the state hypothesis,
The appropriate manipulated variable can be determined.

As described above, in the second embodiment, the time condition and the time difference condition are further added to the first embodiment, and the more detailed
Moreover, the control that reflects the empirical knowledge of the experienced operator can be realized even for the delicate control required under special circumstances. However, in both the first and second embodiments, the collected information is collected. It doesn't mention maintaining the.

However, in reality, maintenance of the collected information is important for realizing better control. Therefore, an example in which maintenance can be performed on the collected information will be described below as a third embodiment.

(Third Embodiment) FIG. 13 shows the configuration of the third embodiment. This configuration is, in addition to the configuration shown in the first embodiment, each state information of the control target S recorded in the case recording means 2, the operation amount U for the control target, and each information of the correspondence relationship with the previous step. Import and register these information,
A record maintenance means 7 for erasing and correcting is added. The record maintenance means 7 may be an editing terminal such as a personal computer or an editor device such as a microcomputer computer having keys and a display. The information recorded in the case recording means 2 is changed, deleted, or edited in accordance with the editing instruction from the record maintenance means 7. The other structure is basically the same as that of the first embodiment.

Next, an embodiment of the system having such a configuration will be described below. Learning mode operation is performed as the first step of system introduction. This will be described with reference to the flowchart of FIG.

[First step of introduction (learning mode operation)] The learning mode operation step is a step of constructing a system that collects empirical knowledge of control possessed by a skilled operator and makes it available for automatic operation. It is carried out at the stage when the equipment is installed. This mode is performed by the operator performing an input operation through the input / output means 1. When this mode is set and the operation is started, the procedure as shown in FIG. That is,
In step St61, the pattern of the above-mentioned measured value of the controlled object is generated from the controlled object and is taken into the system. The captured pattern is input to the input / output unit 1 and the case recording unit 2.

Next, the input / output means 1 outputs and displays the fetched pattern. Then, the operator sees the pattern information of the controlled object displayed on the input / output means 1 and operates the input / output means 1 if necessary to input the operation amount U for the controlled object. , This is given to the control target as a command (step St6 in FIG. 14).
2, St63). The operation amount U is an appropriate operation amount determined by a skilled operator in accordance with the state of the control target S described above.

On the other hand, the manipulated variable U output from the input / output means 1 is sent to the case recording means 2 in addition to the controlled object (FIG. 1).
4 step St64).

As a result, the state S of the controlled object and the manipulated variable U are given to the case recording means 2. The state S of the control target is label information, and the input / output unit 1 gives this label.

As a result, the case recording means 2 takes in a pattern of measured values from the outside and also takes in the state S of the controlled object and the appropriate manipulated variable U from the input / output means 1 and records them. Further, the order relationship between the recording one step before and the current recording is also recorded (step St65 in FIG. 14).

By repeating this until the learning mode is released (step St66 in FIG. 14), the actual information of the control operation performed by the operator for the state of the controlled object can be collected. If the operator determines that the case has been sufficiently sampled, at that stage the operator operates the input / output means 1 to end the learning mode and sets the system to the automatic operation mode.

As a result, the system starts the automatic operation reflecting the empirical knowledge of the skilled operator stored in the case recording means 2. The basic operation of this automatic operation has already been described in detail in the first embodiment, so it will be referred to and will not be described again here.

Next, the operation of the recording maintenance function newly added in the third embodiment will be described.

[Processing Flow for Recording Maintenance] When information maintenance is carried out, an editing command is given to the recording maintenance means 7 to activate it. Then, the record maintenance means 7 is shown in FIG.
The process of (b) is implemented. That is, when the edit command is received (step St71 in FIG. 14B), the record maintenance means 7 gives a read command to the case recording means 2 so that the case recording means 2 can retrieve information necessary for maintenance.

As a result, the information S of the state S of the controlled object, the manipulated variable U for the controlled object, and the corresponding information of the previous step, which are held, are extracted from the case recording means 2 (FIG. 14 (b)). Step St72). Upon receiving this, the record maintenance means 7 takes in the data, and changes the record in accordance with the editing instruction (step St73 in FIG. 14B).
That is, of the information obtained from the case recording means 2, unnecessary information is deleted, and information requiring change is corrected. If there is a change, the case recording means 2 is rewritten and controlled to be updated, and the information is updated.

As a result, of the information accumulated in the case recording means 2, unnecessary information is removed, and information that needs to be corrected is corrected and used for the subsequent automatic operation. It is possible to perform a rational wasteless operation amount generation process.

