JPH08335101A - Device and method for managing intelligent process - Google Patents

Abstract

PURPOSE: To provide guidance to an operator by presenting only one guidance value by integrating the plural models of different characteristics in a plant operation so as to have a mutually substitutive and complementary relation. CONSTITUTION: This device is provided with a theoretical expression model 30 for judging the guidance value of the plant operation according to a theory based on the operating conditions of a plant, neuromodel 31 for judging the guidance value according to learning based on the history of running operations, fuzzy control model 32 for estimating the guidance value and the degree of certainty according to inference based on the running conditions and the tendency of movement, integral module 34 for averaging and integrating two kinds of judged results of the theoretical expression model 30 and the neuromodel 31, and integral module 35 for correcting and integrating this integrated result, based on the judged result of the fuzzy control model 32.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an intelligent process control device and an intelligent process control method for integrating a plurality of alternative models having different characteristics to control a target complex industrial process.

[0002]

2. Description of the Related Art For example, Japanese Unexamined Patent Publication No. 3-95603 discloses a method for supporting driving of a process that requires operator's judgment by multidimensional judgment, and fuzzy reasoning is applied to a theoretical judgment model in the judgment. At the same time, there is a technique in which a neural network is applied to an intuitive judgment model so that the judgment results can be exchanged freely by combining the two models.

In the above fusion, if the decision results are simply exchanged, the choices of the decision results increase depending on the types of models, but as in the above method, the theoretical decision model and the intuitive decision are used. When we try to eliminate mutual influences by changing the method of calculating the degree to which the conclusion part holds in each inference mechanism with the model, the human right peculiar right brain from the interactive operability of this logical judgment result and intuitive judgment result. It is possible to simulate the function of the brain that interacts with the left brain.

However, as described above, even if the technical improvement is required for the qualitative improvement of each model, the interactive operation alone entrusts the human operator with the decision of the model selection, and as a result, an alternative and complementary model is obtained. Further improvement is necessary before the guidance of the only model through multiple judgments, because it is impossible to obtain a positive relationship.

Therefore, before reviewing the roles of the various models as described above, consider a process control aiming at quality control as an example and consider a human behavior norm in the control.

[0006] In many industrial processes for producing industrial products, the following factors can be cited as factors that make process control for quality control of produced products difficult.
That is, (1) the variety and complex entanglement of factors that affect the characteristics to be managed, (2) noise associated with measurement, (3) non-linearity that appears in observable information that substitutes the characteristics for target, and (4) individual characteristics There are four points: lack of norm for handling different information.

As a plant, a concrete concrete (hereinafter referred to as "green concrete") manufacturing process is taken as an example and it is as follows. Aggregate such as gravel, sand, etc., and some additives and water are calculated in a predetermined proportion in cement, and mixed in a mixer to produce raw materials, which are one of the main quality control items of raw concrete produced and shipped. There is a liquidity index called concrete slump. This liquidity index corresponds to the managed characteristic of (1) above.

The management of raw concrete and slump using the slump value measured by the raw concrete and slump test method defined in JIS A1101 as a control index is not easy for the following reasons. Generally, the factors that determine deciduous concrete and slump are water mixing ratio, particle size distribution and particle shape of aggregate, mixing time, temperature condition and various other factors. Above all, the total amount of water contained in the produced ready-mixed concrete is an important target for management as the largest factor. The weighing of various materials for producing ready-mixed concrete is usually automated and a certain range of accuracy is guaranteed, but it is contained in the aggregate, especially the sand called fine aggregate, which has a great influence on the determination of the total water content. Fluctuations in the amount of water taken up are a major issue in slump management.

In the blending calculation for determining the blending ratio of the ready-mixed concrete, assuming the surface water percentage (surface water percentage) of each aggregate, multiplying it by the usage amount of each aggregate, the amount of water brought into the aggregate is calculated. The water value is calculated by calculating and subtracting from the total amount of water required. The setting of the surface water ratio of each aggregate, which is a prerequisite here, is not only the regular inspection at the time of sampling at the time of using the material based on the above JIS regulation, but also the adjustment by the dynamic grasp for the purpose of correction of the composition at the time of use. Although it is possible to use finely grained wood, it is difficult to make accurate settings. In particular, the amount of water brought in per unit weight is large due to the size of the specific surface area, and it is not easy to grasp the surface water dynamics of fine aggregate (sand) that has a decisive effect on slump management. Accurate slump management is difficult because of the subtle changes in properties (particle size distribution, particle shape, etc.), temperature and other factors.

