JPH05204406A - Process controller - Google Patents

Abstract

PURPOSE:To execute process control with high precision by selecting optimum knowledge and an optimum method from plural different kinds of knowledge and plural processing methods so as to decide controlled variable. CONSTITUTION:The neural network 170 of a knowledge selecting means 140 is made to learn so as to recollect the one kind or more than two kinds of knowledge which is appropriate for deciding controlled variable at some point of time based on the operation condition of process. The neural network 205 of a knowledge processing method deciding means 180 is made to learn soas to recollect the one kind or more than two kinds of knowledge processing methods which is appropriate for deciding controller variable at some point of time based on the operation condition of process. The knowledge selecting means 140 and the knowledge processing method deciding means 180 permit the respective neural networks 170 and 205 which are made to learn to recollect so as to select knowledge and to decide the knowledge processing method. Thus, process control with high precision is executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process control device for monitoring and controlling a process having a change over time, and in particular,
From the knowledge base group that has multiple types of knowledge, follow the changes in process operating conditions, select the most effective type of knowledge at that time, and process the knowledge processing method that handles it. The present invention relates to a process control device capable of optimally controlling a process while selecting from a method base.

[0002]

2. Description of the Related Art Water purification process, sewage treatment process,
In process control such as chemical processes, there is a method in which an operator's empirical rule is converted into a fuzzy rule and a fuzzy control is performed by using an inference mechanism. For example, JP-A-2-261506 and JP-A-2-18719.
This example is disclosed in gazettes such as No. 3 and the like.

Since the control by the fuzzy engineering is performed based on the fuzzy rule and the membership function obtained by the interview from the operator, various controls can be performed with the reasoning ability equivalent to that of the operator. It is possible. It is also a fact that the inference process by fuzzy rules and membership functions is logically clear, and the results obtained are sufficiently satisfactory for the operator. However, on the other hand, there is also a problem that acquisition of the above-mentioned rule of thumb and maintenance are very complicated.

As a solution to these problems, Japanese Patent Laid-Open No. 1-
In the technique disclosed in Japanese Patent No. 224804, a control applying a neural network model is introduced. In this model, the process operation history data
It is possible to determine the control amount by learning the knowledge inherent in the data and recollecting it in the model after learning. Therefore, it is possible to perform control in accordance with the process performance without using fuzzy rules.

Further, in the technique disclosed in Japanese Patent Laid-Open No. 63-200807, etc., not the empirical rule of the operator or the operation history data of the process, but the existing principle / law regarding the phenomenon in the process (for example, expressed by a mathematical expression). Applied physical model) or a control device (for example, a simulator) based on the control theory.

According to this method, since universal principles and rules are used, control is performed even when the empirical rule of the operator cannot be obtained sufficiently or even in the process where the history data immediately after the start of operation is insufficient. It is possible.

[0007]

The conventional process control device having such fuzzy rules, neural nets or physical models as knowledge has the following problems.

First, although each of the above-mentioned knowledge has operating conditions of the process in which the knowledge is most effectively utilized, the controller for handling each knowledge has a single device or a plurality of devices. Since it is operated as an integrated device with a fixed combination of, it is difficult to flexibly change the knowledge to be used according to each operating condition of the process.

The operating conditions described here are (1)
The state of the input pattern such as measurement information that affects the determination of the control amount, that is, the state that indicates how the process is operating, (2) the incidental environmental conditions related to the operation of the process, for example, It refers to the degree of accumulation of process operation history data, the breakdown of equipment used for process control, measuring sensors, and maintenance status.

Secondly, when a plurality of types of knowledge are used, the optimum or effective combination of the knowledge processing methods depends on the operating conditions of the process. The processing method of knowledge is either single or the combination is fixed.
It was not easy to change to an optimal knowledge processing method in response to changes in operating conditions.

In view of the above problems, an object of the present invention is to provide a process control device capable of flexibly selecting a plurality of different kinds of knowledge according to operating conditions and selecting the most effective knowledge. Especially.

Another object of the present invention is to determine the most effective knowledge processing method in accordance with changes in operating conditions when utilizing a plurality of types of selected knowledge. To provide a process control device.