The recording maintenance means 7 can be applied to the apparatus of the second embodiment as it is, and the action and effect in that case are also the same as those of the second embodiment.

As described above, in the control device of the third embodiment, the control device of the first or second embodiment is further loaded with the case data or the like recorded in the case recording means, and these are registered, The configuration is such that a record maintenance means for correcting and deleting is added, and in such a configuration, the record maintenance means is configured to record the state of the control target recorded in the case recording means and the operation amount for the control target. Correspondence is taken in and new registration and deletion of these records are performed. Further, in the control device of the third embodiment, in addition to these operations, the same operation as that of the control device of the first or second configuration is performed. Therefore, if there is an error in the rule of thumb of the operator stored in the case recording means, the error can be removed.

The above is a description of the basic concept of the present invention. Then, what will happen when it is applied to an actual device will be described below by giving concrete examples.

(Specific Example) In order to explain the present invention,
The application to specific cases will be described. This case is ventilation control in a highway tunnel. In tunnel ventilation, maintenance and management in pursuit of economic efficiency is an issue. The purpose of control is to provide a safe driving environment, more specifically, visibility is deteriorated by soot in the tunnel, air pollution concentration due to harmful substances such as carbon monoxide, and at an acceptable range with minimum cost. It is to fit. As a tunnel ventilation system, in recent years, a longitudinal flow system that ventilates a tunnel in a longitudinal direction is mainly used. Since the wind speed in the tunnel fluctuates according to natural conditions such as vehicle volume and seasonal winds, the diffusion process of soot is affected by disturbance. To control tunnel ventilation,
Conventionally, sequence control has been used such as "if the carbon monoxide concentration exceeds a threshold value, increase the number of blowers in operation and the number of revolutions". However, such a control system is wasteful, and it has been required to automate the economical control by a machine, which an experienced operator uses empirical knowledge.

FIG. 15 shows a tunnel TN to which the present invention is applied.
It is a cross-sectional view of the vertical direction. As shown in FIG.
2 horizontal blowers (HB1, HB2) and dust collector 2 as ventilation equipment
A group (QC1, QC2) and two vertical blowers (VB1, VB2) are installed. Horizontal blowers HB1 and HB2 are installed to increase the wind speed in the tunnel TN. Dust collectors QC1 and QC2 purify soot and smoke in the tunnel TN. Horizontal blowers HB1 and HB2 and dust collectors QC1 and QC2 are installed on the ceiling of the tunnel TN. The vertical blowers VB1 and VB2 are
In particular, in order to secure visibility in the lower part of the tunnel and to keep the concentration of carbon monoxide below a certain level, it is installed by being embedded in the road surface in order to diffuse pollutants that easily fall and settle to the upper part of the tunnel. In addition to ventilation equipment, in order to monitor the soot state in the tunnel TN, a total of 6 monitoring cameras (C1L, C2L
, C3L, C1U, C2U, C3U) are installed.

The operator (operator) uses the monitoring cameras C1L and C2.
From the images captured by L, C3L, C1U, C2U, and C3U, the soot concentration at each location, the direction of smoke flow, the flow velocity, etc. were judged, and the current situation of the entire tunnel TN was comprehensively grasped and in the near future. After predicting the change of the situation of 1, the increase / decrease of the rotation speed of the blower and the operation / stop / decrease of the rotation speed of the dust collector are properly instructed to control the ventilation in the tunnel TN.

Skilled operators can monitor cameras C1L and C2L
, T that can be monitored by C3L, C1U, C2U, and C3U
Economical control can be performed by making full use of not only the local state in N but also the knowledge about the ventilation characteristics of the entire tunnel TN obtained from the operating experience of the ventilation equipment.

For example, “At a certain point, if the density of soot at the point where it can be seen from the camera is high at all 6 points, if the density immediately before that is relatively low, it is a transient phenomenon, so it is a blower. It is not necessary to increase the number of rotations of "."

The knowledge obtained from such experience can be converted into the if-th in the programming language once it can be verbalized (programmed into a verbal language) as in the above-mentioned example shown in brackets.
It can be expressed in the form of en rule and an automatic control device can be constructed by the technique of expert system. However, such knowledge is rarely held in a logically organized form, and generally it is extremely difficult to extract knowledge from a driver (knowledge acquisition).

Therefore, it is difficult to automate the tunnel ventilation control by applying the conventional technique such as the expert system. The present invention can automatically acquire the empirical knowledge of skilled operators.