The surface water ratio of the aggregate is measured in accordance with JIS A1125. However, in addition to the fixed point observation that is usually twice a day, the sample is sampled from the target aggregate. It is difficult to completely eliminate noise, and it remains at the material acceptance inspection stage. There is a slump monitor as an execution means for the dynamic grasp of the change of the surface water ratio, and a method of inferring it from the change of the viscosity of the ready-mixed fresh concrete which is grasped by the slump monitor is generally performed. Although this slump monitor has been evaluated practically, it cannot be said to be complete because it is impossible to deny the possibility that the above-mentioned other factors change to reduce all the slump change factors to the surface water change.

Therefore, various types of moisture meters are used as measuring instruments for measuring the surface water ratio of aggregates at the time of use. There is a big problem in the absolute accuracy of measured values due to the bias due to the sampling method.

[0012] Even when the moisture meter measurement is used together with the slump monitor in order to further improve the slump management, at present, only the slump and moisture measurement results are presented in parallel. There is no standard for connecting various measurement results and organically integrating various models, and it is the fact that it is inevitable to rely on the management operation centered on the experience and intuition of a skilled human operator.

As described above, a method theoretically supported by the above problem of integrating a plurality of measurement results to determine a safe and effective guidance value suitable for the situation has not been established yet. However, skilled human operators perform operations that are feasible and satisfy the target level through daily driving.

Therefore, in order to extract the points of model integration from the action standards performed by the human operator, the action standards are summarized as follows. That is, (1) comprehensive understanding of the situation, (2) development of a general operation policy, (3) recall of multiple alternatives from different viewpoints, (4) integration considering the mutual relationship of each alternative, ( 5) Stepwise execution of the determined operation amount can be summarized in five.

The internal norm that supports the behavior of the human operator through these five behavioral standards is a stable solution orientation aiming at more aspects than at the point of avoiding extremes.
Balanced orientation that balances the burden and effect of each decision by diversifying risk, information seeking orientation to increase confidence by taking a stepwise approach, satisfaction that holds yield based on expected effect of operation and risk estimation It can be summarized into the above four standard normative elements.

[0016]

When the roles of various models are considered based on the above-mentioned human behavioral norms, the internal normative elements up to the above are the behavioral patterns of the operator in the actual plant. Even if it is obtained from the decision-making requirements and decision-making requirements, the situation-aware stabilization mechanism that can be learned by human operators cannot be incorporated as one form of relaxation (relaxation method). Then, there was a problem that various models could not be integrated so as to obtain mutually alternative and complementary relationships.

The present invention has been made in order to solve the above-mentioned conventional problems, and integrates a plurality of models having different characteristics in plant operation so as to have an alternative and complementary relationship with each other. It is an object of the present invention to provide an intelligent process control device and an intelligent process control method capable of realizing guidance to an operator by presenting only one guidance value.

[0018]

In order to solve the above problems and achieve the object, an intelligent process control apparatus according to the invention of claim 1 is a plant based on an operation status and a history of operation. The operation amount is determined and these are averaged and integrated, and the dynamic tendency is added to the operating situation to further determine the plant operation amount and reliability, and the plant operation amount and the integrated plant operation amount are determined. Integrate according to reliability.

In the intelligent process control apparatus according to the second aspect of the present invention, a theoretical formula is applied to the plant, a learning formula is applied to the operation history, and an inference formula is applied to the operating condition and its dynamic tendency with respect to the plant. To do.

The intelligent process control apparatus according to the invention of claim 3 handles the plant operation amount as a guidance value of the plant operation.

An intelligent process control device according to a fourth aspect of the present invention constructs a theoretical model corresponding to a driving situation, a neuro model corresponding to a history of driving operations, and a fuzzy control model corresponding to a driving situation and its dynamic tendency. Then, the integrated guidance values obtained from the theoretical model and the neuro model are averaged and integrated, and the integrated guidance value and the guidance value from the fuzzy control model are integrated again according to the certainty factor from the fuzzy control model. To construct.

In the intelligent process control apparatus according to the fifth aspect of the present invention, the integrated module group performs the integration by referring to the reliability of the guidance value of the plant operation obtained by the neuro model.

