JPH1143677A - Production plant control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize an efficient productivity increase and also to control a production line at a low cost, without requiring expensive analyzing equipment. SOLUTION: At least one of temp., flow rate and pressure at a specific point of a production line is measured with measuring apparatuses T-1 to T-3 and F-1 to F-3. Next, data for an estimated equation used to prepare an estimated equation 4 are selected from the measured values thus accumulated and property values of a fluid actually measured, and the estimated equation 4 for properties regarding the fluid is prepd., utilizing these selected data for an estimated equation through an estimated equation generating device 3 which applies a neural network. After that the estimated equation 4 is inputted to a control device 5 wherein an estimated value calculated from measured values using the estimated equation 4 and a predetermined target value are compare- computed. Then all or any one of limp., flow rate and pressure at a specific point of chemical facilities in a production line is controlled based on the result of this comparison-computation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a production line for producing several types of products using a chemical apparatus, and to a production plant control system for performing predetermined control on the production line.

[0002]

2. Description of the Related Art Conventionally, a plurality of products such as LPG, naphtha or light oil are produced by refining a raw material such as crude oil in a manufacturing plant equipped with a chemical device.

In the processing or refining of a raw material performed in this manufacturing plant, generally, the raw material is heated by a plurality of distillation apparatuses and the like.
These raw materials are classified into LPG, naphtha, kerosene, and residual oil according to the difference in boiling point.

[0004] Since these intermediate purified products usually contain a trace amount of corrosive sulfur compounds, these compounds are removed by washing with a chemical such as sulfuric acid or soda or desulfurization by hydrogenation to improve the quality. Aim. Further, the residual oil is processed into lubricating oil, asphalt, and the like by a vacuum distillation apparatus.

[0005] In addition, various cracking methods for producing gasoline from heavy oil or light oil have been used to produce gasoline. Here, distillation apparatuses and the like are roughly classified into atmospheric distillation apparatuses and reduced-pressure distillation apparatuses. Of these, the atmospheric distillation apparatuses perform distillation under a pressure near ordinary pressure, and various kinds of fractions are obtained from raw materials. Is a device for separating The filling temperature that is important in the distillation operation varies depending on the type of the raw material.
30-365 ° C, and the naphtha distilling temperature is 100-
130 ° C, the withdrawal temperature of kerosene is around 210-240 ° C, and the withdrawal temperature of light oil is around 290-330 ° C.

[0006] The vacuum distillation apparatus is used to further distill high-boiling oils such as residual oil at normal pressure. When these high-boiling oils are heated to 350 ° C or more at normal pressure, the oil is decomposed. It causes deterioration of quality and yield.

Therefore, in order to distill the high-boiling fraction without decomposing it, it is necessary to reduce the pressure inside the distillation apparatus. Therefore, this vacuum distillation apparatus is used in a vacuum distillation method in which distillation is performed while reducing the pressure in the distillation apparatus.

[0008] The products distilled by this vacuum distillation apparatus are:
It is mainly used in the production of lubricating oil, asphalt or pitch using the residual oil from atmospheric distillation as a raw material, or in the production of a raw material for catalytic cracking and a raw material for indirect desulfurization.

By the way, in the operation of a manufacturing plant, it is possible to accurately and in real time grasp the properties of the oil flowing in the piping, which leads to energy saving of the plant and an increase in production of high value-added products. Is also an important factor in optimal operation.

When the operator determines that the control device used in the plant operation is out of the desired property based on the parameters obtained through empirical judgment or trial and error, the operator changes the property. A target value (set point) such as a certain temperature or flow rate is input (replaced) to control the manufacturing plant.

However, it is necessary for a plant operator or the like to grasp the properties of oil in real time, easily and accurately, but it is often technically difficult. Here, an example of control performed in a conventional manufacturing plant will be described. This control is performed as shown in FIG.
The product sent from 01 is further LP in the rectification column 100
It is to be fractionated into G and naphtha.

Then, the LPG discharged from the top of the rectification tower 100 is cooled by the cooler 103, and
The liquid is returned to the liquid in 4 and returned to the rectification column 100 again. At this time, the liquid level of the rectification tower receiving tank 104 into which the liquefied LPG flows is measured by the liquid level measuring device 105 provided in the rectification tower receiving tank 104,
If it is larger than the set value based on this measurement result,
The control valve 106 and the like are opened to remove the liquefied LPG out of the system.

If the value is smaller than the set value, the control valve 106 and the like are squeezed and the liquefied LPG is returned to the rectification column 100 again. Next, the refluxed liquefied LPG is forcibly returned to the rectification tower 100 by a pump 107 provided in the middle of the pipe. The rectification column 100 is provided with a control valve 108 and the like to prevent reflux,
The flow rate of PG is controlled.

Next, a part of the naphtha liquid fractionated in the rectification column 100 is sent to the gas absorption column 102. On the other hand, the gas discharged from the rectification column receiving tank is sent to the gas absorption column 102.

Then, propane and butane remaining in the gas during the dropping in the gas absorption tower 102 are absorbed. Here, when the flow of the naphtha liquid into the gas absorption tower 102 is small, the liquid level in the rectification tower 100 rises. Therefore, it is necessary to control the liquid level in order to keep the liquid level constant.

The control of the liquid level in the rectification column 100 is as follows.
Measured by a liquid level measurement device 109 provided in the rectification column 100,
If it is larger than the set value based on this measurement result,
By opening the control valve 110 and the like, the naphtha liquid flows out of the system other than the gas absorption tower 102.