FIG. 16 shows a block diagram in which the present invention is applied to the ventilation control device of FIG. Control signals to and from the ventilation equipment, and monitoring cameras C1L and C2 installed in the tunnel TN
Images from L, C3L, C1U, C2U, and C3U are transmitted via a bus (communication line). The console A and the control operation computer B are the input / output means 1 in FIGS. 1, 7, and 13.
Equivalent to the soot state case database C and tunnel TN
The internal state transition recording database D corresponds to the case recording means 2, the neural network E corresponds to the state estimating means 3,
The control device F corresponds to the control path determining means 4 and the manipulated variable determining means 5. And each function is basically as shown in FIG.
This is as described in FIGS. 7 and 13.

Next, the operation will be described.

[Processing when collecting a case] First, a process of collecting control knowledge of an operator will be described. The images sent from the six cameras C1L to C3L monitoring the lower part of the tunnel and the cameras C1U to C3U monitoring the upper part of the tunnel are
It is input to the control operation computer B.

This image shows the tunnel TN to be controlled.
It is a pattern of the measured value showing the soot state of. Tunnel T
An image showing a soot state of a key portion in N is displayed on the monitor television connected to the console A operated by the operator by the control operation computer B. The operator determines an appropriate operation command by referring to the image showing the soot state, and inputs the determined operation command by operating the keyboard connected to the console A. The input operation command is sent to each ventilation device and executed.

This operation command corresponds to the operation amount U. The operator specifies an appropriate state name (label) for the current soot state through the keyboard of the console A. Here, for example, Xij is used, and numbers are assigned to i and j.

The computer B for control operation registers a pair of the image Pij in the soot state and the name Xij of the state S in the soot state case database C. Furthermore, the state transition record database D in the tunnel is formed by combining the state one hour before, the current state, and the operation command U instructed by the operator one hour ago.
Register with.

By repeating this, the soot-state case database C accumulates examples of soot-state images and states. Further, in the in-tunnel state transition record database D, a state transition record is stored in which a transition to another state is made by adding an appropriate operation command U in a certain state.

FIG. 18 shows a part of the record accumulated in the tunnel state transition record database D. For example, it is shown that the state 1 is changed to the state 2 by adding the manipulated variable U _1,2 to the ventilation equipment. As shown in this figure, the operation for making a transition from one state to another is a state transition record.

The soot state images and state examples accumulated in the soot state example database C are used for learning of the neural network E. FIG. 17 shows the structure of the neural network C. In this figure, circles represent neuron elements, and arrows connecting the circles represent connecting lines between neuron elements. The connecting lines connect between all the neuron elements in the input layer and all the neuron elements in the middle layer, and between the neuron elements in all the middle layer and all the neuron elements in the output layer. Because of this, only a part of it is drawn. In the figure, symbol DU _j is a neuron element in the input layer, DU _i is a neuron element in the intermediate layer, DU _k is a neuron element in the output layer, DX _j is a grayscale value DU _j on the pixel of the synthesized image. Input to W _ij is DU _j and D
A coupling weight value that is the signal transmission efficiency on the coupling line connecting U _i , W _ki is a coupling weight value that is the signal transmission efficiency on the coupling line connecting DU _i and DU _k , and a _k is the input of the input layer of the neural network. Represents the reaction intensity value of DU _k for the captured image. The neural network E includes, for example, Japanese Patent Application No. 4-
Those described in the specifications of No. 45614 and Japanese Patent Application No. 5-102980 may be used.

The learning procedure may be the same as that described in the present invention. Therefore, a detailed description will not be given here, and only a brief description will be given. This neural network has a three-layer structure, and the grayscale value of each pixel of an image obtained by arranging six images showing the state of soot in a horizontal row and combining them is input to each element of the input layer.

That is, the input layer is composed of the same number of elements as the number of pixels of the combined image. The image information input to the input layer is propagated to the element of the intermediate layer through the coupling line,
Further, it propagates to the element of the output layer. The elements of the output layer include the same number of elements as the number of state names, and each element has a one-to-one correspondence with a state name. As a result of the propagation of the reaction, the state name corresponding to the element of the output layer that reacted is the hypothesis of the soot state of the tunnel.

When an image is input to the input layer of this neural network, a state appears in the output layer. At the same time, the certainty factor of the state of the image is obtained as the reaction strength of the element representing each state.