[0013]

In order to achieve the above object, according to one aspect of the present invention, a process controller for monitoring and controlling a process having a change over time is used to determine a control amount. , One of a plurality of kinds of knowledge accumulating means and a plurality of kinds of knowledge accumulating by the knowledge accumulating means, which is suitable for determining a control amount at a certain time based on the operating condition of the process, or Knowledge selecting means for selecting two or more kinds of knowledge, means for accumulating a plurality of kinds of knowledge processing methods for determining a control amount, and plural kinds for accumulating means for accumulating knowledge processing methods And a knowledge processing method determining means for determining a knowledge processing method or a combination of the processing methods from the knowledge processing methods, the knowledge selecting means and the knowledge processing method determining means respectively being a neural network. Or a reasoning mechanism, and a process control device characterized in that the neural network is recalled or the reasoning mechanism is caused to infer based on the operating condition of the process to select knowledge and determine a knowledge processing method. Will be provided.

Further, the following means can be considered in order to make the control device of the present invention more effective.

First, the knowledge selecting means is a process control device provided with operating condition input means and knowledge priority calculating means for calculating the priority of knowledge to be used.

Further, the knowledge priority calculating means is a process control device realized by learning and recall by a neural network model.

Further, the knowledge processing method determining means is a process control device having operating condition input means and method priority calculating means for calculating the priority of the knowledge processing method.

Further, the method priority calculating means is a process control device realized by means of learning and recall by a neural network model.

The type of knowledge used is at least one of the data obtained from the process, the mathematical model for explaining various phenomena in the process, or the knowledge described in the rule format. The apparatus and the knowledge processing method can also configure at least a process control apparatus which is one of a learning and recall means of a neural network, a simulation means, and a knowledge inference means.

[0020]

In the knowledge accumulating means, a plurality of kinds of knowledge used for determining the controlled variable are accumulated in advance. Also,
In the means for accumulating the knowledge processing method, a plurality of kinds of knowledge processing methods used in determining the control amount are accumulated in advance.

The knowledge selecting means is one or more kinds suitable for determining the controlled variable at a certain time based on the operating conditions of the process from among a plurality of kinds of knowledge accumulated by the means for accumulating knowledge. Select the type of knowledge of.

The knowledge processing method determining means determines a knowledge processing method or a combination of processing methods from a plurality of types of knowledge processing methods accumulated by the means for accumulating knowledge processing methods.

The knowledge selecting means and the knowledge processing method determining means each include a neural network. The neural network of the knowledge selecting means is trained to recall one or more kinds of knowledge suitable for determining the controlled variable at a certain time point, based on the operating condition of the process. Further, the neural network of the knowledge processing method determining means is trained so as to recall one or more kinds of knowledge processing methods suitable for determining the controlled variable at a certain time point, based on the operating conditions of the process. The knowledge selecting means and the knowledge processing method determining means recall the learned neural network and select the knowledge and determine the knowledge processing method.

According to the present invention, a plurality of knowledge types of different types, which match the operating conditions of the process to be controlled,
It is possible to select the knowledge processing method and the knowledge processing method, thereby dynamically changing the most effective knowledge and the knowledge processing method in response to the operating conditions that change with time, and at least equal to or higher than that of the operator. Control can be performed with high accuracy.

[0025]

BEST MODE FOR CARRYING OUT THE INVENTION In the control of a process having a time-dependent change, the present invention follows the change of operating conditions and is used for determining a control strategy such as determination of a control amount, and a knowledge type and a knowledge processing method. It is possible to construct a process control device that can change and to be optimal. More generally, it is a process control device that can be applied to the control of a process that has the property that the combination of components changes with time. Therefore, the present invention has various types of processes that have a plurality of kinds of knowledge and have a change over time, for example, a water purification process, a sewage treatment process, a chemical reaction process, a bioprocess, a nuclear power process, a heat energy supply process, It can be applied to financial processes such as stock prices and exchanges.

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

FIG. 1 is a block diagram showing the overall construction of an embodiment in which the process control device according to the present invention is applied to a control device for a water purification process.