If the value is smaller than the set value, the control valve 110 or the like is throttled to stop the outflow of the naphtha solution or to reduce the outflow amount. By doing so, the rectification column 100
The purpose is to stabilize and optimize the fractionation of the product by adjusting the amount of fractionated product.

Next, a sample of the naphtha liquid is collected while performing such control, and the vapor pressure is analyzed from the sample using a vapor pressure measuring device or the like. When the vapor pressure is higher than the reference value, the control valve 111 and the like are opened to increase the amount of heating in the reboiler 112 to increase the amount of heat in the rectification column 100.

Then, the amount of LPG having a low boiling point is increased, the liquid level in the rectification column receiving tank 104 rises, and the liquefied LPG escapes to the outside. As a result, the LPG content of naphtha at the bottom of the rectification column decreases, and the vapor pressure of naphtha decreases.

When the vapor pressure is lower than the reference value, the control valve 111 and the like are throttled to reduce the amount of heat in the reboiler 112 to reduce the amount of heat in the rectification column 100. In this way, the vapor pressure is controlled.

[0021]

However, in reality, the operator collects and analyzes a sample of naphtha one by one and analyzes it, and further, the operator opens and closes a control valve and the like for increasing or decreasing the calorific value of the reboiler.

By the way, since a certain time is required until the analysis of the sample, it is necessary to suspend the operation of the plant until the analysis result is obtained, or to perform the full operation even if the operation is continued. Instead, the company only secured the production volume without being influenced by the analysis results.

Therefore, until the analysis result is obtained,
There is a problem that the productivity of the product is reduced. Next, usually, to understand the properties of oil, expensive analyzers are installed or samples are frequently collected and analyzed, but this requires a large amount of maintenance costs and requires a lot of labor. There is a problem of doing.

In recent years, a technique for estimating the properties of a plant from the state of temperature, flow rate, pressure and the like by a regression equation has been introduced in some cases. In addition, it is necessary to reconstruct the estimation formula every time there is a change in quality, and there is also a problem in management and operation that the construction of this estimation formula is difficult to deal with unless it is an experienced operator and cannot be handled easily. is there.

Further, when modifying the parameters of the estimation formula, it is necessary to refer to the average value and the previous results. However, since the quality estimation of the mixed base material has essentially no logic,
There is also a problem that it takes a lot of labor and time to correct the parameters.

Next, attempts have been made to estimate the properties of oil using neural network technology. However, at present, researchers use a specific computer and a unique method, so that any number of people can easily handle it. Not something.

In the estimation result of the vapor pressure of naphtha produced from the distillation apparatus by the conventional regression equation, the standard deviation of the difference between the test value and the estimated value is 0.0472 Kg / cm _2, which is inferior in accuracy. There is also a problem that there is.

The present invention was invented in order to solve the above-mentioned problems. If only the actual property values of the product obtained by tests and the like and the operating data of the manufacturing plant at that time are prepared, the product can be used. A property estimation equation can be constructed, and this estimation equation can be automatically and highly accurately constructed by a general-purpose computing device without using a specific computing device. It is a technical object to provide a manufacturing plant control system capable of directly controlling a plant by directly inputting it to a control device.

Further, the temperature, flow rate,
It is a technical object to provide a manufacturing plant control system capable of automatically controlling pressure and the like in accordance with a control command from a control device.

Further, it is an object of the present invention to provide a manufacturing plant control system capable of preventing a decrease in product productivity due to a sample analysis time, and requiring no expensive analyzer and capable of performing plant control at low cost. Is a technical issue.

[0031]

The present invention employs the following means to solve the above-mentioned problems. The manufacturing plant control system according to the present invention is used for a manufacturing plant control system that performs predetermined control on a manufacturing line that manufactures products from raw materials in a chemical device.

A measuring device for measuring at least one of temperature, flow rate and pressure at a predetermined position in the production line and outputting the measured value, a measured value outputted from the measuring device and stored, and the measured value An estimation formula creation device having estimation formula creation means for creating a property estimation formula for a fluid using a neural network by using a property value of the fluid corresponding to the value as estimation formula creation data, and an estimation formula creation device In addition to inputting the estimated expression, the target value of the fluid property is input, and the estimated value calculated by the estimation expression created using the measured numerical values input from the measuring device is compared with the target value to perform the comparison operation. Temperature at a predetermined location of the chemical device in the production line based on the results,
And a control device for outputting a control signal for controlling at least one of the flow rate and the pressure.

According to the present invention, at least one of a temperature, a flow rate, and a pressure (a numerical value obtained by calculation of a sum, a difference, a product, a ratio, and the like thereof) at a predetermined portion of a production line is measured by a measuring device. . Then, from the accumulated measured values and the property values of the actually measured fluid, the estimation formula creation data to be used for creating the estimation formula is selected, and the estimation formula creation device using the neural network selects the selected data. Using the estimation formula creation data, an estimation formula for the properties of the fluid is created.

Next, this estimation formula is input to the control device,
An estimated value calculated from the estimation formula using the measured value is compared with a preset target value. And
Based on the result of the comparison, all or any of the temperature, the flow rate, and the pressure at a predetermined location of the chemical device in the production line is controlled.

As described above, since the estimated value of the property of the fluid can be calculated from the equation for estimating the property of the fluid, it is not necessary to perform sampling at any time to analyze the property of the fluid.
There is no delay in control until the analysis result is obtained.

The chemical apparatus includes a petroleum refining apparatus in which a refined raw material is heated by a rectification device and petroleum is refined based on a difference in boiling point of the refined raw material. In addition, the chemical apparatus includes all apparatuses for fractionating raw materials or primary products into secondary products and final products, such as atmospheric distillation apparatuses, vacuum distillation apparatuses, debutane apparatuses, or deisobutane apparatuses. .

