WO2006126647A1

WO2006126647A1 - Petroleum refining plant parameter predicting device and parameter predicting method

Info

Publication number: WO2006126647A1
Application number: PCT/JP2006/310481
Authority: WO
Inventors: Masami Konishi; Tatsushi Nishi; Takashi Nukina; Katsuya Sato; Tatsukiyo Ikeda; Ryusaku Hasegawa; Yutaka Saito
Original assignee: National University Corporation Okayama University
Priority date: 2005-05-25
Filing date: 2006-05-25
Publication date: 2006-11-30
Also published as: JP2007004769A

Abstract

A prediction model optimized in respect of the prediction time and first and second parameters used for making a model is made, and a specific parameter of a petroleum refining plant is accurately predicted using the prediction model. The first and second parameters are measured, PLS (Partial Least Squares) analysis is conducted on at least part of the first and second parameters under model making condition, and a prediction model optimized in respect of the prediction time and the first and second parameters used for making the model. When a model update condition is met, an update command to update the optimized prediction model is outputted, an optimized prediction model is newly made according to the update command.

Description

Specification

Parameter prediction apparatus and parameter prediction method for oil refinery plant

[0001] The present invention creates a prediction model capable of predicting a specific parameter power that characterizes an oil refinery plant and a specific parameter ahead of a predetermined time, and uses this prediction model to predict a specific parameter with high accuracy. The present invention relates to an apparatus and a prediction method.

Background art

[0002] Once an abnormal state occurs in an oil refinery plant, it takes a lot of time and cost to resolve it. Therefore, there is a demand for a technology that can detect an abnormality ahead of a predetermined time in a plant of interest at an early stage.

[0003] As a method for predicting parameters of an oil refinery plant, there are a method based on a physical model of the plant and a method based on a mathematical model represented by a neural network model and a regression model.

[0004] In the physical model, phenomena occurring inside the oil refinery plant can actually be reflected in the model, but since there are many complicated phenomena occurring inside the plant, this is all taken as a physical model. It was difficult to express. In addition, there are problems that the amount of calculation becomes enormous and the prediction accuracy decreases.

[0005] In contrast, mathematical models are created based on available data collected from an oil refinery plant, eliminating the need for model building and tuning as with physical models. It can be said that the use of is a practically effective prediction method.

[0006] Various prediction methods have been disclosed as a technology for predicting parameters of such an oil refinery plant. For example, Patent Document 1 describes a plant abnormality diagnosis apparatus using a neural network. In this method, in order to ensure the accuracy of the estimation results used for abnormality diagnosis, a plurality of operating area force data are collected for a plurality of preset plant data and learned through a Yural network.

[0007] The parameter prediction method using the -Ural network described in Patent Document 1 is a plant that is as strongly related to the prediction target as possible in order to ensure and improve the prediction accuracy. It is necessary to model using data. However, in this prediction method, which type of plant data is used in which range is arbitrary, and is set mainly based on the knowledge and experience of the designer. For this reason, depending on the selected plant data, the degree of relevance to the prediction target is strong in one operating region, and the degree of relevance is weak in another operating region. May cause.

[0008] Further, in the oil refining plant, the parameters are intricately correlated with each other. For this reason, there is a linear relationship depending on the parameters selected (multicollinearity), and when using a regression model, the variance increases and the prediction accuracy may drop significantly.

[0009] Therefore, PLS (Partial Least Squares) is a mathematical model that can be accurately estimated even when there is no multi-collinearity between parameters without being affected by the parameters selected during model creation. ) Analysis is used.

[0010] For example, Patent Document 2 discloses a parameter prediction method using PLS analysis. This method creates a main model for property estimation using a statistical linear model such as PLS, and expresses the average output change. Specifically, when creating the main model, the input data used for the main model is selected using the accumulated data, and the delay time between the input parameter and the output parameter reflected in the main model is optimized as necessary. To do. In addition, the error between the actually measured output parameter and the output parameter of the main model is compensated using the error correction model. This error correction model is created by NN (Neura 1 Network) nonlinear model. The output parameters are predicted by the main model and the error correction model.

[Patent Document 1] Japanese Patent Laid-Open No. 9-113671

[Patent Document 2] US6, 278, 962

Disclosure of the invention

Problems to be solved by the invention

[0012] In order to perform PLS analysis, first measure the indicators (input data) used for the prediction calculation in the prediction model and the indicators (output data) predicted after a predetermined time from the time of measurement of the input data. . Then, the relationship between the input data and the output data is created as a prediction model. In the conventional PLS analysis, as the data used for the prediction calculation when creating this model, Therefore, it was necessary to determine the power to use which data and the expected time to predict the data to be predicted.

[0013] However, some of these data may contain specific outliers, and if a model is created using data containing outliers, the prediction accuracy may be greatly reduced. It was.

[0014] Therefore, as a result of intensive studies, the present inventors have been able to predict the parameters with high accuracy by optimizing the prediction time and the parameters used for model creation using PLS analysis. It is a thing.

[0015] In addition, depending on the data to be predicted, the result of PLS analysis may be greatly affected by disturbance and time series history of past input / output data. Therefore, the present inventors have found that RNN analysis is excellent as an analysis method that has a characteristic that it is strongly influenced by disturbances and is not easily affected by disturbances, and that can effectively reflect the influence of time series history of past data. did. Therefore, by combining the PLS analysis and the RNN analysis to predict the output data, we discovered that the parameters can be predicted with higher accuracy by effectively combining the characteristics of both.