The water purification process to which the embodiment of the present invention is applied is shown in FIG.
And a group of measuring instruments M arranged in each part of the water purification facility F to measure the state thereof, a control means 50 for controlling the operation of the water purification facility F, and a control device 100 for giving a control amount to the control means 50. And a man-machine interface means I of the device to the operator.

In the water purification facility F, a landing well 5, a rapid mixing basin 10, a floc formation basin 15, a settling basin 20 and a filtration basin 25 are arranged, a coagulant tank 11 and a coagulant injection pump 12 It has a stirring blade 13, a stirrer 14 for driving the stirring blade 13, and a stirring paddle 16.

Here, the flow of the water purification process will be briefly described.

Raw water is led to the landing well 5 from a river or lake. In the quick mixing pond 10, the water in the landing well 5 is received, the coagulant in the coagulant tank 11 is injected by the coagulant injection pump 12, and the stirring blades 13 are driven by the stirrer 14.

In addition, an alkaline agent which promotes the formation of flocs may be injected.
An agitating paddle 16 is installed at 5, and is gently driven to rotate. The flocs settle in the settling tank 20, and the supernatant liquid is filtered in the filter tank 25.

Next, the measuring instrument group M will be described. A measuring instrument 5M is installed in the landing well 5 to measure the quality of raw water. The measurement items are water temperature, turbidity, alkalinity, pH,
Electrical conductivity, residual chlorine concentration, required chlorine amount, water amount, water level, etc.

A measuring instrument 15M is installed in the flock formation pond 15. The measuring instrument 15M includes an imaging means such as an underwater camera in addition to the items measured by the measuring instrument 5M.
A measuring instrument 20M is installed in the sedimentation tank 20. In addition, if necessary, the rapid mixing tank 10 and the filtration tank 25 have 5M,
The measuring instruments 10M and 25M similar to 15M are installed. These measurement items are the same as 5M and 15M described above.

The above measurement data is used for the control device 1 described later.
It is taken into the database 350 in 00.

The control means 50 is composed of, for example, a control computer and a PID controller, and the control device 10
The operation amount of the process is determined based on the control amount output from 0, and for example, the drive of the agitator 14, the coagulant injection pump 12, the agitation paddle 16 and the like described above is controlled.

The man-machine interface means I has a keyboard 95 as an input means and a CR as an output means.
It has a T display 90. Of course, in addition to this, input means such as a mouse and a touch panel, and output means such as a printer can be provided.

Next, a synergistic process control device 100 (hereinafter, simply referred to as "control device") for performing process control.
Will be described). The control device 100 is
Generally, it is built on a workstation different from the control computer. Although the hardware configuration is not particularly shown, for example, a workstation having a central processing unit, a main storage device, an auxiliary storage device, and various interfaces is used.

The control device 100 selects a database 350 for storing measurement data and input data, a knowledge selection means 140 and a knowledge base group 220 for selecting the type of knowledge to be used, and a knowledge processing method. The knowledge processing method determining means 180 and the knowledge processing method base 260, and the control calculating means 300 for performing the control calculation based on the obtained knowledge and the processing method thereof.

The database 350 stores the measurement data sent online from the measuring instrument group M and the data input from the keyboard 95. The latest stored data is sent to the knowledge selecting means 140 and the knowledge processing method determining means 180, and is used to select the most appropriate knowledge type and the knowledge processing method at that time. Further, it is sent to the control calculation means 300 and used for control calculation.

The knowledge selecting means 140 is a database 3
Referring to the first operating condition input means 150 for inputting operating conditions from 50, the latest data of the database 350 and the knowledge base group 220, one of the most suitable ones for determining the control policy at that time or It has a knowledge priority calculating means 160 for selecting two or more kinds of knowledge.

The knowledge priority calculating means 160 has a meta-neural network 170 and a meta-rule base 171 used for calculating the priority of knowledge to be used.
One of these can calculate the priority.
In this embodiment, two types of bases for calculating the priority are mounted and either one is used depending on the selection, but only one may be mounted.