The measuring device for measuring the temperature, the flow rate and the pressure is preferably arranged at a place where the estimation formula for the properties of the product can be accurately prepared with the highest accuracy. Can be determined in a timely manner in consideration of the arrangement of the components.

Next, this production plant control system
The estimation formula creation device compares and calculates the error between the estimated value obtained from the measurement value used as the estimation formula creation data and the property value of the fluid corresponding to the measurement value, and the allowable error of the estimation value regarding the fluid input in advance, Neural network learning means is provided for learning by correcting the weight and threshold value of the network coefficient when the error is out of the range of the allowable error based on the result of the comparison operation.

Thus, when the error is out of the range of the allowable error, the weight of the network coefficient and the threshold value are successively corrected, and the error is always within the range of the allowable error to improve the learning accuracy.

In addition, the estimation formula creation device includes a measured value searching means for searching the measured values for the maximum value and the minimum value of the measured values, and a property value searching for searching the property values for the maximum and the minimum values of these property values. Means for selecting at least one of the maximum value and the minimum value searched by the measured value search means and the property value search means as a part of the estimation formula creation data.

Further, the estimation formula creation device uses the time of measurement, arbitrarily extracts a measurement value to be used from the measurement time and a property value corresponding to the measurement value, and estimates the measurement value and the property value. It has a selection means for selecting as formula creation data.

Next, the estimating formula creation device calculates the maximum error from among the calculated errors in a comparison operation between the estimated value obtained from the measured value used as the estimating formula creating data and the fluid property value corresponding to the measured value. And a deletion unit that deletes the measured value and the property value used for calculating the estimated value having the maximum error from the estimation formula creation data.

As described above, by deleting the measured value and the property value used for calculating the estimated value having the maximum error from the estimation formula creation data, the error can be reduced at any time and the learning accuracy can be improved.

Further, the estimation formula creation device uses the error between the estimated value obtained from the measurement value used as the estimation formula creation data and the property value other than the fluid property value corresponding to the measurement value as verification data, and A comparison unit for successively comparing verification data is provided, and a storage unit for storing an estimation formula that calculates an estimated value having the smallest error based on the comparison unit.

As described above, by storing the estimation formula that calculates the estimated value having the smallest error while successively comparing the verification data, the reliability of the estimation formula can be improved and the control accuracy can be improved.

[0046]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A manufacturing plant control system according to the present embodiment will be specifically described below with reference to FIGS.

This production plant control system controls a production line for producing products from raw materials using a chemical device, and uses a petroleum refining device as the chemical device.

As shown in FIG. 1, the structure of the control system for the production plant is as follows: a rectification column 1 for distilling a naphtha from a fluid distilled in an atmospheric distillation column and flowing into an atmospheric distillation column receiving tank 2.
And a measuring device T- for measuring the temperature and flow rate of the rectification column 1.
1, T-2, T-3, F-1, F-2, F-3 and measuring devices T-1, T-2, T-3, F-1, F-2, F-3
Using the measured values measured by the method and the property values of the fluid corresponding to the measured values as estimation formula creation data, an estimation formula creation device 3 for creating a fluid estimation formula 4, the estimation formula creation device 3 and the measuring device T- 1, T-2, T-3, F-1, F-2, F-
3 and a control device 5 for inputting the estimation expression 4 created by the estimation expression creation device 3 and comparing the estimated value calculated from the estimation expression 4 with a preset target value.

In this production plant, the fluid temporarily stored in the atmospheric distillation column receiving tank 2 is taken in through the valve 6 and is further heated by the heat exchangers 7 and 7 made of still to be partially vaporized to form the rectification column 1. The steam is allowed to flow in.

Here, the steam is decomposed into naphtha and LPG inside the rectification column 1, and the LPG is taken out from the top of the column and taken into the rectification column receiving tank 9 through the valve 8. The rectification column receiving tank 9 is for cooling LPG condensed and liquefied in the condenser 10.

Then, part or all of the liquefied LPG is returned to the rectification column 1 again. At this time, the liquid level inside the rectification tower receiving tank 9 is measured by the float type liquid level measuring device 11 or the ultrasonic sensor 11 utilizing the acoustic impedance of the ultrasonic wave to control the liquid level.

That is, when the liquid level is larger than the set value, the control valve 12 and the like are opened to remove the liquefied LPG out of the system. If the value is smaller than the set value, the control valve 12 and the like are squeezed and the liquefied LPG is returned to the rectification column 1 again.
Next, the refluxed liquefied LPG is forcibly returned to the rectification column 1 by a pump 13 provided in the middle of the pipe.
A control valve 14 and the like are provided so that the liquefied LPG does not reflux more than necessary to the rectification column 1 between the liquefied LPG and the rectification column 1.

Next, in the rectification column 1, the float type liquid level measuring device 15 is used.
The control of the liquid level is performed by, for example, the control of the liquid level. If the liquid level is larger than the set value, the control valve 16 is opened to allow the naphtha liquid to flow out of the system other than the gas absorption tower (not shown).

When the value is smaller than the set value, the control valve 16 is throttled to stop the outflow of the naphtha solution or to reduce the outflow amount. Further, the naphtha decomposed in the rectification tower 1 is caused to flow out of the system or flow into a gas absorption tower (not shown) through the valve 19.

Next, the estimation formula 4 created by the estimation formula creation device 3 for creating the fluid estimation formula 4 is input to the control device 5, and the control device 5 executes the measurement provided at a predetermined location in the manufacturing plant. Devices T-1, T-2, T-3, F-1, F-2, F
-3, an estimated value is calculated by the input estimation equation 4 using the measured value, a comparison operation between the estimated value and a preset target value is performed, and a control signal is output based on the comparison operation result. . In the present embodiment, the naphtha vapor pressure is used as the target value.