In the learning process, the elements in the output layer are
Information is given that specifies the correct response of the output layer elements, called the teacher signal. The teacher signal is information indicating the state name for the image input to the input layer, and is given to correct the error in the estimated state value for the image that appears in the output layer of the neural network. The learning of the neural network is repeated for all the case images accumulated in the soot state case database until a signal matching the teacher signal appears as a reaction of the element of the output layer of the neural network. Learning here means to correct the weights on the coupling lines of all the elements so that an output that matches the teacher signal appears.

[Processing at the time of control] Next, the process of performing ventilation control of the tunnel using the ventilation control device of which the example has been learned as described above will be described.

A pattern consisting of six images showing the soot state is sent from the surveillance cameras C1U to C3U, C1L to C3L and presented to the input layer of the neural network E through the bus. The neural network E calculates and outputs the state corresponding to the pattern and its certainty factor. Here, in general, it is rare that an image that exactly matches the image has been experienced in the past. At this time, the neural network E outputs a plurality of state hypotheses of images learned in the past that are close to the current pattern. Therefore, note that there are multiple pairs of state hypotheses and certainty factors.

The control device F reads the state and its certainty factor from the neural network E, and reads the target state T designated by the operator via the console A and the control operation computer B. Next, the control device F searches the tunnel state transition record database D, and determines the target state T from the current state hypothesis.
Read all records of state transitions, which are the routes leading to.

For example, it is assumed that the records registered in the tunnel state transition record database D are as shown in FIG. When the current state hypotheses output by the neural network E are "9" and "12", and the respective confidence factors are "0.6" and "0.2" and the target state T is "14", the target The route to the state T is "9"->"10"->"14""12"->"13"->"14""12"->"15"-> There are 3 routes of "16"->"14". The control device records these as control route candidates.

Next, the control device F determines the operation amount for the ventilation equipment. The operation amount collectively represents the increment direction of the operation amount for each facility e. All of the horizontal fan, dust collector, and vertical fan that are the ventilation equipment of the applicable tunnel have the number of rotations of the motor as the operation amount, and the operation amount U _{a, b} for transitioning from the state a to the state _b is 6 units. Motor V
B1, VB2, HB1, HB2, QC1, QC2 rotation speed increment direction U _{a, b} = (ΔVB1 _{a, b} , ΔVB2 _{a, b} , ΔHB
1 _{a, b} , ΔHB2 _{a, b} , ΔQC1 _{a, b} , ΔQC2 _{a, b} ). Here, the values taken by the respective components of U _{a, b} are NB, ZZ, PB, which “reduce” the rotation speed of the motor,
It corresponds to "maintain" and "increase". The relationship between the direction of the motor revolution speed increment and the actual value of the revolution speed increment is shown in FIG.
It is determined by the membership function shown in FIG.

Since the state hypotheses are "9" and "12",
The control device determines the rotation increment from the operation commands U _9,10 and U _12,13 for transitioning to the next states 10 and 13 on the two paths. In these two operation commands, the vertical blower VB1
Will be described as an example. It is assumed that the command relating to VB1 of U _9,10 is NB, that is, the rotational speed is “decreasing”, and the command relating to VB1 of U _12,13 is PB, that is, increasing the rotational speed. At this time, the rotation speed increment of VB1 is obtained by the centroid method in fuzzy inference, as shown in FIG.

At this time, the certainty factor “0.6” of the state hypothesis “9” and the certainty factor “0.2” of the state hypothesis “12” are used as the grade of each membership function. 5 remaining
After determining one motor speed increment, change the speed of each ventilator.

Next, the image from the surveillance camera is presented to the input layer of the neural network E again. As a result, the current state hypotheses are “13” and “16”, and the certainty factors are “0.
8 "and" 0.2 ". The control device F sets the route candidates" 9 "->" 10 "->" 14 "," 12 "->" 13 "that were established one time earlier. Of the “−−>“ 14 ”, the first route from the state“ 9 ”, which did not reach the state“ 10 ”expected as the next state, is excluded from the route candidates. , Is determined only from the operation command U _13, 14 from the state “13” to the state “14.” The determination is performed by fuzzy inference as in the case of one time ago.

The ventilation device is controlled by repeating the above until the target state T is reached.

Next, as another example, an example in which the state transition of the plant has dynamics will be described.