The first operating condition input means 150 can be realized by using an electronic circuit such as a microprocessor and an interface, or an interface of a workstation.

The knowledge base group 220 is divided into a plurality of sub-knowledge bases for each type of knowledge, for example, a neuro sub-knowledge base, a rule sub-knowledge base, and a physical model sub-knowledge base. For example, the neuro sub-knowledge base stores a plurality of files of weighting coefficient matrices that will be described later, which form a model. In the case of the present embodiment, one file corresponds to one operating condition such as high turbidity and low turbidity. By using the parameters of this file, the operation amount such as the coagulant injection rate can be adjusted. It can be calculated.

Further, the rule sub-knowledge base stores a plurality of rules relating to the manipulated variable determination as shown in the following expression (1).

When the raw water turbidity is high, the coagulant injection rate is increased. If the raw water temperature is high, reduce the coagulant injection rate to a slightly lower level. (Equation 1) In the case of such a fuzzy rule, the membership function of the fuzzy variable is also stored in the rule sub knowledge base.

In the physical model sub-knowledge base, a model based on physical knowledge such as the following equation (2) is stored.

[0048]

[Equation 1]

The knowledge processing method determining means 180 includes the second operating condition input means 190 for inputting operating conditions from the database 350, the latest data of the database 350, the knowledge base group 220 and the knowledge processing method base 260. With reference to the method, the method priority calculating means 200 for selecting an optimum processing method for the knowledge selected by the knowledge selecting means 140.

The method priority calculating means 200 has a meta-neural network 205 and a meta-rule base 202 used for calculating the priority of the knowledge processing method to be used. One of these can calculate the priority. In this embodiment, two types of bases for calculating the priority are mounted and either one is used depending on the selection, but only one may be mounted.

The second operating condition input means 190 is, for example,
It can be realized by using an electronic circuit such as a microprocessor and an interface or an interface of a workstation.

The knowledge processing method base 260 registers a plurality of types of knowledge processing methods. An example of the registered knowledge processing method is shown.

{Fuzzy reasoning}, {Neural network recollection}, {Model simulation} {Correcting with neural network as fuzzy reasoning} Information on the kind of knowledge selected based on priority and knowledge processing method is as follows: It is sent to the control calculation means 300.

The control calculation means 300 calculates the control amount of the process based on the data indicating the state of the process stored in the database 350 by using the selected knowledge and the processing method of the knowledge, and the calculation is performed. The result is output to the control means 50 and the CRT 90.

Next, the input / output relationship of data and the like related to the control device 100 will be described. Inputs to the control device 100 include online input from the measuring instrument group M and manual input from the keyboard 95. The measuring instruments 5M and 10
M, 15M, 20M, 25M, etc. sample data at fixed time intervals. The sampling data is
It is transmitted to and stored in the database 350. In the database 350, the sampled data is stored in a format arranged in time series for each data item. In addition, hand analysis data and visual observation data that cannot be measured online with a measuring instrument are, for example,
The operator inputs data interactively through the keyboard 95 while referring to the CRT 90. This data is also stored in the database 350.

Next, the output from the control device 100 will be described. The control device 100 determines a control measure (for example, a control amount) based on the input data. Control device 1
The calculation result by the control calculation means 300 in 00 is the CRT.
Displayed to the operator via 90.

The output result of the control calculation means 300 is transmitted to the control means 50, and desired control is performed. In addition, by switching the control mode, automatic control,
It is also possible to set various operation modes such as semi-automatic operation via an operator, for example, using the control calculation result only as guidance information and manually setting the control amount.

Next, the knowledge selecting means 140 and the knowledge processing method determining means 180 will be described in detail.

The operation of the knowledge selecting means 140 will be described with reference to the flow shown in FIG.

First, in the operation condition input step 151, the first
The latest data regarding the process is fetched from the database 350 as the operating condition by the operating condition input means 150. The operating conditions taken in here are data indicating the operating state of the process itself (for example, raw water pH, raw water alkalinity, number of flocs, etc.) and incidental facts related to the operation of the process (years of operation, Sensor maintenance status, data storage amount, etc.).