Means for solving the problem

[0016] In order to solve the above problems, the prediction device of the present invention is characterized by having the following configuration. That is, the present invention uses the first and second parameters measured in the oil refinery plant to create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter. A prediction device for predicting a second parameter using the prediction model,

Input means for inputting first and second parameter measurement conditions and model creation conditions; measurement means for measuring first and second parameters from an oil refinery plant based on the first and second parameter measurement conditions;

Parameter storage means for storing the first and second parameters;

PLS (Partial Least Squares) analysis is performed on at least a part of the first and second parameters based on the model creation conditions, and the prediction optimized with respect to the prediction time and the first and second parameters used for model creation Model creation Dell creation means,

Model storage means for storing the optimized prediction model;

An output means for outputting a second parameter predicted by the optimized prediction model; an update condition input means for inputting a model update condition for updating the optimized prediction model;

Update command means for issuing an update command for updating the optimized prediction model when the model update condition is satisfied,

The present invention relates to a prediction apparatus configured to newly create an optimized prediction model when the update command is issued.

The present invention uses the first and second parameters measured at the oil refinery plant to create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter, and the prediction model is A prediction device that uses the second parameter to predict

Input means for inputting first and second parameter measurement conditions, model creation conditions and model correction conditions;

Measuring means for measuring the first and second parameters from the oil refinery plant based on the first and second parameter measurement conditions;

Parameter storage means for storing the first and second parameters;

At least a part of the first and second parameters based on the model creation conditions

Model creation means for performing PLS (Partial Least Squares) analysis and creating the prediction model optimized with respect to the prediction time and the first and second parameters used for model creation;

Based on the model modification conditions! A model correction means for generating a modified prediction model by further performing RNN (Recurrent Neural Network) analysis on the prediction model created by the model creation means and correcting the prediction model;

Model storage means for storing the modified prediction model;

Output means for outputting the second parameter predicted by the modified prediction model; An update condition input means for inputting a model update condition for updating the optimized prediction model;

When the update command is issued, the prediction device is configured to newly create an optimized prediction model and newly create a modified prediction model obtained by correcting the prediction model. About.

[0018] The present invention uses the first and second parameters measured in the oil refinery plant to create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter, A method for predicting a second parameter using a prediction model, wherein the first and second parameters are measured from an oil refinery plant based on input first and second parameter measurement conditions;

A process for storing the first and second parameters;

PLS (Partial Least Squares) analysis was performed on at least part of the first and second parameters based on the input model creation conditions, and the prediction time and the first and second parameters used for model creation were optimized. A process of creating the prediction model;

Storing the optimized prediction model;

A process of outputting a second parameter predicted by the optimized prediction model, and a process of inputting a model update condition for updating the optimized prediction model;

A process for issuing an update command for updating the optimized prediction model when the model update condition is satisfied,

The present invention relates to a prediction method characterized by newly creating an optimized prediction model when the update command is issued.

[0019] The present invention uses a first parameter and a second parameter measured in an oil refinery plant to predict a second parameter ahead by an estimated time from the time of measurement of the first parameter. A method of creating a Dell and predicting a second parameter using the prediction model, wherein the first and second parameters are measured from an oil refinery plant based on input first and second parameter measurement conditions. Process,

A process for storing the first and second parameters;

Based on the input model correction conditions, a process of generating a corrected prediction model by further correcting the prediction model by performing an RNN (Recurren t Neural Network) analysis on the prediction model;

Storing the modified prediction model;

A process for outputting a second parameter predicted by the modified prediction model; a process for inputting a model update condition for updating the optimized prediction model;

The present invention relates to a prediction method characterized in that when an update command is issued, an optimized prediction model is newly created, and a modified prediction model in which the prediction model is modified is newly created.

[0020] The first prediction apparatus of the present invention includes an input unit, a measurement unit, a parameter storage unit, a model creation unit, a model storage unit, an output unit, an update condition input unit, and an update command unit. In addition to these means, the second prediction apparatus further has a model correcting means.

[0021] Parameter measurement conditions (first parameter measurement conditions, second parameter measurement conditions), model creation conditions, and model correction conditions, if necessary, are input to the input means.

Here, the “first parameter measurement condition” is a condition for designating the measurement means used for measurement of the first parameter, the measurement interval, and the number of measurement points of each measurement means in the oil refinery plant. The

The “second parameter measurement condition” is a condition for designating a measurement means used for measuring the second parameter in the oil refinery plant. In this prediction device, the measurement interval of the second parameter and the number of measurement points of each measuring means are set to be the same as the value input as the first parameter measurement condition. Further, the number of parameters measured as “first parameters” is set to be larger than the number of parameters measured as “second parameters”.

[0024] “Model creation conditions” refer to the maximum time difference between the two parameters when the PLS analysis is performed using the measured first parameter and second parameter (maximum prediction). This is a condition that specifies the number of model configuration variables that make up the prediction model and the model that creates the model with time.

[0025] The "model correction condition" defines conditions for creating a corrected prediction model in which the prediction model obtained by PLS analysis is further corrected by RNN analysis. Specifically, RNN analysis is performed. When creating a modified prediction model, which residue is used in the PLS analysis, which first parameter is used, and which first parameter is used as the RNN solution. It is a condition that specifies the power used for prayer and the maximum number of intermediate layers.