In the intelligent process control apparatus according to the invention of claim 6, reliability is improved by adjusting the adoption rate of the guidance value of the plant operation obtained by the neuro model according to the contribution rate by extrapolation data at the time of integration. To get

In the intelligent process control apparatus according to the invention of claim 7, the contribution rate and the adoption rate are expressed as follows: NK = (VK-VG) / VK NK: Contribution rate of neuro model VK: Teacher data of extrapolation data Variance VG: Variance of error of extrapolation data NW = NRM * NK NW: Neuro result adoption ratio NRM: Neuro result adoption maximum ratio

The intelligent process control apparatus according to the invention of claim 8 sets an allowable range according to the level of certainty at the time of correction, and obtains a guidance value so as to be within the allowable range.

In the intelligent process control method according to the invention of claim 9, the guidance value of the plant operation is judged based on the theory based on the operating condition of the plant to be operated, and the plant operation is performed by learning based on the history of the operating operation for the plant. Judgment of the guidance value of, and judgment including the guidance value of the plant operation by inference based on the operating condition of the plant, integration by averaging the judgment results of these theory and learning, and the integration result and the judgment result of inference. The reliability value is added to the integration, and in the judgment by inference, the guidance value according to the inference rule and the certainty factor according to the dynamic tendency of the plant are estimated for this guidance value.

In the intelligent process control method according to the tenth aspect of the invention, the integration is performed by referring to the reliability of the guidance value of the plant operation obtained in the second process in the fourth process. .

In the intelligent process control method according to the eleventh aspect of the present invention, the reliability is improved by adjusting the adoption rate of the guidance value of the plant operation obtained by the neuro model according to the contribution rate by the extrapolation data at the time of integration. To get

In the intelligent process control method according to the twelfth aspect of the present invention, the contribution rate and the adoption rate are: NK = (VK-VG) / VK NK: Contribution rate of the second process VK: Extrapolation data teacher Variance of data VG: Variance of error of extrapolation data NW = NRM * NK NW: Adoption rate of judgment result in the second process NRM: Maximum adoption ratio of judgment result in the second process.

In the intelligent process control method according to the thirteenth aspect of the present invention, the allowable range is set according to the level of certainty at the time of correction, and the guidance value is obtained so that the allowable range falls.

[0031]

[Example] In general, a logical relationship, non-linearity, or ambiguity relationship is introduced into a description of a target process that cannot be expressed only by a physical relationship description, and state estimation and management or control action suitable for the actual condition of the target are performed. The approach to be performed can be called an intelligent process management based on human intelligent information processing. Introducing the concept of intelligent process control is an important viewpoint for realizing the problem-solving model that incorporates the above-mentioned human behavioral norms, but its effectiveness has already been confirmed by a clear basis. The objective model based on the existing theory is included, and the effect as a reflection of the above-mentioned yield orientation can be expected.

Therefore, the intelligent process control will be described in detail by taking a raw plant as an example of a concrete target process. Normally, a monitoring means for the kneading process is used in the raw plant like the slump monitor, and the estimated value of the slump presented as an output from the monitoring means and the fine aggregate (sand) that greatly affects the slump management are provided. ) To optimize the water management by presenting an appropriate guidance value for the surface water rate by adding the actual results (history) such as the operation records of the operator to the measured value of the surface water rate. Is the management item for raw conslump. This guidance value indicates a guideline effective as information for determining the plant operation amount.

First, the basic concept will be described. FIG. 1 is a conceptual diagram showing an embodiment of an intelligent process control method according to the present invention. In FIG. 1, 30 is a theoretical formula model for determining an average guidance value for setting surface water ratio by a theoretical formula, and 31 is a surface. A neural network (hereinafter referred to as neuro) model that determines the average guidance value for water rate setting by a learning formula, 3
Reference numeral 2 is a fuzzy control model that judges the guidance value for setting the surface water ratio by an inference formula. This fuzzy control model 32
Is composed of an operation policy model 33A and a certainty factor model 33B. 34 is an integration module that averages the judgment results of the theoretical model 30 and neuro model 31 to obtain a guidance value, and 35 is an integration module 34 that verifies the integration result with the judgment result of the fuzzy control model 32 to obtain the guidance value. An integrated module for correction, 36 is a module for displaying a guidance value which is an integrated result of the integrated module 35, 37 is an operator operation module for performing an operation according to the guidance value, and 38 is based on an operation result of the module 37, that is, an operator set value. This is an operation record module for recording operation records (history).