The knowledge priority calculation step 161 can be realized by a neural network model, an inference mechanism, etc., and the operation condition input step 151 can perform data conversion depending on it.

In the knowledge priority calculating step 161, it is possible to calculate the priorities of a plurality of different types of knowledge registered in the knowledge base group 220. That is, the knowledge suitable for the process control at that time is given to the knowledge base group 22.
Select from 0.

When the knowledge priority calculating step 161 is realized by a neural network, it is desirable to have the steps shown in the flowchart of FIG.

Hereinafter, an operation example when the knowledge priority calculating step 161 is realized by a neural network will be described with reference to the flow shown in FIG.

First, in the automatic learning step 162, it is judged whether or not the latest operating condition fetched from the operating condition inputting step 151 is reflected in the neural network model for calculating the priority of knowledge, and it is necessary. If you decide,
Calculate (learn) to reflect.

Learning and recall of the neural network model used in the calculation here will be briefly described.

The basis of the model is the nerve cells (neurons) that make up the brain. A mathematical model of a single neuron that reflects the physiological knowledge of this neuron is widely used. The basic operation of a neuron is expressed by a function having a threshold characteristic of a multi-input / single-output system as shown in the following equations 3, 4, and 5.

Yt = F (Ut) (Equation 3) Ut = ΣWt, iXi (Equation 4) (where Σ means the sum of i) F (u) = 1 / {1 + exp (-u + Θ) } (Equation 5) where Yt is the output signal of the neuron T, Ut is the product sum of the inputs to the neuron T, Wt, i is the weight coefficient between the neurons T and I, and Xi is the input from the neuron I Signal, F is a sigmoid function, and Θ is a threshold of the sigmoid function.

The signal from a certain neuron to the focused neuron is multiplied by the coupling strength (also called weighting coefficient or synapse strength) Wt, i between the two. The sum Ut of these weighted input signals Wt, i · Xi is a threshold value Θ
, The neuron is excited and outputs the output signal Yt. At this time, it is the sigmoid function of Equation 5 that determines the threshold characteristic.

The neural net model is a network type model having the above-mentioned neurons as basic constituent elements. Although a plurality of models have been proposed, the present invention uses a hierarchical model (also referred to as Rumelhart model) in which neurons are connected in a hierarchical manner. As shown in FIG. 4, this model has a three-layer structure including an input layer, an intermediate layer, and an output layer.

The learning of the neural network model is performed by outputting the model output and the desired output (referred to as teacher data, generally,
It is performed by modifying the weighting coefficient so that the error with the history of operating conditions or the like) is reduced.

The conventional BP (back propagation) method, which is most frequently used as a learning algorithm, defines the sum of squares of recall errors as an evaluation function and modifies the weighting coefficient so that the evaluation function decreases. Is. For details of the algorithm, see Rumelhart, DE, etc (1986) “Lea
rning Representations by Backpropagating Errors '',
Nature, vol. 323, p533-536, etc.

The term "recall" as used herein refers to a calculation in which an input is applied to the input layer of the neural network model and an output value is output from the neuron in the output layer. The output value from the output layer neuron is the recall value. That.

In the automatic learning step 162, new learning for calculating the priority of knowledge is automatically performed using the above-mentioned neural net model. As an input of the model, for example, in the water purification process, various water quality data, image processing information of flocs, various event information (information about typhoons and rainfall, process operating years, process facility renovation, and extension information, etc.) as shown in FIG. ) Is placed.

Further, the type of the sub-knowledge base in the knowledge base group 220 (neuro, rule, physical model, etc.) is placed in the output.

The teacher data for learning here is created from the past history data stored in the database 350. Specifically, in typical operating conditions, data arranged in the input layer of the neural network model,
Creating multiple sets of sub-knowledge base types used at that time.

This preparation work is automatically performed based on a predetermined standard, but it can be given by an operator from the outside.

When the current data is input, the neural network model that learned the teacher data created in this way outputs the largest recall value from the output layer neurons corresponding to the most suitable sub-knowledge base.

The weighting coefficient matrix of the learned neural network model is stored in the meta-neural network base 170.