Then, the control device 5 transfers the control signal to a control valve 17 provided in the rectification column 1, and the control valve 17
The degree of opening and closing of the valve is adjusted based on the control signal from the control device 5, the flow rate of the heating fluid is changed, the heating amount of the reboiler 18 is adjusted as needed, and the increase or decrease of the heat amount in the rectification column 1 is changed.

The opening / closing degree of the control valve 17 and the flow rate of the heating fluid according to the opening / closing degree are also constantly input to the control device 5. Here, the control valve 17 adjusts the flow rate of the heating fluid into the reboiler 18, and the degree of opening and closing of the control valve 17 is adjusted so that the amount of heat applied in advance can be controlled.

Next, the rectification column 1 for further fractionating the fluid into naphtha performs an operation of causing the steam generated in the reboiler 18 to flow in countercurrent to a part of the condensate or to make a contact in a countercurrent stage. is there.

Then, as shown in FIG. 2, the fluid flowing through the pipe 20 is heated by the reboiler 18 to be turned into steam, and the steam generated by the reboiler 18 flows into the rectification column 1 through the pipe 21. I do.

Next, the vapor comes into contact with the reflux liquid from the top of the tower while rising inside the tower. And the LPG that came out of the tower
Is condensed and liquefied in the condenser 10 through the pipe 22 and flows into the rectification column 11. Part of the LPG condensate is taken out of the system as a distillate through the pipe 23,
The remainder is again forcibly returned to the inside of the tower from the top through the pipe 24 by the pump.

As described above, a part of or the whole of the condensate is returned to the inside of the column, so that the condensate comes into contact with the condensate from the top of the column, which is rich in low-boiling components, and the low-boiling components move into the vapor phase. As it rises, it becomes richer in lower boiling components. Also, the liquid becomes richer in higher boiling components as it descends in the column.

The flow rate of steam in the tower is determined by the heat transfer rate in the reboiler 18, the feed rate of the raw material, or the amount of liquid refluxed from the top of the tower. Next, a measuring device T- for measuring the temperature, flow rate and pressure at a predetermined location in the manufacturing plant
1, T-2, T-3, F-1, F-2, and F-3 are temperature sensors T-1, T-2, T-3, and a flow sensor F, respectively.
-1, F-2, F-3, and pressure sensors, and the temperature sensors T-1, T-2, and T-3 are roughly classified into a contact type and a non-contact type. A thermistor, a platinum resistance temperature detector or a thermocouple having a large width can be used. Further, as the non-contact sensor, a radiation thermometer that collects radiation energy by an optical system or an electronic circuit can be used.

Here, the temperature sensors T-1, T-2, T-
1, the temperature sensor T-1 is provided at the top of the rectification tower as shown in FIG. 1, and the temperature sensor T-1 is connected to a pipe for flowing fluid from the heat exchangers 7 and 7, which are stills, to the rectification tower. In addition, the temperature sensor T-3 is provided on a pipe through which a fluid flows from the reboiler 18 to the rectification tower.

Next, the flow sensors F-1, F-2, F-3
Uses a differential pressure type flow meter in which a throttle (orifice) is provided in a pipe and a flow rate is obtained from a differential pressure before and after the throttle. Furthermore, an electromagnetic flowmeter that measures the electromotive force in proportion to the flow velocity to determine the velocity, a hot-wire flowmeter that establishes a temperature-sensitive resistance line in the pipe through which the fluid passes, and determines the flow velocity by the temperature difference between two points, or the moving speed of the fluid An ultrasonic flowmeter utilizing the ultrasonic Doppler effect caused by the above can also be used.

Then, the flow rate sensors F-1, F-2, F-
The arrangement of 3 is such that a flow rate sensor F-1 is provided on a pipe 20 connected from the atmospheric distillation column receiving tank 2 to the heat exchangers 7 and 7 made of still, and the liquefied LPG is supplied from the rectifying column receiving tank 9 to the rectifying column 1. Along with providing the flow sensor F-2 in the piping 24 for reflux,
A flow sensor F-3 is provided on a pipe 26 for allowing naphtha taken out of the rectification column 1 to flow into the gas absorption column.

Although a pressure sensor is not provided in the present embodiment, a pressure sensor can be provided at an appropriate place when the estimation formula 4 is created by adding pressure to the estimation formula creation data.

Here, the pressure sensor for detecting the mechanical energy acting between the substances of the fluid uses the characteristics of a Bourdon tube for extracting a mechanical quantity as an electric signal and a strain gauge or a diaphragm for directly converting the mechanical quantity to an electrical quantity. The capacitance type pressure sensor described above can be used.

Next, the estimation formula creating device 3 for creating the fluid estimation formula 4 is a personal computer having a learning function using a neural network, and stores the measured values stored in advance and the properties of the fluid corresponding to the measured values. The estimation formula 4 is created from the values.

Here, the principle of the estimation formula creating device 3 is shown in FIG.
The measuring devices T-1 and T-
2, measured values of temperature and flow rate from T-3, F-1, F-2, and F-3 and property values corresponding to the measured values (property values used in the present embodiment are steam pressure tests. ) Is input to the estimation formula creation device 3.

Next, as necessary, the input measured values and the property values corresponding to the measured values are rearranged, and thereafter, an operation of creating the estimation formula 4 is performed. Note that these estimation equations 4 are created in three languages: CL language, FORTRAN language, and BASIC language, and are input to the control device 5.