In FIG. 1, SD is slump data including the estimated slump, the target slump and the smoothed value of the slump difference, OH is the latest set value set by the operator, and SD ₁ is the slump data SD. In the latest smoothed value of slump difference, SD ₂ is the smoothed value of slump difference for several batches in the past, IH is the latest measured value with a moisture meter, and LM is the theoretical model 30. A guidance value indicating the calculated surface water rate, NM is a guidance value indicating the surface water rate calculated by the neuro model 31, and FM is a guidance value indicating the surface water rate calculated by the operation policy model 33A of the fuzzy control model 32. , KM is confidence calculated in confidence model 33B fuzzy control model 32, UM ₁ guidance value integrated by the integration module 34, UM ₂ integrated modular Integrated guidance values Lumpur 35, UM ₃ is a guidance value after being operated by the module 37 of the operator's operation. The guidance value is the latest set value of surface water ratio.

In the above configuration, basically, a kind of relaxation mechanism is introduced through integration of duplication caused by multiple models described in parallel, alternatively or competitively for the same target process. In order to realize an intelligent process control model, a theoretical model 30, a neuro model 31, a fuzzy control model 3
Two types of different models are adopted, and each of the models 31 to 33 has objective, learning, and subjective characteristics which are mutually alternative and complementary. In addition, the integrated modules 34 and 35 at each stage are connected to respective models 30 to 3
This is for reflecting the information about the characteristics and the confidence intervals of No. 2. Therefore, the characteristics of each model are summarized in Table 1 below.

[0036]

[Table 1] In integrating each model, it is necessary to reflect the characteristics expressed in “expectation” and “problem” separately from the above-mentioned Table 1 in the calculation mechanism of the integration part in a form that can be treated numerically. This is shown in Table 2 below.

[0037]

[Table 2]

The individual processes of the model and the module will be described based on the problems and evaluation indexes of the various models described above. Theoretical model 30 The cause of the slump difference, which is obtained by subtracting the target slump value from the slump estimated value, is the difference from the true value of the surface water rate estimation of the fine aggregate, which is the basis of the compounding correction calculation when measuring the kneading materials. Suppose that. Therefore, the true fine aggregate surface water rate can be estimated based on the fine aggregate surface water rate set in the current batch.

In this theoretical formula model 30, the slump difference between the estimated slump value and the target slump value is calculated based on the input slump data SD as if it were due to the overmeasured amount of water obtained from the latest measurement result. A calculation is performed to calculate the surface water, calculate the surface water to be operated in the next measurement, obtain the amount of change from the currently set value, and stabilize the amount. The amount of change is the guidance value LM.

Neuro model 31 In the neuro model 31 which learns the operation behavior of the operator based on various presented information and general situation grasp from the actual operation data, the slump data SD which is the latest slump estimation value, the moisture meter A guidance value NM, which is a change amount for the next batch measurement, is calculated and output on the basis of input variables such as the surface water rate IH from the above and the operator's surface water rate OH. In this neuro model 31, the mutual relation of the input and output is expressed as a network, and the nonlinear characteristic is expressed with respect to the level of the realized value of each variable and the mutual relation of the variables to accumulate the actual data of the operation by the operator. Along with improving accuracy. Back-propagation is adopted here.

Fuzzy control model 32 In the operation policy model 33A, the first step is to change the set value of the surface water ratio of the fine aggregate from the smoothing value of the current batch of the slump difference and its tendency according to a plurality of inference rules. Decision is made.

Therefore, first, the smoothing value SD ₂ of the slump difference in the past several batches is input as a dynamic tendency, and the tendency of the slumping difference is calculated from this and the latest smoothing value SD ₁ of the slump difference. A change amount (guidance value FM) of the surface water ratio is obtained according to the fuzzy rule shown in FIG. 2 based on the smoothed value SD ₁ of the slump difference and the tendency of the slump difference. This calculation is shown in Figure 3.
It can be expressed by the membership function shown in.

In FIGS. 2 and 3, for example, when the slump difference is on the rise and the smooth value of the slump difference is "positive" (see), it is determined that the degree of change is increased in the positive direction (up to plus). ), And in the same case if the smooth value of the slump difference is near zero (see),
By giving a judgment that the degree of change is weakened (a small change to the plus), the smooth value of the slump difference is "negative" in the same case.
In the case of (see), the judgment that the degree of change is strengthened in the negative direction (down to negative) is given.
It should be noted that the same method is used when the tendency of the slump difference is constant or when the slump difference is declining. In FIG. 3, the membership function of the amount of change means a change to plus when it is located on the right side of the center, and shows that the amount of change increases as the distance from the center increases. Also, as in the case of, if it is left and right across the center,
Indicates no change.