These conditions do not need to be entered all, and some or all of the conditions may be set so as to be fixed together.

[0026] The measuring means includes a first measuring means capable of measuring the first parameter and a second measuring means capable of measuring the second parameter. The first measuring means measures the first parameter from the oil refinery plant at the measurement interval 'number of measurement points entered as the first parameter measuring condition. The second measuring means measures the second parameter from the oil refinery plant at the same measuring interval * number of measuring points as the first measuring means.

Here, the “first parameter” and the “second parameter” are parameters representing the characteristics of the oil refinery plant, and are not particularly limited as long as they represent the characteristics of the plant. For example, the pressure at a predetermined position in the plant, the temperature, the flow rate of the raw material 'product, etc., the product properties (flash point, initial fraction, distillation point) and the like can be mentioned.

[0028] The parameter storage means stores the first and second parameters measured by the first measurement means and the second measurement means, respectively. [0029] The model generation means performs PLS analysis on at least a part of the measured first parameter and second parameter, and is optimized for the first and second parameters used for prediction time and model generation. Create a predictive model.

[0030] The model correction means creates a modified prediction model by further modifying the prediction model created by the model creation means by further performing RNN analysis on the result of the PLS analysis.

The model storage unit stores the prediction model created by the model creation unit or the model correction unit. The output means outputs the second parameter ahead of the prediction time by using the optimized prediction model. The update condition input means inputs a model update condition for updating the optimized prediction model. The update command means outputs an update command for updating the optimized prediction model when the model update condition input by the update condition input means is satisfied.

[0032] It should be noted that in the present invention, it is only necessary for various means to be formed so as to realize the function. For example, dedicated hardware that exhibits a predetermined function, and a predetermined function provided by a computer program. It can be realized as a prediction device, a predetermined function realized in the prediction device by a computer program, a combination thereof, or the like.

[0033] Further, in the present invention, various means are formed as a plurality of members that do not need to be independent of each other, and some means may be a part of other means. It is possible that a part of the means overlaps with a part of the other means.

[0034] Further, the storage means of the present invention may be hardware in which a computer program for causing the prediction device to execute various processes is stored in advance. For example, ROM (Read Only) fixed to the prediction device. Memory (HDD), HDD (Hard Disc Drive), CD (Compact Dies) —ROM and FD (Flexible Disc—cartridge), which are interchangeably mounted on the prediction device, and combinations thereof can be implemented.

[0035] Further, the prediction device of the present invention may be hardware that can read a computer program and execute a corresponding processing operation. For example, a CPU (Central Processing Unit) is mainly used as a ROM, Good hardware such as RAM (Random Access Memory), l / F (Interface) unit, etc. [0036] In the present invention, causing the prediction device to execute various operations corresponding to the computer program also means causing the prediction device to control the operation of various devices. For example, storing various data in the prediction device means that the CPU stores various data in an information storage medium such as a RAM fixed to the prediction device, and is loaded in the prediction device interchangeably. For example, various data can be stored in the information storage medium with CPU power SFDD (FD Drive).

The invention's effect

[0037] The first prediction apparatus of the present invention constructs a prediction model optimized for the prediction time and the first and second parameters used for model creation by using PLS analysis, and sequentially executes the prediction model. While updating, it is possible to accurately predict specific parameters such as oil refinery plants. In addition, the first prediction device of the present invention is that the number of the first parameter is optimized when the prediction model is optimized, and the prediction model is updated according to the model update condition. This is different from the prediction device using PLS analysis by 2.

[0038] The second prediction apparatus of the present invention combines the PLS analysis and the RNN analysis to generate the characteristics of both analysis methods and to be influenced by disturbances. By reflecting the sound, it is possible to predict specific parameters such as an oil refinery plant with high accuracy. The second prediction device of the present invention is different from the prediction device that corrects the error of the prediction model using the NN analysis according to Patent Document 2 described above in that the error of the prediction model is corrected using the RNN analysis. . The RNN analysis according to the present invention is performed in consideration of the time delay of the first parameter due to the normal RNN, and the residual and evaluation value of the second parameter by the PLS analysis is considered! / ヽThe

Brief Description of Drawings

FIG. 1 is a schematic diagram showing an example of an internal structure of a prediction device of the present invention.

FIG. 2 is a schematic diagram showing a distillation apparatus.

FIG. 3 is a conceptual diagram showing a measurement state of parameters of the present invention.

FIG. 4 is a conceptual diagram showing a sampling state of parameters of the present invention.

FIG. 5 is a conceptual diagram showing a sampling state of parameters of the present invention. FIG. 6 is a schematic diagram showing the internal structure of the prediction device of the present invention.

FIG. 7 is a flowchart showing a process of creating an optimized prediction model.

FIG. 8 is a schematic diagram showing an example of the internal structure of the prediction device of the present invention.

FIG. 9 is a conceptual diagram showing the RNN analysis method of the present invention.

FIG. 10 is a flowchart showing a process of creating an optimized prediction model.

FIG. 11 is a diagram showing measurement positions of parameters in the distillation column in Examples.