Next, the control unit 5 calculates the estimated value (the estimated value used in the present embodiment is the vapor pressure value) by the estimation formula 4 using the measurement value input in advance, and sets it in advance. A comparison operation is performed with a target value (a target value used in the present embodiment is a vapor pressure value), and a control signal is output to a manufacturing plant.

In principle, the estimation formula 4 is created only once off-line in principle, and the created estimation formula 4 is used again except when the fluid to be produced is largely changed to an unprecedented value. 4 is not created.

The control device 5 continuously controls the manufacturing plant based on the estimation formula 4, and receives continuous operation data (measured values of temperature and flow rate) from the manufacturing plant. The measured value is stored in the memory, and the value of the vapor pressure estimated by the estimation formula 4 is continuously calculated using the measured value.

Next, in the neural network used to create the estimation formula 4, the signal given to the input layer is output as a unit value in the output layer while being converted by the weight of the connection between the layers.

However, with this only, a specific output pattern is output for a certain input pattern. Therefore, the weight is adjusted so that a desired output is output for the input pattern. A typical example of this adjustment method is a back propagation method.

Here, a desired output pattern is used as a teacher signal, and an error from the actual output pattern is propagated toward the input layer to change the weight between the respective layers so that the learning is performed in a direction to reduce the error.

In the learning stage, for a certain input pattern,
First, the output pattern obtained by operating the network is calculated. An error between the output pattern and a desired output pattern (teacher signal) is calculated, and the calculated error is propagated to the input layer as a learning signal to improve the weight of the connection. When the learning signal is calculated, a correction amount of the connection weight is determined, and the connection weight is corrected according to the correction amount. By repeating the above, the error between the output value from the output layer and the teacher signal is reduced.

The process described below simply explains the principle of such a neural network. Here, as shown in FIG. 4, the first layer is an input layer, the Nth layer is an output layer, and the N-2 layer between them is an intermediate layer. It is assumed that the connection between the layers is only the connection from the layer number n to the next layer n + 1, and that there is no jump connection.

The bonding between these layers is
Fully connected type, where all units of the layer are connected to each other
And Then, as shown in FIG. 5, the i-th unit in the n-th layer
Output value of knit is Xⁿ _iAnd the j-th unit of the (n-1) th layer
Weight of the connection from the packet to the i-th unit in the n-th layer is W
^{n, n-1} _{i, j}And

Next, considering the signal propagation from the input layer to the output layer as shown in FIG. 6, focusing on the i-th unit of the n-th layer, the output value X ⁿ of the i-th unit of the n-th layer is considered. _i becomes ^{u = ΣW n, n-1} i, j · X n-1 j ··· (1) X n i = f (u n i -h n i) ··· (2).

[0081] Here, u ⁿ _i is the internal state of the i-th unit of the n layer, h ⁿ _i is the threshold. Further, f is called a transfer function and is generally represented by f (x) = 1/1 + exp (ax) (3).

This function is also called a sigmoid function.
Is substituted with an appropriate real number, and usually 0 may be used. And X ⁿ _i
Is calculated by summing the product of the output value from the unit j of the previous layer n-1 and the weight specific to the connection path that this will follow from equation (1) for all the units of the layer n-1. I do. Then, the value u ⁿ _i is the internal state of the unit of interest.

[0083] obtain an output value X ⁿ _i by applying a value obtained by subtracting therefrom the threshold h ⁿ _i in the transfer function. The threshold h ⁿ _i from equation (1) _{^{u n i = ΣW n, n}} -1 i, j · X n-1 j -h n i ··· (4) X n i = f ( u ⁿ _i) a ... (5).

[0084] In addition, the -h ⁿ _i from equation (4), -h ⁿ _i
Unit for outputting a constantly 1.0 Given the × 1.0 is regarded as that which it is joined with a bond weight -h ⁿ _i. The signal propagates in this way, but only the output value from the input layer is output as it is to the input layer. That is, only the input layer has no relation to the transfer function and the threshold value.

Next, in the learning stage, an output pattern obtained by operating the network with respect to the first input pattern is calculated. An error between the output pattern and the desired output pattern is calculated, and the weight of the connection is corrected while propagating the error as a learning signal toward the input layer. Here, the desired output pattern is called a teacher signal.

Next, the way of giving learning is different between the learning signal returned by the output layer and the learning signal returned by the unit of the preceding layer. Learning signal returned from unit i in the output layer uses the output value and the teacher signal from the output layer as shown in FIG. _{^{7, δ N i = (d}} i -X N i) f '(u N i) (6)

Next, a learning signal returned by a layer before the N-1th layer is determined. learning signal [delta] ^N _i from unit i of the n-layer back towards the unit of n-1 _{^{layer, δ N i = f '(}} u N i) Σ k δ n + 1 k · W n + 1, n k _{, i} ... (7)

That is, as shown in FIG. 8, the learning signal from the (n + 1) th layer that is transmitted from the unit (i) of the nth layer to the (n + 1) th layer in the opposite direction and converges on the unit is
The sum is multiplied by the inherent weight of the combination. The product of this and the derivative of the transfer function is taken and treated as a learning signal.

After the learning signal is calculated in this manner, the correction amount of the weight of the connection is then calculated as ΔW ^{n, n-1} _{i, j} (t) = ηδ ⁿ _i X ^n-1 _j + αΔW ^{n, n-1} _{i , j} (t-1) (8).

Here, ΔW ^{n, n−1} _{i, j} (t) indicates the amount of correction to the weight of the connection between the unit j of the n−1 layer and the unit i of the n layer, and ΔW ^{n, n -1} _{i, j} (t-1) indicates the previous correction amount.