(Operation when plant state transition has dynamics) Next, according to the present invention, an explanation will be given of the point that appropriate control can be performed even when the plant state transition has dynamics. FIG. 21 shows a tunnel N
7 shows a part of the routes accumulated in the tunnel state transition record database D when the state transition of the soot of T has dynamics. Circles indicate states, symbols inside circles indicate state names, and arrows indicate transitions between states. The character string enclosed by "(", ")" attached to the arrow means an operation command that causes a transition.

The operation command is expressed by combining the name of the ventilation equipment for changing the rotation speed and the symbols "+" and "-" indicating the direction of increment. For example, such as (VB1 +, HB1 +), which means increasing the rotational speeds of the vertical blower VB1 and the horizontal blower HB1. Further, (0) indicates that the rotation speed is maintained for all ventilation equipment.

Further, the alphabetical string of 2 rows and 3 columns above the circle indicating the state is an observed controlled variable, that is, a character string representing the state of the image corresponding to the state captured by 6 cameras. is there. The upper row shows the images of the three cameras C1U to C3U installed at the top of the tunnel, and the lower row the 3 installed at the bottom.
The image of one camera C1L-C3L is shown. The alphabet is
"L" is low soot concentration, "m" is medium, "h"
Means high. Actually, a set of 6 images is the controlled amount, but here, for simplification of description, instead of showing 6 images, the images are graded into 3 levels by the soot concentration and the degree thereof is shown. The controlled variable will be represented by a set of letters.

Further, the information actually read by the operator from the image includes not only the soot concentration but also the smoke flow direction and the wind speed, but this is not dealt with in this description for the sake of simplicity. .

At time k, the six cameras C1U, C
The images obtained from 2U, ..., C3L are assumed to be mmmhh. The neural network E reads this image and outputs "a", "1", "A" as the state hypothesis. The control device F starts from these three states from the in-tunnel state transition record database D, and has three routes from the three states to the given target state "a"->"b"->"ha""1"-->"2"->"3""A"->"B"->"C" is searched. The three paths are shown by solid lines in FIG. These routes are control routes which are respectively in order from the top of the figure when the vehicle amount decreases, when the vehicle amount is constant, and when the vehicle amount increases. Please note that it is not possible to determine if there is a situation.

The control device F then determines the operation amount by the fuzzy inference described above and controls the ventilation device. In this case, since the operation command of any path is (0), the increments of the rotation speed for all ventilation equipments are all 0.

It is assumed that the image obtained from the camera is hhhhhh at time k + 1, which is one time later. The neural net reads this image and outputs "2" and "B" as the state hypothesis. Of the three candidates for the route that are currently present, the control device F determines that the state next to the state one hour before is
Routes "2" and "B" that are not "B" are excluded from the route candidates.

The controller F then proceeds to state hypothesis "2",
The certainty factor of "B" and the operation command (HB1 +,
HB2 +), route "A"->"B" operation command (VB
1+, VB2 +, QC1 +, QC2 +), the operation amount is determined by the fuzzy inference described above. In this explanation, a specific numerical value of the manipulated variable is not necessary and will not be described. Control each ventilator based on the fixed operation amount.

At time k + 2, it is assumed that the image obtained from the camera is hhhmhm. The neural net reads this image and outputs "C" as the state hypothesis. The control device selects the route “2” that is not the state “C”, which is the next state one time before, from the two routes that are present.
Remove “3” from the route candidates.

As shown in the above example, when the plant state transition has dynamics, the plant state cannot be identified only by observing the instantaneous controlled variable at a certain time. In the above example, the controlled variable mm at time k
Although there are three possibilities for states corresponding to mhhh, the present invention makes it possible to refine the hypothesis of states gradually. The present invention can perform appropriate control by handling the time series of states, that is, state transitions, even when the state of the plant cannot be identified from the value of the instantaneous controlled variable observed at a certain time. .

The above is the details of the embodiment of the present invention.

As described above, the function of directly handling the data, such as images, which the operator refers to during control as a controlled variable, the function of acquiring the control knowledge of the operator, and the generation of the best control path from the acquired knowledge. By having the function and the function together, it becomes possible to take in the control knowledge of a skilled operator and automatically perform the control according to the optimum control route.

The present invention is not limited to the above-described embodiments, but various modifications can be made without departing from the scope of the invention. For example, although a neural net is used as the state estimating means, any means may be used as long as it is a means for classifying patterns. For example, a statistical pattern recognition means based on principal component analysis (for example, a composite pattern A similarity method) or an arbitrary clustering method may be adopted.