Explanation of symbols

[0040] 50 Oil refinery plant

51 Input means

52 First measurement means

53 Second measuring means

54 Parameter storage means

55 Model creation means

56 Model storage means

57 Output means

58 Input section

59 Update command means

61 Predetermined time

62 Sampled first parameter

64 Second parameter sampled

71 Model correction method

BEST MODE FOR CARRYING OUT THE INVENTION

[Embodiment 1]

The first prediction apparatus of the present invention includes input means, measurement means (first measurement means, second measurement means), parameter storage means, model creation means, model storage means, and output means. Each component means is connected by a cable so that data can be exchanged.

FIG. 2 shows an example of an oil refinery plant that performs parameter prediction using the prediction device of the present invention. A distillation column is shown. This distillation tower 14 is an apparatus for separating crude oil into fractions such as light gasoline, heavy gasoline, kerosene, light oil, heavy light oil, in addition to the top gas and bottom oil. In the distillation column 14, raw materials are supplied from the predetermined positions 11 to 13 to the preliminary distillation columns 15 to 17, respectively. The pre-distillation towers 15 to 17 are separated into tower top gas 23 to 25 and tower bottom oil 26 to 28 from the supplied raw materials and components discharged from predetermined positions 20 to 22 of the distillation tower, respectively. The top gas 23 to 25 is returned to the distillation column 14 again, and the bottom oil 26 to 28 is taken out as a product. The raw material is also supplied from a predetermined position 19 near the tower bottom. Further, a tower top gas 29 and a tower bottom oil 30 are taken out from the distillation tower 14. The intermediate material in the distillation column is transported from 31 to 33 to 34 to 36 by the compressor!

[0043] In this distillation column 14, the measuring means are arranged at 11 to 13 and 15 to 22, and are arranged so as to measure the characteristics of 23 to 30 tower top gas or tower bottom oil. Of these measuring means, the first measuring means measures the first parameter and the second measuring means measures the second parameter.

[0044] It should be noted that the parameters include the raw material supply flow rate, the product discharge flow rate, temperature, pressure, parameters calculated by combining these parameters, parameters obtained by performing special arithmetic processing on these parameters, etc. Can be used. In addition, if these combinations are calculated, special calculation processing can be performed by calculation means built in the measurement means. These parameters can be arbitrarily assigned as the first and second parameters.

1S Since the first parameter is used to predict the second parameter, the first parameter must be a parameter that represents the state ahead of the second parameter in the distillation column. Also, the first parameter and the second parameter cannot overlap, and the number of first parameters must be greater than the number of second parameters.

[0045] For example, in the distillation column of FIG. 2, the parameters obtained by measuring the characteristics of the bottom oil 30 and the top gas 29 are parameters representing the state of the product finally obtained in the distillation step. It can only be used as a two parameter. In addition, the parameters measured at the predetermined positions 11 to 13 and 19 are parameters representing the state of the raw material supplied first, and can only be used as the first parameter! /.

[0046] The PLS analysis method of the present invention may use some or all of the above parameters. good. The parameters are not limited to those shown in FIG. 2, and are not particularly limited as long as they represent the characteristics of the distillation column. Since PLS analysis is used for model creation in the present invention, it is possible to predict the value of a specific parameter for a predetermined time with high accuracy regardless of the parameters used for model creation.

The details of the processing steps performed by the prediction apparatus of the present invention will be described with reference to FIG. First, parameter measurement conditions (first parameter measurement condition, second parameter measurement condition) and model creation conditions are input to the input unit 51. As the first parameter measurement conditions, the measurement means for measuring the first parameter in the oil refinery plant, the measurement interval, and the number of measurement points to be measured by each measurement means are entered.

As the second parameter measurement condition, a measurement means for measuring the second parameter in the oil refinery plant is input (in this embodiment, the case where there is one kind of second parameter) is input. The model creation conditions include the maximum time difference (maximum prediction time) between the first parameter and the second parameter for PLS analysis, and the number of model configuration variables that make up the prediction model. input.

[0048] Next, based on the first parameter measurement conditions input to the input means, the predetermined first measurement means 52 arranged in the oil refinery plant 50 has the first parameter of the predetermined measurement interval and the number of measurement points. Measure. In addition, the second measuring means 53 is configured so that the predetermined second measuring means 53 disposed in the oil refinery plant 50 performs the measurement of the first measuring means 52 on the basis of the second parameter measuring condition input to the input means. Synchronously measure the second parameter with the same measurement interval-number of measurement points as the first measurement means. The measured data of the first parameter and the second parameter are conceptually shown in Fig. 3 (Fig. 3 shows the case of 36 measurement points).

[0049] The first parameter and the second parameter measured in this way are stored after being corrected by the parameter storage means as necessary. As a method for correcting the data of the first parameter X, the method of equation (1) can be mentioned.

[0050] [Equation 1]

One (n = l, 2,-.. M; p = l, 2 ... Q) [0051] where x * is the nth value measured in the pth type among the x types that were measured at the Mth point for each of the Q types, X is the average value of the Xth value for the pth type measured at the M point, σ represents the standard deviation of the ρth type of X.

[0052] Next, PLS analysis is performed on the measured first parameter and second parameter by the model creation means, and the prediction time and the prediction optimized for the first and second parameters used for model creation are performed. Create a model. The process for creating this prediction model will be described with reference to FIG.

