A state quantity prediction method with a trend judgment mechanism introduced in an industrial control process

Abstract

The application provides a state quantity prediction method in an industrial control process with a trend judgment mechanism, comprising the following steps: step S1: dividing the industrial process state quantity into factor state quantity and target state quantity; step S2: creating a preliminary intelligent prediction model of the target state quantity; step S3: extracting target state quantity change trend feature information and building a target state quantity trend classifier in the industrial control process; step S4: establishing a trend judgment mechanism; and step S5: creating a trend error compensation mechanism and correcting the target state quantity. The application provides a state quantity prediction method in an industrial control process with a trend judgment mechanism, the traditional prediction model usually only pays attention to whether the prediction value is reasonable, but ignores whether the prediction value trend change is correct. The target state quantity change trend has a great influence on the industrial control process strategy. The method introduces the trend judgment mechanism to provide the target state quantity trend change information for the industrial control.

Description

Technical Field

[0001] This invention relates to a method for predicting state variables in industrial control processes by introducing a trend judgment mechanism, and pertains to the field of industrial process control. Background Technology

[0002] With the rapid development of computer science, the storage and processing of massive amounts of historical data generated in industrial production processes has become a reality. This data contains a wealth of useful information, and combined with artificial intelligence algorithms, models of industrial production state variables can be built based on this data. These models can then be used to predict and control key state variables in industrial production, thereby optimizing the industrial production process, improving factory efficiency and quality, reducing production costs, and minimizing maintenance and downtime.

[0003] Artificial intelligence algorithms, including machine learning and deep learning, are constantly being improved and refined, leading to increased efficiency and accuracy in industrial process control. Establishing predictive models for key state variables in industrial processes using historical data and advanced AI algorithms is of paramount importance for upgrading industrial informatization. Summary of the Invention

[0004] This invention provides a method for predicting state variables in industrial control processes by introducing a trend judgment mechanism, solving the control instability problem caused by incorrect trends in the target state variables predicted by traditional artificial intelligence models. First, a preliminary intelligent prediction model is established based on preprocessed and feature-selected industrial process data. Then, the target state variables in the industrial process data are divided into three states: rising, unchanged, and falling. A trend classifier is established, and a trend judgment mechanism is designed. If the trend of the target state variable predicted by the preliminary intelligent prediction model is consistent with the classification result of the trend classifier, the predicted target state variable is used as the output of the industrial learning machine. Otherwise, a trend error compensation mechanism is designed to correct the target state variable using the change information of the industrial process state variables and the trend error guidance. The correction result is used as the output.

[0005] To achieve the above objectives, the technical solution of the present invention is a method for predicting state variables in industrial control processes by introducing a trend judgment mechanism:

[0006] A method for predicting state variables in industrial control processes by incorporating a trend judgment mechanism, characterized in that the method specifically includes the following steps:

[0007] Step S1: Preprocess and perform correlation analysis on the collected industrial process state variables, and divide the state variables into factor state variables and target state variables according to industrial control requirements.

[0008] Step S2: Based on artificial intelligence algorithms, create correlation equations between target state variables and factor state variables to build a preliminary intelligent prediction model for target state variables in industrial control processes.

[0009] Step S3: Extract the target state quantity change trend feature information, and build a target state quantity trend classifier in the industrial control process based on the factor state quantity and trend feature information.

[0010] Step S4: Establish a trend judgment mechanism. If the trend of the target state quantity predicted by the preliminary intelligent prediction model is consistent with the classification result of the trend classifier, the predicted target state quantity is used as the output; otherwise, proceed to the next step.

[0011] Step S5: Create a trend error compensation mechanism. Utilize the change information of industrial process state variables and trend error guidance to correct the target state variable. The correction result is used as the output. A method for predicting state variables in industrial control processes by introducing a trend judgment mechanism.

[0012] As a further improvement of the present invention, step S1 includes the following steps:

[0013] 1-1: Collecting Industrial Process Operation Data D * (X * ,Y * ),in X * Let Y be the factor state quantity, and k be the number of factor state quantities. * Let be the target state variable.