Further, although the tunnel ventilation control is targeted in the present embodiment, in addition to this, the plant control device described in the present embodiment can be applied to the temperature inside the blast furnace, the sewage sludge, the steel plate rolling shape, the traffic signal, etc. It can be easily inferred that the same effect can be obtained.

The present invention will be summarized as follows. Firstly, it is composed of an input / output means, a case recording means, an operation amount determining means, a control route searching means, and a state estimating means.
Secondly, in addition to the first configuration, a timekeeping means is provided, and the time information of this timekeeping means is used so that it is possible to focus on and refer to past cases in which temporal elements approximate. And, thirdly, it is configured so that the collected cases can be edited and operated.

In the control device of the first configuration, the pattern of the measured value relating to the controlled object, which is the monitoring information, is taken into the input / output means and presented to the operator of the controlled object, and then the input / output means uses the pattern. The state of the controlled object estimated by the operator and the operation amount in the present state determined by the operator are taken in. This manipulated variable is added to the controlled object. Subsequently, the case recording means captures the pattern, the state and the operation amount determined by the operator, and records these and their corresponding relationship. By repeating such operations, the empirical knowledge of the operator regarding the control of the target is acquired as a correspondence relationship between various states of the control target, the monitoring information in the state, and the operation amount suitable in the state. I did it.

Then, in this control device,
The control target is automatically controlled by using the accumulated empirical knowledge of the operator. That is, when the pattern of observed values regarding the controlled object is input to the state estimating means, the state estimating means selects a pattern close to the currently observed pattern from the past patterns stored in the case recording means,
The state recorded corresponding to the selected pattern is used as a hypothesis of the current state of the control target. At the same time, the degree of similarity between the currently observed pattern and the selected past pattern is taken as the certainty factor of the state hypothesis. Subsequently, the control path searching means takes in the record stored in the case recording means, searches for a record corresponding to the current state hypothesis of the control target supplied by the state estimating means, and controls to reach the target state of the control target. Search for route candidates. Next, the manipulated variable determining means
The control route candidate supplied by the control route searching means and the certainty factor supplied by the state estimating means are taken in to calculate the operation amount for the controlled object.

Therefore, according to this configuration, in the control of the target in which it is not clear which information included in the pattern is used by the operator as the control amount to determine the operation amount,
It is possible to control the controlled object semi-automatically by acquiring the empirical rule regarding the control of the object without performing the difficult work of describing the empirical rule of the operator in the rule format. Furthermore, the logical structure of the empirical rules, such as the order of application of the empirical rules concerning the control of the operator, can be automatically obtained by observing the case of the control by the operator.

In the control device having the second configuration,
The pattern of measured values regarding the controlled object is captured by the input / output means and presented to the operator under the controlled object. The input / output unit takes in the state of the controlled object estimated by the operator by looking at the pattern and the operation amount in the present state determined by the operator. This manipulated variable is added to the controlled object. Subsequently, the case recording means captures the pattern, the state and the operation amount determined by the operator, and further captures the current time from the clock means. The time, the time elapsed from one step before to the present, the pattern, the state, the operation amount, and the correspondence between these are recorded. By repeating the above operation, the empirical knowledge of the operator regarding the control of the object that changes depending on time can be used to determine the time, elapsed time, various states of the controlled object, patterns in that state, and suitable operations in that state. Acquired as a correspondence with quantity. The control device automatically controls the controlled object as described below, using the empirical knowledge of the operator acquired and accumulated in this way. The pattern of observed values regarding the controlled object is input to the state estimating means. Next, this state estimation means
From the patterns stored in the case recording means, a pattern close to the currently observed pattern is selected, and the state recorded corresponding to the selected pattern is set as the current hypothesis to be controlled. At the same time, the degree of similarity between the currently observed pattern and the selected pattern is used as the certainty factor of the state hypothesis. Subsequently, the control path searching means takes in the record stored in the case recording means, searches for a record corresponding to the current state hypothesis of the control target supplied by the state estimating means, and controls to reach the target state of the control target. Search for route candidates. Next, the manipulated variable determining unit takes in the control route candidates supplied by the control route searching unit and the certainty factor supplied by the state estimating unit, and calculates the manipulated variable for the controlled object.

Therefore, in addition to the effect of the device having the first configuration, the same result is obtained when the controlled object is in the same state due to the difference in time and the difference in the time elapsed from one step before to the present. Even if a control target has complicated dynamics that transits to different states even if an operation amount is added, a hypothesis of the state of the control target is generated from the instantaneous pattern measured at a certain time, and the generated state From the hypothesis, the appropriate manipulated variable can be determined.