Here, since the convergence calculation is repeated, if this time is represented by t, the previous time is t-1. Η is a learning constant and is related to the speed of convergence. α is a stabilization coefficient, and has the effect of suppressing vibration at the time of convergence by using the correction amount of the previous weight. Then, appropriate values are given to η and α in the expression (8) within a positive real number range of 1.0 or less.

Next, when the correction amount is determined, W
^{n, n-1} _{i, j} (t + 1) = W ^{n, n-1} _{i, j} (t) + ΔW ^{n, n-1} _{i, j}
The weight of the connection is corrected by the equation (t).

Regarding the correction of the threshold value, it is considered that a signal is output from each unit that always outputs 1.0 with the weight of the connection corresponding to the threshold value. Therefore, a learning signal corresponding to this is obtained from the equations (6) and (7), and the following equation is obtained ^assuming that X ^n-1 _i = 1 in the equation (8).

[0094] _{^{Δh n i (t) = ηδ}} n i αΔh n i (t-1) Therefore, formula for _{^{modification, h n i (t + 1}} ) = h n i (t) + Δh n i (t) and Become.

The above procedure is repeated to reduce the error between the output value from the output layer and the teacher signal. Next, a procedure for creating the estimation formula 4 using the neural network according to the present embodiment will be described with reference to the flowchart in FIG.

First, reading from the memory and designation of a storage destination are performed (step 40), and a menu is selected from the operation list displayed on the display. Here, the operation list includes the continuous learning using the saved learning result file (step 6).
7), new learning (step 41), creation of a list of estimated values and actual values (test values of vapor pressure) (step 69), creation of estimation formula 4 from the learning result file (step 71), estimation and actual values (step 71). The vapor pressure test value) is plotted in the figure (step 73), the operation data (estimation formula creation data) is manually input, the estimation calculation is performed (step 75), and the processing is completed (step 7).
8) Each item is included.

When the new learning (step 41) is selected and executed, a screen for inputting necessary items is displayed (step 42). Here, the necessary items include the number of measured values accumulated during the operation of the plant (here, the temperature of the distillation apparatus, the flow rate of the raw material, the pressure), the number of operation data used for learning (estimation formula creation data), verification Number of operation data (accuracy confirmation data of estimation formula), tolerance,
The maximum number of times of learning and constants (α, ε) required for learning.

Next, when these values are input, it is selected whether or not to rearrange the operation data (estimation formula creation data) (step 43). When the rearrangement is performed, the rearrangement is performed according to the following procedure (step 44).

The maximum value and the minimum value of the actual value (test value of vapor pressure) to be estimated are selected, and these are always used as operation data (estimation formula creation data). Accumulated temperature,
The maximum and minimum values of flow rate, pressure, etc. are selected and used without fail as operation data (estimation formula creation data). The operation data (estimation formula creation data) is arbitrarily selected from the date of the accumulated value.

Here, the method of selecting the operation data (estimation formula creation data) arbitrarily from the date is determined by selecting the operation data (estimation formula creation data) at predetermined intervals, every few days, For example, it is selected every hour.

Next, when the rearrangement is completed or when returning to step 43 and selecting no rearrangement of the operation data (estimation formula creation data), the operation data (estimation formula creation data) is set to the learning data and the verification data. (Step 45).
Then, a random number is substituted for the weight and the threshold (step 46). When the above operation is performed, the estimation formula creation device 3 starts learning (step 47).

Next, a screen for selecting whether to forcibly stop the learning appears on the screen, and selects whether to forcibly stop the learning (step 48). Here, when the learning is not forcibly stopped, it is determined whether or not the learning data (estimation formula creation data) having the maximum error is excluded from the learning target (step 49).

Here, the maximum error is the maximum error among the errors calculated by comparing the estimated value obtained from the measured value used as the estimation formula creation data with the fluid property value corresponding to the measured value. Error. The learning data refers to a measured value and a property value used for calculating the estimation formula.

If no exclusion is to be performed, it is determined whether or not the current learning result is forcibly saved (step 5).
0). Next, when the forced saving is not performed, it is selected whether or not there is a forced check of the verification system (step 51).

Further, when the forced check is not performed,
The first to n-th learning data (estimation formula creation data) estimation calculations are performed (step 52). After performing the estimation calculation, it is determined whether or not the error is within the range of the allowable error (step 53).

Here, the error refers to the estimated value calculated from the estimation formula using the measurement value used by the estimation formula creation device as the estimation formula creation data, and the fluid property value (actual measurement value) corresponding to this measurement value. ), And the permissible error refers to the permissible error of the estimated value relating to the fluid input in advance.

If the error is out of the range of the allowable error in this judgment, the total weight and the threshold are corrected (step 54). Also, when the error is within the range of the allowable error or when all weights and thresholds are corrected in step 54,
It is determined whether the learning data is the n-th data (step 5).
5).

Here, if the learning data is not the nth one, the operation returns to step 52 and the operation is repeated. If the learning data is the nth one, it is determined whether or not the timing of the verification accuracy check is met. (Step 5
6) If the timing is correct, the operation data used for the verification data (here, the operation data is the vapor pressure, temperature, The verification accuracy is checked based on the flow rate and the pressure (step 57). Then, after checking the verification accuracy, it is determined whether the verification accuracy is the highest in the past (step 58).

If the verification accuracy is the highest in the past, the learning result file (estimation formula 4 for calculating the estimated value having the smallest error) is stored (step 59). Returning to step 56, if there is no timing for the verification accuracy check, if the verification accuracy is not the highest ever, or if the learning result file (estimation formula 4) is stored, It is determined whether or not the number of times of learning is equal (step 60).