The automatic learning algorithm here corrects the learned neural network stored in the neural network base on the basis of the latest operating conditions, and has been already proposed by the inventors. Known materials can be used. For details of the algorithm, for example, “Study on Extraction of Plant Operating Rule Using Neural Network” (Proceedings of the Institute of Electrical Engineers of Japan D, Vol.111, N
o.1, p20-28).

Here, by performing the automatic learning, it is possible to always remember the priority of the knowledge with a good meta-neural net model even when the characteristics of the process change with time.

In the next priority recall process 164, the operating conditions are input to the neural network model obtained in the previous process 162, and the priority of each type of knowledge arranged in the output layer is calculated. In this case, the larger the output value, the higher the priority. The calculated priority is sent to the next knowledge selection step 166.

In the knowledge selecting step 166, the knowledge having a high priority is selected from the sub-knowledge bases in the knowledge base group 220 in which the actual knowledge is stored, and the control computing means 300 is selected.
Also, the knowledge information to be used is output to the knowledge processing method determination means 180.

In this step 166, a method such as selecting a predetermined number in a descending order of priority or higher priority may be considered.

Further, when the knowledge priority calculating means 160 is realized by the inference mechanism of knowledge processing, it is desirable to have the steps shown in the flowchart of FIG.

First, in the priority inference process 163, the data from the first operating condition input means 150 is used by referring to the rule regarding the priority determination stored in the meta-rule base 171 as shown in, for example, Expression 6. Thus, the priority can be calculated.

The inference method depends on the type of rule stored in the meta rule base 171. If the rule is a crisp rule, the reasoning mechanism is forward reasoning, and if the rule is fuzzy reasoning, the reasoning mechanism is fuzzy reasoning.

Expression 6 shows an example in which the meta rule is a fuzzy rule.

If the history data is small, the priority of the rule sub-knowledge base and the physical model sub-knowledge base is raised (Equation 6) In the knowledge selection step 165, the knowledge base group 220 in which the actual knowledge is stored is stored. From among the above, a predetermined number of sub-knowledge bases having a high priority or a sub-knowledge base having a priority higher than a threshold value are selected, and the knowledge information to be used is output to the control calculation means 300 and the knowledge processing method determination means 180 To do.

Next, the knowledge processing method determining means 180 will be described.

FIG. 6 is a flow chart showing the operation of the knowledge processing method determining means 180. Here, the knowledge processing method corresponding to the sub-knowledge base selected by the knowledge selection means 140 is selected. First, in the second operation condition input step 191, the latest operation condition of the process is fetched from the database 350. The type of data to be taken in is the same as that in the operation condition input step 151 described above.

In the next method priority calculation step 201, it is possible to calculate the priorities of a plurality of knowledge processing methods registered in the knowledge processing method base 260 and their combinations. Similar to the knowledge priority calculation step 161, this step 201 can be realized by a method using a neural network as shown in FIG. Of course, the inference based on the rules shown in FIG. 5 is also possible. Here, the knowledge base group 2
The details of the method are the same as those of the knowledge priority calculating step 161 described above, except that the knowledge processing method base 260 is referred to instead of 20.

For example, when a neural network is used, the weighting coefficient matrix stored in the meta-neural network base 205 is used, and each of the knowledge processing methods as described above arranged in the output layer neurons is Recall priority. The knowledge processing method with the highest priority recalled is selected.

When the number of selected sub-knowledge bases is one, the knowledge processing method is naturally determined. For example, if only the rule sub-knowledge base is selected,
"Inference" is selected as the knowledge processing method.

Information regarding the knowledge processing method determined by the knowledge processing method determination means 180 is sent to the control calculation means 300.

The knowledge selection means 1 that characterizes the present invention is as described above.
40 is a description of the function of the method priority calculation step 180.

Next, the control calculation means 300 will be described. In this step, the knowledge selected from the knowledge base group 220 by the knowledge selecting means 140 and the knowledge processing method base 26 by the knowledge processing method determining means 180.
The process control amount is calculated by using the knowledge processing method determined from 0, and is output to the control means 50 and the CRT 90.