Next, the estimation of the certainty factor KM can be expressed by a certainty factor estimation rule (FIG. 4) and a membership function (FIG. 5) corresponding to the rule.
Here, as shown in FIG. 4, the operation policy of the change amount of the surface water ratio (guidance value FM: up / small change / down) and the tendency of surface water (up / small change / down) given from the moisture meter are shown. Is calculated to estimate the certainty factor.

In FIG. 4 and FIG. 5, when the change amount is up and the tendency of surface water is up (see), it makes sense and a judgment with a high degree of certainty is given. If the tendency of surface water is small in a case (see), a judgment with a certainty factor, for example, is given, and in the same case, if the tendency of surface water is down (see), Try to give a judgment with less certainty. It should be noted that the same method is used when the change amount is small or when the change amount is down. As shown in the membership function of FIG. 5, the certainty factor is distributed in the range of 0 to 100%.

In this way, the calculation for estimating the guidance value FM and the certainty factor KM is executed.

As described above, the fuzzy control model 3 in which the operation policy model 33A and the certainty degree model 33B are combined.
The output of 2 (judgment result) obtains the evaluation information of the validity of the guidance based on the common sense of the field that has long been recognized as a qualitative relationship.

Integration Module 34 Integration module 34 provides guidance value LM from theoretical model 30 and guidance value N from neuro model 31.
By integrating M and M, one guidance value UM ₁ is obtained. Here, the set value of the surface water ratio of the fine aggregate obtained as the output of the above two models 30 and 31 is weighted according to the degree of reliability that can be expected in each calculation result, and the integration by averaging is performed. Done.

The theoretical model 30 and the neuro model 31 have contrasting characteristic differences, but if they are assumed to have the same reliability, for example, if it is assumed that weights of 0.5 are given, the theory The expression model 30 has reliability as unbiased estimation, while the neuro model 31 is evaluated by an extrapolation test in the process of model learning (maximum 100%).
Is calculated. Therefore, a weight for the weighted sum is given in a form that reflects the extrapolation test evaluation of the neuro model 31. This is because the accuracy of the neuro model 31 can be expected to gradually improve with the accumulation of data, so that the learning content can be reflected in the result by increasing the weight of the weighted sum as the extrapolation evaluation becomes higher.

Specifically, the integration module 34 executes an operation for adjusting the degree of adoption of the neuro result depending on whether the contribution rate of the extrapolation data of the neuro model 31 is good or bad. Therefore, the contribution rate and the adoption rate are calculated according to the following equations (1) and (2), respectively. NK = (VK-VG) / VK (1) NK: contribution rate of the neuro model 31 VK: variance of extrapolation data teacher data VG: variance of extrapolation data NW = NRM * NK ... (2) NW: Ratio of adoption of neuro results NRM: Maximum ratio of adoption of neuro results

[0051]Integration module 35 This integration module 35 is obtained by the integration module 34.
Guidance value UM ₁Of the fuzzy control model 32
The results of both models are integrated by verifying the results of disconnection.
It Guidance value F is used for the judgment result that is a guide for the verification.
There is M and certainty factor KM.
Then, an allowable range according to the certainty factor KM of the change amount is calculated.
The guidance value of the integrated module 34 is within this allowable range.
If it fits, output it, that is, the guidance value UM
₂If the guidance is displayed as
If you want to make a correction within the allowable range,
Is displayed.

More specifically, it shows a case where the confidence level of 0% is completely unknown, and in this case, for example, the standard deviation ± of the guidance value KM of the surface water change operated by the operator is ±. As long as the guidance value UM ₁ is within the double allowable range, it is statistically 95%.
Reliable range. Further, when the certainty factor is 100%, the guidance value FM may be adopted, but it may be set within ± 1% of this guidance value FM, for example. In this way, it is possible to arbitrarily set the allowable range between the certainty factors of 0% and 100% and calculate the allowable range within the certainty factors of 0 to 100%.
This allowable width can be calculated by the following equations (3) and (4). y = {(a−b) / 100} * KM + b (3) FM−y ≦ HABA ≦ FM + y (4) a: Permissible width percentage when confidence is 100% b: When confidence is 0% Permissible Width Percentage y: Confidence KM Permissible Width Percentage HABA: Allowable Width