Further, in the control device of the third configuration,
The record maintenance means fetches the state of the controlled object and the operation amount and the correspondence relationship with respect to the controlled object recorded in the case recording means, and newly registers and deletes these records. In addition to these operations, operations similar to those of the control device having the first or second configuration are further performed.

Therefore, if there is an error in the empirical rule of the operator accumulated in the case recording means, the error can be removed.

[0190]

According to the present invention, it is possible for an operator to acquire the control knowledge of a plant that has been accumulated empirically and control the plant based on that. Moreover, efficient control can be performed by adjusting the acquired control knowledge and generating an optimum control path.

[Brief description of drawings]

FIG. 1 is a diagram for explaining an embodiment of the present invention,
The block diagram for demonstrating the 1st Example of this invention.

FIG. 2 is a diagram for explaining an embodiment of the present invention,
3 is a flowchart showing the flow of processing when recording a control case according to the first embodiment of the present invention.

FIG. 3 is a diagram for explaining an embodiment of the present invention,
The flowchart which shows the flow of the process at the time of control in the 1st Example of this invention.

FIG. 4 is a diagram for explaining an embodiment of the present invention,
FIG. 4 is a partially enlarged view of the flowchart of FIG. 3 showing the flow of processing at the time of control in the first embodiment of the present invention.

FIG. 5 is a diagram for explaining an embodiment of the present invention,
FIG. 4 is a partially enlarged view of the flowchart of FIG. 3 showing the flow of processing at the time of control in the first embodiment of the present invention.

FIG. 6 is a diagram for explaining an embodiment of the present invention,
FIG. 4 is a partially enlarged view of the flowchart of FIG. 3 showing the flow of processing at the time of control in the first embodiment of the present invention.

FIG. 7 is a diagram for explaining an embodiment of the present invention,
The block diagram for demonstrating the 2nd Example of this invention.

FIG. 8 is a diagram for explaining an embodiment of the present invention,
The flowchart which shows the flow of a process at the time of control case recording in the 2nd Example of this invention.

FIG. 9 is a diagram for explaining an embodiment of the present invention,
The flowchart which shows the flow of the process at the time of control in the 2nd Example of this invention.

FIG. 10 is a diagram for explaining the embodiment of the present invention and is a partially enlarged view of the flowchart of FIG. 3 showing the flow of processing at the time of control in the second embodiment of the present invention.

FIG. 11 is a diagram for explaining the embodiment of the present invention and is a partially enlarged view of the flowchart of FIG. 3 showing the flow of processing at the time of control in the second embodiment of the present invention.

FIG. 12 is a diagram for explaining the embodiment of the present invention and is a partially enlarged view of the flowchart of FIG. 3 showing a flow of processing at the time of control in the second embodiment of the present invention.

FIG. 13 is a diagram for explaining the embodiment of the present invention and is a block diagram for explaining the configuration of the third embodiment of the present invention.

FIG. 14 is a diagram for explaining the embodiment of the present invention and is a flowchart showing the flow of processing at the time of recording a control case in the third embodiment of the present invention.

FIG. 15 is a diagram for explaining an embodiment of the present invention, and is a diagram for explaining the present invention by applying it to a plant as an actual case.

FIG. 16 is a diagram for explaining the embodiment of the present invention and is a diagram showing an example of a plant as an example to which the present invention is applied.

FIG. 17 is a diagram for explaining the embodiment of the present invention and is a diagram showing the structure of a neural net which is an embodiment of the state estimating means shown as a constituent element of FIG. 16;

FIG. 18 is a diagram for explaining the embodiment of the present invention, and is a diagram showing an example of the control route recorded in the control route database means of the present invention shown as a constituent element of FIG. 16;

FIG. 19 is a diagram for explaining the embodiment of the present invention and is a diagram showing an example of a membership function used when the control unit shown as a constituent element of FIG. 16 determines an operation amount.

FIG. 20 is a diagram for explaining the embodiment of the present invention and a diagram for explaining a method for determining a manipulated variable in an application example.