First, a plurality of temporary prediction models are created by the temporary model creating means. In this process, first, specify which part of the measured first parameter and second parameter to use as data when creating the prediction model (data area).

[0054] In order to specify this data area, it is necessary to determine the prediction time. First, the prediction time is changed stepwise within the range of the maximum prediction time specified in the model creation conditions input in advance. (Sl, S2). However, the estimated time must be less than or equal to the time represented by (measurement interval) X {(number of measurement points) -1} entered as the first parameter measurement condition. The rate of change in steps must be at least a multiple of (measurement interval) and less than (measurement interval) X (number of measurement points) Z2.

[0055] For example, if the measurement interval is 1 minute and the number of measurement points is 100, (measurement interval) X {(number of measurement points) — 1} = 1 X (100— 1) = 99 minutes, and the estimated time is Can vary from 1 to 99 minutes. For example, if the prediction time is changed every minute (measurement interval), the generated prediction time is 1, 2, and 3 99 minutes. Also, if you change the forecast time every 2 minutes,

1, 3, 5. 99 minutes.

[0056] Sampling of the first and second parameters is performed according to a predetermined condition for each predicted time thus changed. To do this, first specify the data area for sampling. Examples of specified data areas are shown in Fig. 4 (a) to (c). In Fig. 4, the data area is represented by a gray area. 4 (a) to 4 (c) show data areas when different prediction times 61 are generated by the above process. At this time, the last part 63 of the data area of the second parameter is set to coincide with the end of the measurement, and the measurement interval and the number of measurement points are specified in advance as the parameter measurement conditions. If determined, the data area of the first parameter and the second parameter is automatically determined.

[0057] The data area of the first parameter and the data area of the second parameter may overlap each other as shown in Fig. 4 (a) or contact as shown in Fig. 4 (b). May be so far away.

Next, the first parameter used for creating the prediction model is sampled under a predetermined condition of the data area force of the first parameter specified in this way (S 3). The sampling conditions are not particularly limited. For example, the sampling method may be one or more times at a sample number that falls within a predetermined range, a method of sampling at random, a method of sequentially sampling at a predetermined measurement interval, or other sampling according to a predetermined law. Can be mentioned.

[0059] Preferably, random sampling is performed. In random sampling, sampling is not performed according to the prescribed law, so it is difficult to capture abnormal values. In addition, the number of samples can be set freely. The sampling condition may be input as a model creation condition, or a predetermined condition may be set in the prediction device in advance. Figure 5 (a) shows randomly sampled data62.

Next, the second parameter 64 ahead is sampled from the first parameter 62 sampled in this way for the estimated time determined by the above-described processing (FIG. 5 (b): S3). As the second parameter, the parameter that is ahead of the first parameter by the estimated time 61 is specified, so if the first parameter 62 is determined, the second parameter 64 to be automatically sampled is determined.

Next, the first parameter force sampled in this way also creates a first parameter matrix X (Equation (2)) and a second parameter force second parameter matrix Y (Equation (3)).

[0062] [Equation 2]

[0063] where X is the first parameter sampled nth in the pth type, Y is nth

np n

Indicates that it is a sampled second parameter. Next, a model configuration variable matrix (equation (4)) having the same power as the sampled first parameter is created (S4).

[0064] [Equation 3]

Here, W represents the first model configuration variable. After this, this first parameter

1 Create a matrix t expressed by the product of the matrix matrix X and the model component matrix (4), and calculate the inner product φ of the second parameter matrix Y and the matrix t.

[0066] [Equation 4]

[0067] The larger the inner product φ is, the more the second parameter predicted using the model configuration variable W

1

It can be evaluated that there is little error with the measured second parameter. Therefore, the model configuration variable W that maximizes Φ is obtained. Specifically, the model configuration variable W is set to norm 1 (w ²

(1 1 1 1 = 0)

Is determined by the undetermined multiplier method under the conditions (S5).

[0068] [Equation 5]

J = Y ^T -XW,-λ \ ν "'W.-l) (7)

[0069] Next, let Y ⁽¹⁾ be the パラメータ part of the second parameter matrix Y that cannot be represented by the model configuration variable W

1

And Υ ⁽¹⁾ can be expressed from W and its weighting factor r (Equation (10)). Y ⁽¹⁾ and r XW are

1 1 1 1 From the intersection, r expressed by equation (11) is determined.

1

[0070] [Equation 6] = Y-r ₁ XW _l (10)

'

[0071] When r expressed by Eq. (11) is used, the model configuration variable W and the first parameter X are used for prediction.

1 1

The measured second parameter Y (= r XW) can be expressed.

1 1

Here, when the number of model configuration variables input in advance as model creation conditions is 2 or more (S6, S7), the next model configuration variable W is obtained. The following is the model configuration variable W

2 2 shows the process. First, let X ⁽¹⁾ and Y ⁽¹⁾ be the parts of the Y created using the sampled first and second parameters that cannot be represented by the model configuration variable W.

1

α ⁾ and Υ ⁽¹⁾ are expressed as equations (12) and (13) (S8).

[0073] [Equation 7] t, ^T X

2 (12)

[0074] Next, by substituting Y ⁽¹⁾ and X ⁽¹⁾ as Y → Y ⁽¹⁾ and Χ → Χ ^{(1) into} Eqs. (2) to (11), the model configuration variable W Ask for. Number of model configuration variables entered as model building conditions

2

If is 3 or more, the process is performed in the same way. Then, the process ends when the number of model configuration variables input as model creation conditions is reached.