[0014] 1-2: Industrial process operation data D * (X * ,Y * Perform data preprocessing operations such as missing value imputation and outlier handling.

[0015] 1-3: Preprocessed data D * (X * ,Y * Correlation analysis is performed, and feature selection is carried out on the factor state variables based on the correlation analysis results, finally obtaining the data D(X,Y), where X=(X1,X2,…X…). j ), where j is the number of factor state variables after feature selection, j < k.

[0016] As a further improvement of the present invention, step S2 includes the following steps:

[0017] 2-1: Divide the preprocessed and feature-selected data D(X,Y) into training set D. train (X train ,Y train ) and test set Dtest (X test ,Y test ). Where X train and X test Y represents the factor state quantities in the training and test sets, respectively. train and Y test These are the target state variables in the training set and the test set, respectively.

[0018] 2-2: Select a suitable artificial intelligence regression algorithm equation Y = φ(W, B, X). Where W and B are the parameters of the algorithm. Let be the loss function of the algorithm, where loss is the loss value.

[0019] 2-3: Using the factor state variables X of the training set train and target state variable Y train Training intelligent algorithms:

[0020] Building Y train =φ(W,B,X) train )

[0021] The goal is to minimize the loss function. Among them W * and B * These are the model parameters obtained after training.

[0022] The model Y = φ(W) is obtained. * B * ,X).

[0023] 2-4: Using the factor state quantity X of the test set test and target state variable Y test The effectiveness of the obtained model is verified. Finally, the preliminary prediction model for the target state variables, Y = φ(W), is obtained. * B * A method for predicting state variables in industrial control processes by incorporating a trend judgment mechanism.

[0024] As a further improvement of the present invention, step S3 includes the following steps:

[0025] 3-1: The trend of the target state variable curve is divided into three types: rising, unchanged, and falling. These are represented by label1, label2, and label3, respectively.

[0026] 3-2: Based on the changing trend of the target state variables and the factor state variables, the dataset U(X,label) is obtained.

[0027] 3-3: Divide U(X,label) into training set U train (X train ,labeltrain ) and test set U test (X test ,label test ).

[0028] 3-4: Choose a suitable artificial intelligence classification algorithm, label = β(C,V,X), where C and V are the parameters of the algorithm. loss = η(C,V,X) is the loss function of the algorithm, where loss is the loss value.

[0029] 3-5: Using X from the training set train and label train Training intelligent algorithms:

[0030] Building Labe train =β(C,V,X) train )

[0031] The objective is to minimize the loss function: min(loss) = η(C) * V * ,X train ). Among them, C * and V * These are the model parameters obtained after training.

[0032] The model label = β(C) is obtained. * V * ,X).

[0033] 3-6: Using test set X test and label test The effectiveness of the obtained model is verified. Finally, the target state variable trend classifier label = β(C) is obtained. * V * A method for predicting state variables in industrial control processes by incorporating a trend judgment mechanism.

[0034] As a further improvement of the present invention, step S4 includes the following steps:

[0035] 4-1: Preliminary prediction model for existing target state variables Y = φ(W) * B * The target state variable trend classifier label = β(C) * V * ,X).

[0036] 4-2: Suppose there is a factor state variable x at time t. t The preliminary prediction model for the target state variables yields the following result: y t+1 =φ(W * B * ,x t), where y t+1 This represents the predicted value of the target state variable at time t+1.

[0037] 4-3: The label is obtained from the target state variable trend classifier. t+1 =β(C * V * ,x t ), where label t+1 The target state trend is at time t+1.

[0038] 4-4: Based on y t and y t+1 Obtain the target state quantity trend state

[0039] y t+1 -y t >0 definition At this point, the target state is in an ascending state;

[0040] y t+1 -y t =0 definition At this point, the target state variable remains unchanged.

[0041] y t+1 -y t <0 definition At this point, the target state is in a decreasing state;

[0042] 4-5: If The output y of the industrial learning machine est =y t+1 ;if Then proceed to the next step. A method for predicting state variables in industrial control processes by introducing a trend judgment mechanism.