Then, in the process of calculating the guidance value UM ₂ , if the guidance value UM ₁ is a value within the allowable width HABA, the guidance value UM ₁ is output as the guidance value UM ₂ . If the guidance value UM ₁ is larger than the allowable width HABA, the guidance value UM
₂ is corrected to the upper limit value (FM + y) corresponding to the certainty factor KM, and if the guidance value UM ₁ is smaller than the allowable width HABA, the guidance value UM ₂ is changed to the lower limit value (FM-y) corresponding to the certainty factor KM. to correct. With this correction, the guidance value UM ₂ is contained within the allowable width HABA.

As a basic function for effectively operating the above-mentioned intelligent process control device 1, data generated in association with kneading of ready-mixed concrete, which is an example thereof, is recorded as an operation record of an operator. At the same time, it is effective to have a self-organizing mechanism that accumulates the data in a database, analyzes it periodically, and updates the neuro model 31 by learning while gradually increasing the practicality.

Next, a specific configuration of an apparatus that realizes intelligent process control based on the above concept will be described.

FIG. 6 is a block diagram showing an embodiment of the intelligent process control device according to the present invention. In the figure, 21 is an input unit for inputting / outputting data to / from a plant, 22 is a CPU for controlling the entire apparatus according to a program, 23 is an operation unit including an input device such as a keyboard and mouse, and 24 is a measured value or guidance. A display unit for displaying information such as values and an operation menu, 25 is a first storage unit storing various programs for realizing the concept shown in FIG. 1, and 26 is data such as operation results and measured values. Second storage unit for recording
Is a bus line in this device.

The first storage unit 25 has a main program MP that plays a central role in all programs, a slump estimation program SP, and a theoretical formula corresponding program LP for executing the process of the theoretical formula model 30. Neuro-compatible program NP for executing the process of the neuro model 31 and fuzzy compatible program FP for executing the process of the fuzzy control model 32
And an integration program IP for executing each process of the integration modules 34 and 35.

In the second storage section 26, the operation record database CDB for recording the operation record of the operator as a history and the measurement record database for recording the measurement value from each measuring device such as a moisture meter as a history. VDB is provided.

FIG. 7 is a block diagram schematically showing the positioning of the intelligent process control device shown in FIG. 6 in the ready-mixed concrete plant.
It includes an automatic weighing device that controls the production of ready-mixed concrete by setting the kneading capacity and the surface water ratio of fine aggregate (sand) which is the kneading material.

The batch type plant shown in FIG. 7 includes the intelligent process control device 1, a plurality of storage bins 2 for separately storing two or more kinds of fine aggregates, and a plurality of storage bins 2 in each storage bin 2. The moisture meter sensor 4 for measuring the moisture content of the aggregate, the moisture meter measurement control unit 5 for calculating the surface water ratio based on the moisture content, and the moisture meter 4 and the moisture meter measurement control unit 5 are connected. , The interface section (hereinafter referred to as I / F) 7 that connects the intelligent process control device 1 and the moisture meter measurement control unit 5, and the surface water ratio given by the intelligent process control device 1. The automatic metering device 8 controls the metering of water and the metering of other kneading materials based on the set value, a mixer 9 for kneading, a concrete hopper 10 and the like. Reference numeral 11 is an agitator vehicle for loading concrete.

First, the management procedure of the intelligent process in the above batch type plant will be described with reference to FIG. In this example, the intelligent process control device 1 sets the measured value in which the surface water of sand is corrected in each measuring bottle (cement, water, fine aggregate, admixture, etc.)
Used to do against. First, the automatic weighing device 8
When the start of weighing is instructed after confirmation of various set values in step (step 1002), each storage bin 2 ...
The kneading material is dropped into the container. At this time, when the kneading material is fine aggregate, the water content is measured by the water content sensor 4 when falling, and the data is sent to the water content measurement control unit 5. The moisture meter control unit 5 calibrates the measured moisture content with a model registered in advance to calculate the surface water rate of each storage bottle.

These surface water ratios are output to the intelligent process control device 1 via the I / F 7 in the latter stage, and are averaged there according to the mixing ratio of the fine aggregate. In this intelligent process control device 1,
A process of obtaining a surface water rate value and a dynamic graph of its transition from the averaged surface water rate and displaying it as a visible image on the display unit 24 is performed.