That is, when the number of model configuration variables input as model creation conditions is Κ, processing is performed as follows to obtain model configuration variables sequentially from W to W.

1 Κ

[0076] [Equation 8] X ⁽⁰⁾ = X -X

y ^((,) = o N

(k = l, 2 .... K

[0077] The following x ^(Q) and Y ^(Q) can be set as appropriate according to the type of data to be used.

Each r XW is calculated from each model configuration variable W obtained in this way, and each r XW is used.

k k k k k can be used to create a temporary prediction model. Through the above process, the provisional model creation means creates a plurality of provisional prediction models in which the first and second parameters are sampled according to a predetermined condition for each changed prediction time. Next, for each temporary prediction model, the first parameter used to create the temporary prediction model is input, and the second parameter ahead by the prediction time is predicted. Then, an optimum model with the smallest error between the predicted second parameter and the sampled second parameter is determined from the created temporary prediction models.

A statistical index can be used for error evaluation. For example, the following indicators can be used as statistical indicators.

[0079] [Equation 9]

[0080] where is (second parameter predicted by provisional prediction model)-(sampled second parameter), e is {(sampled second parameter)-(average value of sampled second parameter) } The average value of}, y is the sampled second parameter, and y is the average value of the sampled second parameter. These statistical indicators may be used alone or in combination to evaluate errors.

Thus, the temporary prediction model evaluated as the model with the smallest error by the statistical index is set as the optimal prediction model and stored in the model storage unit. Further, when the output command is input, the output means predicts a specific parameter ahead by the prediction time by using the optimum prediction model stored in the model storage means.

[0082] (Embodiment 2)

The second prediction apparatus of the present invention further includes an update command means, and is configured to update the model by creating a prediction model again when the model update condition previously input to the update condition input unit is satisfied. ing. Thus, the second parameter can be accurately predicted over a long period of time by updating the prediction model according to the model update condition.

This process will be described with reference to FIG. The update command means 59 has an update condition input unit 58 for inputting model update conditions. Then, an update command is issued to update the prediction model when the model update condition is satisfied. When this update command is issued, the first measuring means and the second measuring means automatically start measuring the first parameter and the second parameter, respectively, and the prediction model optimized again by the process of Embodiment 1 above. Create

[0084] The update condition includes, for example, a case where the prediction model is automatically updated every predetermined time. In this case, the predetermined time can be calculated from the time when the prediction model is created. In addition, the second parameter is predicted periodically, and the second parameter is actually measured at the time when the second parameter was predicted (the time advanced by the predicted time from the time of measurement of the first parameter used for prediction). When the error between the predicted second parameter and the measured second parameter is calculated using a statistical index such as equations (16) to (18) and the error exceeds a predetermined value, It may be a case where an update command is issued after determining the state. [0085] (Embodiment 3)

The third prediction device of the present invention is the one in which the first and second prediction devices are further provided with model correction means. This prediction device performs further RNN analysis on the result of PLS analysis. RNN analysis has characteristics different from PLS analysis, such as being less susceptible to disturbances and being able to effectively reflect past input / output data time series history in output data. For this reason, it is possible to construct a prediction model with further improved prediction accuracy by performing RNN analysis on the prediction model that has been subjected to PLS analysis. In the RNN analysis, the first parameter is used as an input, but in addition to the first parameter used in the PLS analysis, multiple evaluation values represented by equations (16) to (18) may be used! The reason for using this evaluation value is that it is possible to calculate the evaluation value shown in Eqs. (16) to (18) by using the residue of the predicted value and the actual value obtained by PLS analysis. This force is considered to be effective in improving accuracy.

FIG. 8 shows an outline of a prediction apparatus in which model correction means is further added to the second prediction apparatus. This prediction apparatus has input means, measurement means (first measurement means, second measurement means), parameter storage means, model creation means, model correction means, model storage means, output means, and update command means.

In the prediction apparatus of FIG. 8, the prediction model created by the model creation means 55 is further obtained from the model correction condition input to the input means 51 and the measured value of the first parameter stored in the parameter storage means. The corrected predicted model force is generated by the model correcting means 71. This modified prediction model is stored in the model storage means 56 and can be output by the output means 57 as necessary.

The method of creating a modified prediction model that combines this PLS analysis and RNN analysis is described in detail below.

FIG. 10 is a flowchart showing the process of RNN analysis. First, the residue Y ^(k) (Equation (14)) that cannot be expressed by the model configuration variable W among the sampled second parameters is calculated from the PLS analysis method as in the first or second embodiment. Then get in this way k

Determine the type of residue to be used in the creation of a modified prediction model by RNN analysis. This residue is all residues obtained by PLS analysis. Some residue may be sufficient. When some residues are used, for example, the intermediate force of all residues obtained by PLS analysis can be determined by random sampling (Sl l). In this case, the number of samples should be set freely as long as it does not interfere with the RNN analysis.

Can do.

[0089] Next, a first parameter used for creating a prediction model by RNN analysis is determined. The first parameters may be all the first parameters used for model creation by PLS analysis or some of the first parameters. When some first parameters are used, for example, sampling can be performed randomly (Sl l). In this case, the number of samples can be set freely within a range that does not interfere with the RNN analysis.