For example, a neuro sub-knowledge base and a rule sub-knowledge base are selected as knowledge, and as a knowledge processing method, a method in which a neural net as shown in FIG. 7 is mainly used and correction is performed by a fuzzy inference mechanism is selected. In this case, the control amount can be obtained as in Expression 7.

In this example, a macro change for a relatively long time is controlled using a neural network, and a micro change for a relatively short time is controlled using a fuzzy inference to correct the macro change. It is a thing. In this case, rules for when the value output by the neural network is likely to be output high or when it is output low are prepared for fuzzy inference.

Control amount a (t) = predicted value a (t) + correction value a (t) predicted value a (t) = recall value (t) of neural network a correction value a (t) = fuzzy reasoning a Corrected inference value (t) (Equation 7) However, a: type of control device (in the case of water purification process, coagulant injection rate, chlorine injection rate, etc.) t: variable indicating time Other knowledge of another type, Even when the knowledge processing method is selected, the control amount can be calculated in the same manner. By the above-mentioned method, it is possible to always realize highly accurate control (or presentation of guidance information) while dynamically changing the type of knowledge most suitable for the operating condition and the processing method thereof.

A series of operations of the present invention will be described with reference to the general flow chart shown in FIG.

First, in step 1000, the apparatus is activated, and in step 1001, the process measurement value is read into the database 350.

Next, in Step 10 for selecting knowledge to be used.
Perform from 02 to 1005. In step 1002, data sampling and data format conversion are performed so that the sampled data can be used in the workstation for the control device 100.

At step 1003, new learning for calculating the priority of knowledge is automatically performed using the neural network, and at step 1004, the priority of knowledge is recalled by the neural network. As a result of this recall, in step 1005, the sub-knowledge base to be used is selected.

Next, in steps 1006 to 1010, the knowledge processing method is selected.

In steps 1006 and 1007, data conversion is performed for reading the list of the selected sub-knowledge base and for the workstation used by the control device 100.

Next, in step 1008, automatic learning is performed, and in step 1009, the neural network
Recall the priority of knowledge processing methods. In step 1010, a knowledge processing method is selected based on the recall result.

Next, the control calculation is performed from steps 1011 to 1
Perform at 013.

In steps 1011 and 1012, the selected knowledge is read and a processing program, which is a method for processing this knowledge, is called. And step 1
At 013, the control value is calculated.

Finally, in step 1014, the calculation result is output to the CRT 90 and the control means 50.

When the number of sub-knowledge bases selected in step 1005 is one, the knowledge processing method is automatically determined. Therefore, in such a case, step 1
It is possible to branch from 005 to step 1011.

After outputting the calculation result, step 100 is executed again.
It returns to 1 and enters a new control loop.

In this embodiment, the water purification process is targeted, but the same effects can be obtained in other processes.

[0114]

According to the present invention, among different kinds of knowledge used for determining the controlled variable of the process to be controlled according to the operating conditions of the process to be controlled,
It is possible to select the optimum knowledge, select the optimum method from a plurality of processing methods for handling the selected knowledge, and calculate the control amount.

As a result, the control of the process which has been conventionally performed only by the knowledge of the determined type and the processing method thereof can be performed.
The type of knowledge and the processing method can be changed flexibly according to changes in operating conditions, and due to the favorable synergistic effect of multiple types of knowledge, it is possible to perform more precise process control than before. Become.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of an embodiment in which a process control device according to the present invention is applied to control a water purification treatment process.

FIG. 2 is an explanatory diagram showing an operation example of a knowledge selecting unit.

FIG. 3 is an explanatory diagram of an example in which the knowledge priority calculation step is realized by a neural network.

FIG. 4 is an explanatory diagram showing a general structure of a hierarchical neural network model.

FIG. 5 is an explanatory diagram of an example in which a knowledge priority calculation step is realized by an inference mechanism.

FIG. 6 is an explanatory diagram showing an operation of a knowledge processing method determining unit.

FIG. 7 is a diagram showing an example of elements in a knowledge processing method base.

FIG. 8 is an explanatory diagram of an example of a neural net model for automatic learning.