FIG. 21 is a diagram for explaining the embodiment of the present invention and is a diagram showing an example of a route when the state transition has dynamics.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input / output means 2, 2a ... Case recording means 3, 3a ... State estimation means 4 ... Control path search means 5 ... Manipulation amount determination means 6 ... Timekeeping means 7 ... Recording maintenance means

Claims (8)

[Claims] 1. A monitoring information collecting unit for collecting monitoring information about a control target, monitoring information collected by the monitoring information collecting unit, status information of a control target for the monitoring information, and an operator for the status information. Case record means for recording case data represented by a set of operation amount based on an instruction given from the above together with time-series order information for each case, and case data recorded in the case record means. Out of
Between the status information for the other monitoring information and the status information for the extracted monitoring information, the monitoring information having a predetermined degree of approximation to the other monitoring information collected by the monitoring information collecting means is extracted. And a state information for the monitoring information extracted by the state estimating unit, and a target state of a control target that is separately given, by referring to the case data recorded in the case recording unit. Control path generation means for tracing the order information to extract case candidates that can reach the target state, the certainty factor obtained by the state estimation means, and the control path generation means. The desired state information is selected based on the selected case candidates, and the operation amount for the case data for the selected state information is given to the control target. Control apparatus characterized by comprising an operation amount determining means that. 2. A clocking means for generating time information, a monitoring information collecting means for collecting monitoring information about a control target, monitoring information collected by the monitoring information collecting means, and status information of a control target for the monitoring information. A case recording means for recording case data represented by a set of an operation amount based on an instruction given by an operator for the status information and the time information, in time series for each case; Of the case data recorded in the recording means,
Between the status information for the other monitoring information and the status information for the extracted monitoring information, the monitoring information having a predetermined degree of approximation to the other monitoring information collected by the monitoring information collecting means is extracted. And a state information for the monitoring information extracted by the state estimating unit, and a target state of a control target that is separately given, by referring to the case data recorded in the case recording unit. Control route generation means for tracing the time information and extracting case candidates capable of reaching the target state, the certainty factor obtained by the state estimation means, and the control route generation means. The desired state information is selected based on the selected case candidates, and the operation amount for the case data corresponding to the selected state information is given to the control target. Control apparatus characterized by comprising a manipulated variable determining means. 3. Record maintenance for fetching case data recorded in the case recording means, performing editing processing such as registration, correction and deletion of this case data, and updating the case recording means with information after this editing processing. The control device according to claim 1, further comprising means. 4. The control device according to claim 1, wherein a neural net is used as the state estimating means. 5. The control device according to claim 1, wherein the status information is monitoring image information of a control target. 6. The control device according to claim 1, wherein the state information is pattern information in which physical information detected from each part of the controlled object is arranged. 7. A collecting step of collecting monitoring information about a controlled object, monitoring information collected by the collecting step, status information of a controlled object for the monitoring information, and an operation based on an operator's instruction for the status information. A case recording step of recording case data expressed as a set with a quantity together with time-series order information for each case, and of the case data recorded in this case recording step, collected by the collecting step. State estimation that extracts the monitoring information having a predetermined degree of approximation to other monitoring information and obtains a certainty factor of agreement between the state information for the other monitoring information and the state information for the extracted monitoring information Step, and referring to the case data recorded in the case recording step, the state information for the monitoring information extracted in the state estimating step is referred to. And a control path extraction step of extracting the case candidates that can reach the target state by tracking the order information based on the information about the target state of the control target that is separately provided, and the state estimation step. An operation amount determining step of selecting desired state information based on the certainty factor and the case candidates extracted by the control path extracting step, and giving an operation amount of the case data corresponding to the selected state information to the control target. A method for controlling an object to be controlled, comprising: 8. A time measuring means step for obtaining time information, a collecting step for collecting monitoring information about a control target, monitoring information collected by this collecting step, status information of a control target for the monitoring information, and the status A case recording step of recording, in time series for each case, case data represented by a set of an operation amount based on an operator's instruction for information and the time information, and a case recorded in this case recording step. Of the data, the monitoring information having a predetermined degree of approximation to the other monitoring information collected in the collecting step is extracted, and the status information for the other monitoring information and the status for the extracted monitoring information are extracted. A state estimating step of obtaining a certainty factor of coincidence with information, and referring to the case data recorded in the case recording step, the state estimating step Based on the status information for the monitoring information extracted by, and information about the target state of the control target that is separately provided, a control path extraction step of tracking the time information and extracting a case candidate that can reach the target state, The desired state information is selected based on the certainty factor obtained in the state estimation step and the case candidates extracted in the control path extraction step, and the operation amount for the case data for the selected state information is controlled. And a step of determining a manipulated variable given to the control object.