When the measuring process of all used kneading materials is completed,
The kneading material is discharged from the measuring bottle 3 to the mixer 9 in response to the mixing start operation. This kneading material is kneaded by the mixer 9, and at that time, the load of the mixer 9 is continuously measured and the data is output to the intelligent process control apparatus 1. The intelligent process control device 1 performs a process of displaying a load value and its transition as a visible image (step 100).
4).

Further, in the intelligent process control system 1, the mixer 9
When kneading progresses and the load state stabilizes, the slump estimation is performed and this is also displayed as a numerical value (step 1006).

Thereafter, the guidance value calculation process for setting the surface water ratio of the fine aggregate is performed as described above (step S1008). The operator confirms this guidance value and is given three options: to change the setting according to the guidance, to make the setting similar to the previous batch without changing the setting, and to change the setting to an arbitrary value different from the guidance value. For example, when the above is selected by the operation of the operation unit 23, the surface water rate according to the guidance value is instructed to the automatic weighing device 8 for the next batch as a set value (steps 1010, S1012). Further, when the above is selected, the surface water rate adopted in the previous batch becomes a set value, and this is instructed to the automatic weighing device 8 as the surface water rate of the next batch (steps 1010, S1014).

When the above is selected, the operator gives an instruction to change the setting, and an arbitrary input value is set. This input value becomes a set value, which is instructed to the automatic weighing device 8 as the surface water rate of the next batch (steps 1010 and S1016). Since the set surface water rate is a moisture correction amount of the fine aggregate stored in each storage bin 3, ..., Any one of the above items 1 to 3 can be selected for each storage bin 3. When the setting of the surface water ratio is completed, the set value is set to the operation record database ODB.
The process of writing to the is executed (step 1018).

When the kneading is completed in the mixer 9,
The ready-mixed concrete is discharged to the concrete hopper 10 and further loaded on the agitator vehicle 11 for shipment.

As described above, regarding the slump management, the guidance values, which are the plant operation amounts, are calculated from the operation status and the history of the operation operations, and these are averaged and integrated, and the dynamic tendency is added to the operation status. Further, since the guidance value and the certainty factor are judged and the guidance value and the integrated guidance value described above are integrated according to the certainty factor, like the theoretical model 30, the neuro model 31, and the fuzzy control model 32, A group of integrated modules (integrated modules 34, 3) so as to have mutually alternative and complementary relationships with respect to a plurality of models having different characteristics
By integrating in 5), the guidance to the operator can be realized by presenting only one guidance value.

[0069]

As described above, according to the present invention, regarding the plant, the plant operation amount is judged from the operating condition and the history of the operating operation, and these are averaged and integrated. In addition, the plant operation amount and the reliability are determined by integrating the plant operation amount and the integrated plant operation amount according to the reliability. It is possible to obtain an intelligent process control apparatus and an intelligent process control method that realizes guidance to an operator by presenting only one model by integrating so as to have an alternative and complementary relationship with.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing an embodiment of an intelligent process control method according to the present invention.

FIG. 2 is a diagram illustrating a fuzzy rule for deriving a guidance value based on a fuzzy control model.

FIG. 3 is a diagram illustrating a membership function for estimating a guidance value using a fuzzy control model.

FIG. 4 is a diagram illustrating a fuzzy rule for deriving a certainty factor using a fuzzy control model.

FIG. 5 is a diagram illustrating a membership function for estimating the certainty factor using a fuzzy control model.

FIG. 6 is a block diagram showing an embodiment of an intelligent process control device according to the present invention.

7 is a configuration diagram schematically showing the positioning of the intelligent process control device shown in FIG. 6 in a ready-mixed plant.

FIG. 8 is a flowchart illustrating an intelligent process management procedure.

[Explanation of symbols]

 1 Intelligent process control device 22 CPU 24 Display unit 25 First storage unit 26 Second storage unit 30 Theoretical model 31 Neuro model 32 Fuzzy control model 33A Operation policy model 33B Confidence model 34, 35 Integrated module FP Fuzzy compatible program IP Program for integration LP Program for theoretical formula NP Model for neuro ODB Operation record database SP Slump estimation program VDB Measurement record database

Claims (13)