Next, if the current number of times of learning and the maximum number of times of learning are not equal, the process returns to step 48 and the learning is repeated (step 48). Return to the menu and select the menu from the operation list. When the process returns to step 48 and the forced stop of learning is selected (step 48), the process returns to the main menu and selects a menu from the operation list. Further, returning to step 49 and selecting to exclude the learning data (estimation formula creation data) from the learning target (step 49), the operation of removing the learning data (estimation formula creation data) having the largest error at the present time from the learning target is performed. Is performed (step 61), and the process proceeds to step 52 to perform a learning operation (step 52).

Returning to step 50, if the forced saving of the learning result is selected (step 50), the forced saving is performed (step 62), and the process proceeds to step 52 to perform the learning operation (step 52). Then, returning to step 51 and selecting the forced check of the verification accuracy (step 51),
Proceeding to step 57, the verification accuracy of the verification data (test value of vapor pressure) is checked (step 57).

Next, returning to the main menu and selecting continuation learning (step 67) in the saved learning result file, the saved learning result file saved in the step is read (step 68), and the process proceeds to step 47 for learning. Is started (step 47).

Next, when a list of estimated values and actual values (test values of vapor pressure) is selected from the main menu (step 69), a list is prepared (step 7).
0), return to the main menu again. When the user selects to create the estimation formula 4 from the learning result file from the main menu (step 71), the estimation formula 4 is created in three languages, CL language, FORTRAN language, and BASIC language (step 72). Return.

Next, when the plotting of the estimated value and the actual value (test value of steam pressure) is selected from the main menu (step 73), the learning data (estimation formula creation data) and the verification data (test of steam pressure) are obtained. A plot is created for each value (step 74), and the process returns to the main menu again.

Further, when the operation data (estimation formula creation data) is manually input from the main menu and the execution of the estimation calculation is selected (step 75), the operation data input screen is changed and the operation data (estimation formula creation data) is input (step 75). Step 76), an estimation calculation is performed (step 77). Finally, when "end" is selected from the main menu, the process ends (step 78).

Next, the estimation equation 4 input using the measured values
The control device 5 that calculates an estimated value from the DCS uses a plant control computer (hereinafter, referred to as a DCS).
5 is a general name of "Distributed Contr."
ol ・ System.

Although not shown, the DCS body 5
And a keyboard and a display connected to the DCS main body 5, respectively. Furthermore, the DCS body 5
Is a central processing unit 5a (hereinafter referred to as CP) as shown in FIG.
U) and the main storage device 5c (hereinafter, referred to as memory)
And an input unit 5b and an output unit 5d.

Then, the CPU 5a fetches the instruction from the memory 5c to the CPU 5a based on the arithmetic unit for performing the four arithmetic operations, the logical operation, the magnitude comparison, and the like on the given data, and the address of the instruction to be executed. And a control unit for decoding the contents and outputting necessary operation instructions to other devices.

The control unit issues an input control command to the input unit 5b (S-30), issues a memory control command to the memory 5c (S-31), and issues a memory control command to the output unit 5d. Issue an output control command (S-32). Then, the command input from the input unit 5b is first transferred to the memory 5c (S-33). In the memory 5c, the data and the instruction are selected from the given command, and the selected data and the instruction are transmitted to the CPU 5a. (S
-34).

Thereafter, the control section decodes the data and the command transferred from the memory 5c and gives necessary operation instructions to the operation section (S-35). The arithmetic unit performs arithmetic operations such as four arithmetic operations, logical operations, and magnitude comparisons on the given data and instructions.

The data and instructions processed by the CPU 5a are fed back to the memory 5c again (S5).
-36), and transfers the result to the output unit 5d (S-3).
7). Then, the control unit repeats the process of [instruction fetching → decoding → address calculation → data fetching → instruction execution] to sequentially execute the instruction.

The DCS 5 stores the correlation between the amount of heat to be applied and the flow rate of the heating fluid corresponding to the degree of opening and closing of the control valve 17 in the memory 5c in the form of a map, and also sets a preset target target. The correlation between the value (the value of the vapor pressure) and the amount of heat is stored in the memory 5c in the form of a map.

Next, a control method performed by the DCS 5 will be described with reference to a flowchart of FIG. First, the estimation formula 4 created by the estimation formula creation device 3 is input to the DCS 5 (step 80). Next, the measured value used for calculating the estimated value is read from the memory (step 81).

Then, an estimated value (a vapor pressure value) is calculated by the estimation formula 4 using the read measured value (step 8).
2). Next, an actually measured value of the flow rate of the heating fluid to be controlled is always input (step 83). Here, the control valve 17
Calculating the flow rate of the heating fluid for heating the reboiler 18 based on the opening / closing degree of the
May be provided with a function of calculating the heat calories added to the heat.

Next, a preset target steam pressure value is read from the memory (step 8).
4). Then, a comparison operation between the calculated estimated value and the read target value is performed (step 85), and when the estimated value is within the range of the target value, the plant is operated as it is (step 86). . When the estimated value is out of the range of the target value, the opening / closing degree of the control valve is adjusted to change the flow rate of the heating fluid (step 87). Change the amount of heat in 1.

In this way, for example, when the estimated value is larger than the target value range, the control valve 17 is opened to increase the amount of heating in the reboiler 18 to increase the amount of heat in the rectification column 1.

As a result, the amount of LPG having a low boiling point withdrawn is increased, the liquid level in the rectification column receiving tank 11 is increased, and the liquefied LPG is increased.
Will escape to the outside. As a result, the LPG content of naphtha at the bottom of the rectification column 1 decreases, and the vapor pressure of naphtha decreases.