[0090] Next, RNN analysis is performed using the residue Y ^(k) determined as described above and the first parameter X. Figure 9 shows an overview of this RNN analysis. In addition to the first parameter X, it is also possible to perform RNN analysis using a plurality of evaluation values represented by equations (16) to (18). The RNN analysis consists of three processing processes (virtual regions): an input layer, an intermediate layer, and an output layer. The residue and the first parameter determined as described above are first input to the input layer. Is done.

[0091] In the input layer, Y ^(k) 'd, which is the residue Y ^(k) multiplied by the forgetting rate d, is calculated, and this Y ^(k) ' d and the first parameter are added to the sigmoid function (Equation A). The entered calculation results are output (formulas B and C). Here, the forgetting rate d is set to a value of 0 <d≤1 depending on conditions.

Next, the number of intermediate layers is sequentially changed one by one from 1 to the maximum number of intermediate layers previously input as model correction conditions (S12 to S17). Next, a weight function W corresponding to the number of intermediate layers is defined, and a load sum z is output. For example, when the number of intermediate layers is 1, the weighting function is WG ⁾ and the load sum z (formula D) is output as shown in FIG. If the number of intermediate layers is 2, the weight functions for the first and second intermediate layers are W ⁽¹⁾ and W ⁽²⁾ , the load sum is z and _Z , and the number of intermediate layers is 3, The weight functions for the first, second, and third intermediate layers are W ⁽¹⁾ , W ⁽²⁾ , W ⁽³⁾ and the sum of loads is z, z, z ... The weight function W and load sum of the intermediate layer are defined according to

[0093] Next, in the output layer, the load sum is calculated from the load sum z and the weight function W output from the intermediate layer. u is output. For example, if the number of intermediate layers is 1, the weight function of the output layer is W ⁽

jk

²⁾ The load sum u (Equation E) is output. When the number of intermediate layers is 2, the output layer weight function

k

If the number is W ^(¾ , the load sum is u and the number of intermediate layers is 3, the weight function of the output layer is W ⁽⁴⁾ , the load sum is kl 1 lm u... Output layer weight function W and load sum u are defined m

It is.

[0094] [Equation 10]

1

Φ () =-~-; Φ (-∞) = 0, Φ∞) = 1 (A), = Φ („) (l≤n≤m-l) γ _η = φ (ά · Υ _η ^{ " ) (m≤n≤M) = ^φ ∑) () = Φ (∑) ( ^Ε )

[0095] However, it was used to input (formula B), the (Formula C), each input layer (m-1) and the first parameter X species (M-m) types of residues Y ^0. M, n, and M are natural numbers.

[0096] Thereafter, the weight function W of the intermediate layer and the output layer in the RNN analysis for each number of intermediate layers is determined so that the load sum u output from the output layer is minimized. As a method of obtaining these weighting functions W, a force that can use various known methods, for example, a BPTT method, an RTRL method, or the like can be used. Next, a temporary correction prediction model is created for each RNN analysis of each number of intermediate layers using the weight function W thus obtained. Next, we evaluate the errors in each of these temporary correction prediction models. For example, statistical indexes such as equations (16) to (18) can be used for error evaluation.

[0097] Temporary correction prediction model evaluated as a model with the least error by the statistical index as described above, PLS analysis is determined as a correction prediction model by correcting the prediction model by combining RNN analysis, and the model storage means Remembered. Further, when an output command is input, the output means predicts a specific parameter ahead by the prediction time by using the modified prediction model stored in the model storage means.

Note that in this prediction device, the update command means is further connected to the update condition input unit in advance. When the input model update condition is satisfied, the prediction model may be created again by the model creation unit and the modified forecast model may be created by the model correction unit. Thus, the second parameter can be accurately predicted over a long period of time by updating the prediction model according to the model update condition. The model update process can be performed by the same procedure and method as the second prediction device.

Example

[0099] (Experimental method)

With the prediction device of the present invention, the parameters (characteristics) of the distillation tower in the Mizushima Refinery, Nippon Oil Refining Co., Ltd. were predicted. First, sensors were attached to 28 locations in the distillation column, and the characteristics such as temperature, pressure, and raw material flow rate were measured by each sensor. Figure 11 shows the types of parameters measured and their positions in the distillation column. In this measurement, 1200 points of the first parameter (X to X) and the second parameter (Y) were measured every 3 minutes.

1 27 1

[0100] Sample of 1200 points, measured first parameter (X to X), second parameter (Y)

1 27 1

Of these, at least a part of the first 1 to 500 points was used for analysis and a prediction model was created. Next, 501 to 1000 points of data were predicted from 501 to 1200 points using this prediction model. The prediction procedure is shown below.

[0101] Out of 500 points, 200 points were extracted and analyzed with a predicted time of 15 minutes (delay time of 5 samples). In Example 1, the second parameter of the 501st to 600th points was predicted by PLS analysis with 5 latent variables. In Examples 2 and 3, a combination of PLS analysis and RNN analysis was performed. At this time, the number of latent variables in PLS analysis was set to 5, and the number of intermediate layers in RNN analysis was set to 10. In Example 2, the second parameter of the 501st to 600th points was predicted, and in Example 3, the second parameter of the 601st to L000th points was predicted.