FIG. 9 is a flowchart showing a flow of overall functions of the present invention.

[Explanation of symbols]

5 ... landing well, 10 ... rapid mixing pond, 11 ... flocculant tank,
12 ... Flocculant injection pump, 13 ... Stirring blade, 14 ... Stirrer, 15 ... Flock forming pond, 16 ... Stirring paddle, 20 ...
Settling tank, 25 ... Filtration tank, 5M, 10M, 15M, 20
M, 25M ... Measuring instrument, 50 ... Control means, 90 ... CRT,
95 ... Keyboard, 100 ... Control device, 140 ... Knowledge selecting means, 150 ... First operating condition inputting means, 151 ... Operating condition inputting step, 160 ... Knowledge priority calculating means, 161 ...
Knowledge priority calculation step, 162 ... Automatic learning step, 163 ...
Priority inference process, 164 ... Priority recall process, 165 ... Knowledge selection process, 166 ... Knowledge selection process, 170 ... Meta neural network, 171 ... Meta rule base, 1
80 ... Knowledge processing method determination means, 190 ... Second operation condition input means, 191 ... Operation condition input step, 200 ... Method priority calculation means, 201 ... Method priority calculation step, 202 ... Meta rule base, 205 ... Meta neural Network, 220 ... Knowledge base group, 260 ... Knowledge processing method base, 300 ... Control calculation means, 350 ... Database.

Front page continued (72) Inventor Mikio Yoda 5-2-1 Omika-cho, Hitachi City, Ibaraki Hitachi Ltd. Omika Plant, Ltd. (72) Inventor Fumichi Kure 5-2-1 Omika-cho, Hitachi City, Ibaraki Prefecture Incorporated company Hitachi Ltd. Omika Factory (72) Inventor Shoji Watanabe 4026 Kuji Town, Hitachi City, Hitachi, Ibaraki Prefecture Hitachi Research Institute Ltd. (72) Inventor Kazuo Yahagi 4026 Kuji Town Hitachi City, Ibaraki Prefecture Hitachi Research Laboratory (72) Inventor Misako Obuchi 4026 Kuji Town, Hitachi City, Hitachi, Ibaraki Prefecture Hitachi Research Laboratory, Hitachi, Ltd.

Claims (7)

[Claims] 1. A process control apparatus for monitoring and controlling a process having a change over time, wherein a means for accumulating a plurality of types of knowledge, which is used to determine a control amount, and a plurality of types for accumulating the knowledge. From the knowledge,
Knowledge selecting means for selecting one or more kinds of knowledge suitable for determining the controlled variable at a certain time point based on the operating conditions of the process, and a plurality of types for determining the controlled variable, Knowledge processing method determining means for determining a knowledge processing method or a combination of processing methods from a plurality of types of knowledge processing methods accumulated by means for accumulating knowledge processing methods and means for accumulating knowledge processing methods And the knowledge selecting means and the knowledge processing method determining means each include a neural network, or an inference mechanism, based on the operating conditions of the process, recall the neural network, or cause the inference mechanism to infer, A process control device characterized by performing knowledge selection and knowledge processing method determination. 2. The knowledge selection means according to claim 1,
A process control method comprising: means for inputting operating conditions of a process; and knowledge priority calculation means for calculating priority of knowledge to be used. 3. The process control apparatus according to claim 2, wherein the knowledge priority calculation means includes learning and recall means based on a neural network model. 4. The method according to claim 1, wherein the knowledge processing method determining means includes means for inputting a process operating condition and method priority calculating means for calculating a priority of a knowledge processing method. Process control device. 5. The process control apparatus according to claim 4, wherein the method priority calculation means includes learning and recall means based on a neural network model. 6. The type of knowledge selected by the knowledge selecting means is at least data obtained from a process, a mathematical model for explaining various phenomena in the process, and a rule format. A process control device characterized by being one of the described knowledge. 7. The knowledge processing method according to claim 1, 4 or 5, wherein the knowledge processing method determining means determines at least one of a learning and recalling means of a neural network, a simulation means and a knowledge inference means. Process control device characterized by being one of the following.