When the estimated value is smaller than the target value range, the control valve 17 and the like are throttled to reduce the amount of heat in the reboiler 18 to reduce the amount of heat in the rectification column 1. In this way, the vapor pressure is controlled.

[0129]

According to the manufacturing plant control system of the present invention, the estimation formula can be constructed by the estimation formula creation device if only the actual test values and the operation data of the manufacturing plant at that time are prepared. Therefore, there is no need for the operator to sequentially measure the vapor pressure of the fluid and construct an estimation formula based on the measurement result.

Further, it is possible to automatically construct a high-precision estimation formula with a normal computing device without any knowledge of constructing the estimation formula and without using a specific computing device. Further, by directly inputting the estimation formula to the control device, the control device can calculate an estimated value from the estimation formula, perform a comparison operation with a preset target value, and perform appropriate control on the plant. .

Next, since there is no need to limit the amount of production accompanying the analysis time of the sample, productivity can be efficiently improved, and there is no need to provide an expensive analyzer and the production cost can be reduced. Control over the line can be performed.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a manufacturing plant according to the present embodiment.

FIG. 2 is a configuration diagram of a rectification column according to the present embodiment.

FIG. 3 is a block diagram of plant control according to the present embodiment.

FIG. 4 is an explanatory diagram of a neural network according to the present embodiment.

FIG. 5 is an explanatory diagram of a neural network according to the present embodiment.

FIG. 6 is an explanatory diagram of a neural network according to the present embodiment.

FIG. 7 is an explanatory diagram of a neural network according to the present embodiment.

FIG. 8 is an explanatory diagram of a neural network according to the present embodiment.

FIG. 9 is a flowchart of the estimation formula creating apparatus according to the present embodiment.

FIG. 10 is a block diagram of a control device according to the present embodiment.

FIG. 11 is a flowchart of a control device according to the present embodiment.

FIG. 12 is a configuration diagram of a conventional manufacturing plant.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Rectification tower 2 ... Distillation tower receiving tank 3 ... Estimation formula preparation device 4 ... Estimation formula 5 ... Control device 6 ... Control valve 7 ... Heat exchanger 8 ...・ Control valve 9 ・ ・ ・ Rectification tower receiving tank 10 ・ ・ ・ Condenser 11 ・ ・ ・ Liquid level measuring device 12 ・ ・ ・ Control valve 14 ・ ・ ・ Control valve 15 ・ ・ ・ Liquid level measuring device 16 ・ ・ ・ Control Valve 17: Control valve 18: Reboiler 19: Control valve T-1: Temperature sensor T-2: Temperature sensor T-3: Temperature sensor F-1: Flow rate sensor F-2: Flow rate sensor F-3: Flow rate sensor 100: Rectification column 101: Distillation column receiver 102: Absorption tower 103: Condenser 104: Rectification column receiver 105: Liquid level measuring device 106: Control valve 108: Control valve 109: Liquid level control device 111: Control Valve 112 ... reboiler

Claims (7)

[Claims] 1. A production plant control system for performing a predetermined control on a production line for producing a product from a raw material in a chemical device, wherein at least one of a temperature, a flow rate, and a pressure at a predetermined portion of the production line is measured. A measuring device that outputs the measured value, and the measured value output from the measuring device and the property value of the fluid corresponding to the measured value are used as estimation formula creation data, and the neural network is used for the fluid. An estimation formula creation device having an estimation formula creation means for creating a property estimation formula; and an estimation formula created by the estimation formula creation device, and a fluid property target value input and input from the measurement device. The estimated value calculated by the estimation formula is compared with the target value by using the measured value obtained, and the production line is calculated based on the result of the comparison operation. A control device for outputting a control signal for controlling at least one of temperature, flow rate, and pressure at a predetermined location of the chemical device. 2. The production plant control system according to claim 1, wherein said chemical device is a petroleum refining device. 3. An estimation formula creation device, comprising: an error between an estimated value obtained from a measured value used as the estimation formula creation data and a property value of a fluid corresponding to the measured value; And a neural network learning means for learning by correcting the weight and threshold value of the network coefficient when the error is out of the range of the allowable error based on the result of the comparison operation. The manufacturing plant control system according to claim 1 or 2, wherein: 4. The estimation formula creation device, wherein: a measurement value search means for searching the measurement values for maximum and minimum values of the measurement values; and a search for the maximum and minimum values of these property values from the property values. Property value searching means, and selecting means for selecting at least one of the maximum value and the minimum value searched by the measured value searching means and the property value searching means as a part of the estimation formula creation data. The manufacturing plant control system according to any one of claims 1 to 3, wherein: 5. The estimation formula creation device uses a time of measurement of the measured value, arbitrarily extracts a measured value to be used from the measured time and a property value corresponding to the measured value, and calculates the measured value and the property value. 5. The manufacturing plant control system according to claim 1, further comprising a selection unit that selects the data as the estimation formula creation data. 6. 6. The estimation formula creation device according to claim 1, wherein in the comparison operation between the estimated value obtained from the measured value used as the estimation formula creation data and the fluid property value corresponding to the measured value, a maximum error among the calculated errors is calculated. 2. A maximum error search unit for searching for an error, and a deletion unit for deleting a measured value and a property value used for calculating an estimated value having the maximum error from the estimation formula creation data. A manufacturing plant control system according to claim 5. 7. The estimation formula creation device uses an error between an estimated value obtained from a measurement value used as the estimation formula creation data and a property value other than a fluid property value corresponding to the measurement value as verification data. And a comparing means for successively comparing the verification data, and a storing means for storing an estimation formula for calculating an estimated value having the least error based on the comparing means. 6. The manufacturing plant control system according to any one of 6.