[Claims] 1. A first determining means for determining a plant operation amount based on an operating condition of a plant which is an operation target, and a second determining means for determining a plant operation amount based on a history of operation operations on the plant. And a third judging means for judging the plant manipulated variable and its reliability based on the operating condition of the plant and its dynamic tendency, and two plant manipulated variables obtained by the first and second judging means. The first integrating means for averaging and integrating, the plant operation amount obtained by the first integrating means, and the third
Intelligent process control device comprising: second integration means for integrating the plant operation amount obtained by the determination means according to the above reliability according to the reliability. 2. A theoretical formula judging means for judging a plant manipulated variable into a theoretical formula based on an operating condition of a plant which is an operation target, and a plant manipulated variable according to a learning formula based on a history of driving operations for the plant. A learning formula judging means, an inference formula judging device for judging the plant operation amount and its reliability in an inference formula based on the operating condition of the plant and its dynamic tendency, and the theoretical formula judging device and the learning formula judging device. Reliability of the first integration means for averaging and integrating two plant operation quantities, the plant operation quantity obtained by the first integration means, and the plant operation quantity obtained by the inference formula judging means. An intelligent process control device comprising: 3. The intelligent process control device according to claim 1, wherein the plant operation amount indicates a guidance value for plant operation. 4. A theoretical formula model for determining a guidance value for plant operation based on a theory based on the operating condition of a plant to be operated, and a neuro for determining a guidance value for plant operation by learning based on a history of the operation on the plant. A model, a fuzzy control model that makes a judgment including a guidance value for plant operation by inference based on the operating status and dynamic tendency of the plant, and a theoretical formula model and a neuro model
The above-mentioned fuzzy control model is provided with an integrated module group for averaging and integrating the two judgment results, and correcting and integrating the integration result based on the judgment result obtained by the fuzzy control model. While estimating the guidance value of the plant operation according to a plurality of different inference rules, the confidence factor according to the dynamic tendency of the plant is estimated for the estimation result, and the estimated operation amount and the confidence factor are used as the determination result. An intelligent process control device characterized by: 5. The intelligent process control device according to claim 4, wherein the integrated module group performs integration by referring to the reliability of the guidance value of the plant operation obtained by the neuro model. 6. The reliability is obtained in the integration by adjusting the adoption rate of the guidance value of the plant operation obtained by the neuro model according to the contribution rate of the extrapolation data of the neuro model. 5. The intelligent process control device described in 5. 7. The intelligent process control apparatus according to claim 6, wherein the contribution rate and the adoption rate are calculated according to the following equations, respectively. NK = (VK-VG) / VK NK: Contribution rate of neuro model VK: Variance of extrapolation data teacher data VG: Variance of extrapolation data error NW = NRM * NK NW: Neuro result adoption ratio NRM: Neuro result Maximum hiring rate 8. The intelligent process control according to claim 4, wherein in the correction, an allowable width is set according to a certainty level, and a guidance value is obtained so as to fall within the allowable width. apparatus. 9. A first process of determining a guidance value of a plant operation based on a theory based on an operating condition of a plant which is an operation target, and a guidance value of a plant operation based on learning based on a history of the operation operation on the plant. The second process, the third process of making a judgment including the guidance value of the plant operation based on the inference based on the operation status and dynamic tendency of the plant, and the two judgment results obtained by the first and second processes. And averaging and integrating, and a fourth process of correcting and integrating the integration result based on the determination result obtained in the third process, and the third process is set in advance. Guidance values for plant operation are estimated according to multiple different inference rules, and a certainty factor corresponding to the dynamic tendency of the plant is estimated for the estimation result. To, intelligent process control method characterized by the determination result and the confidence that these estimated manipulated variable. 10. The intelligent process management according to claim 9, wherein the fourth process is performed by referring to the reliability of the guidance value of the plant operation obtained by the second process. Method. 11. In the integration, reliability is obtained by adjusting the adoption rate of the guidance value of the plant operation obtained in the second step according to the contribution rate of the extrapolation data in the second step. 11. The intelligent process control method according to claim 10, wherein: 12. The intellectual process management method according to claim 11, wherein the contribution rate and the adoption rate are calculated according to the following equations, respectively. NK = (VK-VG) / VK NK: Contribution rate of the second process VK: Variance of extrapolation data teacher data VG: Variance of extrapolation data error NW = NRM * NK NW: Judgment by the second process Result adoption ratio NRM: Maximum adoption ratio of the judgment result in the second process 13. The intelligent process control according to claim 11, wherein in the correction, an allowable width is set according to a certainty level, and a guidance value is obtained so as to fall within the allowable width. Method.