[0102] From the error with respect to the actual measured value Y of the second parameter predicted as described above,

1

Dell accuracy was evaluated. The error was evaluated by RMSE (Equation (16)) and INDEX (Equation (18)). These results are shown in Table 1.

[0103] [Table 1] Child measurement time Latent change Middle layer RMS EI ND EX analysis method

(Minutes) number of numbers number (one) (one)

Example 1 PLS 15 5 1. 967 1 12.14 Example 2 P L S + R N N 15 5 30 1. 636 77. 59

'Detail 3 PL S + RNN 15 5 30 2. 165 59. 48 From the results of Example 1, it is clear that the second parameter can be predicted with high accuracy by the prediction model created by PLS analysis. In addition, the results of Examples 1 and 2 show that the prediction accuracy is further improved by using PLS analysis and RNN analysis compared to the case of PLS analysis alone. The results of Example 3 show that this prediction model is the 601st to 1000th points, and that the prediction accuracy is maintained even when a certain amount of time has passed after the prediction model is created.

Claims

The scope of the claims

[1] Using the first and second parameters measured at the oil refinery plant, create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter. A prediction device for predicting a second parameter using:

Parameter storage means for storing the first and second parameters;

Model storage means for storing the optimized prediction model;

A prediction device characterized by being configured to newly create an optimized prediction model when the update command is issued.

[2] The prediction device according to claim 1, wherein the model update condition is a case where a predetermined time elapses when the optimized prediction model is created.

[3] The output means periodically predicts the second parameter,

The measuring means measures the second parameter at a time when the second parameter is predicted, and the model update condition is that an error between the predicted second parameter and the measured second parameter is abnormal due to a statistical index. It is a case where it is determined that the condition is present.

The prediction device according to 1. [4] Using the first and second parameters measured at the oil refinery plant, create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter. A prediction device for predicting a second parameter using:

Parameter storage means for storing the first and second parameters;

Model storage means for storing the modified prediction model;

An output means for outputting a second parameter predicted by the modified prediction model; an update condition input means for inputting a model update condition for updating the optimized prediction model;

When the update command is issued, the prediction device is configured to newly create an optimized prediction model and newly create a modified prediction model obtained by correcting the prediction model. .

[5] The model creation means includes:

The prediction time is changed within the range of the prediction time specified by the model creation condition, and the first and second parameters are set according to a predetermined condition from the measured first and second parameters for each changed prediction time. Sampled and said sampled first And provisional model creation means for creating multiple provisional prediction models by performing PLS (Partial Least Squares) analysis on the second parameter,

Model decision means for evaluating the accuracy of each temporary prediction model and determining the optimum prediction model;

The prediction apparatus according to claim 1, comprising:

[6] The sampling of the first parameter according to the predetermined condition is a predetermined number of random samplings,

6. The sampling of a second parameter according to the predetermined condition is a sampling of a second parameter measured earlier by the estimated time than the sampled first parameter. Prediction device.

[7] The prediction device according to claim 5 or 6, wherein the accuracy of each temporary prediction model is evaluated using a statistical index.

[8] The model creation means further includes:

The prediction apparatus according to claim 1, wherein a prediction model in which the number of model configuration variables constituting the prediction model is optimized is created.

[9] The oil refinery plant is a distillation tower,

Each of the first and second parameters is independently at least one kind of selected parameter, pressure, temperature, feed flow rate of raw material, and group discharge force of product discharge flow rate in a predetermined region of the distillation column. The prediction apparatus according to any one of claims 1 to 8, wherein the prediction apparatus is one.

[10] Using the first and second parameters measured at the oil refinery plant, create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter. A method for predicting a second parameter using:

A process for measuring the first and second parameters from the oil refinery plant based on the input first and second parameter measurement conditions;

A process for storing the first and second parameters;

PLS (Partial Least Squares) analysis is performed on at least a part of the first and second parameters based on the input model creation conditions, and the prediction time and model A process of creating the predictive model optimized with respect to the first and second parameters used for creation;

Storing the optimized prediction model;

A prediction method, wherein an optimized prediction model is newly created when the update command is issued.

Using the first and second parameters measured at the oil refinery plant, create a prediction model that predicts the second parameter ahead by the prediction time from the time of measurement of the first parameter. A method for predicting two parameters,

A process for storing the first and second parameters;

Storing the modified prediction model;

A process for outputting a second parameter predicted by the modified prediction model; a process for inputting a model update condition for updating the optimized prediction model; A process for issuing an update command for updating the optimized prediction model when the model update condition is satisfied,

A prediction method characterized in that, when the update command is issued, an optimized prediction model is newly created, and a modified prediction model in which the prediction model is corrected is newly created.

[12] The process of creating the predictive model is:

A process of changing the prediction time within a range of the prediction time specified by the model creation condition;

A process of sampling the first and second parameters from the measured first and second parameters according to a predetermined condition at each changed predicted time;

A process of generating a plurality of temporary prediction models by performing PLS (Partial Least Squares) analysis on the sampled first and second parameters;

The prediction method according to claim 10 or 11, further comprising: a process of evaluating an accuracy of each temporary prediction model and determining an optimal prediction model.

[13] The sampling of the first parameter according to the predetermined condition is a predetermined number of random samplings,

13. The sampling of the second parameter according to the predetermined condition is a sampling of the second parameter measured in advance for the estimated time with respect to the sampled first parameter. Prediction method.