[0043] As a further improvement of the present invention, step S5 includes the following steps:

[0044] 5-1: Set a compensation factor as Δy, and define Δy = y t -y t-1 .

[0045] 5-2: If In an ascending state, label t+1 If the state is unchanged or decreasing, then output y. est =y t+1 +Δy; if In a decreasing state, label t+1 If the state is unchanged or increasing, then output y. est =y t+1 -Δy; if For an invariant state, labelt+1 If it is in an ascending state, then output y. est =y t+1 -Δy; if For an invariant state, label t+1 If it is in a decreasing state, then output y. est =y t+1 +Δy.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] This invention presents a method for predicting state variables in industrial control processes by incorporating a trend judgment mechanism. It utilizes historical industrial process data combined with artificial intelligence algorithms to build a preliminary intelligent prediction model for the target state variables and a classifier for the trend of target state variable changes. The trend classifier is used to determine whether the trend of the target state variables predicted by the preliminary intelligent prediction model is reasonable. If it is unreasonable, a compensation factor is introduced to correct the predicted target state variables.

[0048] Traditional prediction models typically focus only on whether the predicted value is reasonably large, neglecting the accuracy of the predicted value's trend. However, the trend of the target state variable has a significant impact on industrial control process strategies. This method introduces a trend judgment mechanism to provide industrial process control with information on the trend changes of the target state variable. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation

[0050] The present invention will be further explained and described below with reference to embodiments. The specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0051] As shown in the figure, this invention provides a method for predicting state variables in industrial control processes by introducing a trend judgment mechanism, specifically including the following steps:

[0052] Step S1: Preprocess and perform correlation analysis on the collected industrial process state variables, and divide the state variables into factor state variables and target state variables according to industrial control requirements.

[0053] Step S2: Based on artificial intelligence algorithms, create correlation equations between target state variables and factor state variables to build a preliminary intelligent prediction model for target state variables in industrial control processes.

[0054] Step S3: Extract the target state quantity change trend feature information, and build a target state quantity trend classifier in the industrial control process based on the factor state quantity and trend feature information.

[0055] Step S4: Establish a trend judgment mechanism. If the trend of the target state quantity predicted by the preliminary intelligent prediction model is consistent with the classification result of the trend classifier, then the predicted target state quantity is used as the output of the industrial learning machine; otherwise, proceed to the next step.

[0056] Step S5: Create a trend error compensation mechanism, which uses the change information of industrial process state variables and trend error guidance to correct the target state variable, and the correction result is used as the output of the learning machine.

[0057] Step S1 specifically includes the following steps:

[0058] 1-1: Collecting historical operating data of thermal power plants D * (X * ,Y * ),in X * There are 15 factor state variables, Y. * is the target state variable, and is the inlet NOx concentration value.

[0059] 1-2: Missing values ​​are imputed using the upper and lower averaging method. For example, for factor state variables. For missing values, the imputation method is as follows:

[0060]

[0061] Outlier handling is performed using the 3σ principle; data points outside the range (μ-3σ, μ+3σ) are considered outliers, where μ is the mean and σ is the standard deviation. For factor state variables...

[0062]

[0063] 1-3: Use mutual information to perform correlation analysis on the preprocessed data, for factor state quantities. and target state variable Y * The mutual information value is:

[0064]

[0065] in Represents mutual information value, and p(y * ) are the factor state quantities. and target state variable Y * probability distribution for and Y * The joint probability distribution.

[0066] The mutual information values ​​of each factor's state variable and the target state variable are ranked, and the top nine are selected as feature variables based on their mutual information values. These are: total air volume, total coal volume, air-to-coal ratio, air-to-coal ratio of mill A, air-to-coal ratio of mill E, total primary air volume, total secondary air volume, and weighted value of primary air at the mill inlet. The final preprocessed and feature-selected data D(X,Y) is obtained.

[0067] Step S2 specifically includes the following steps:

[0068] 2-1: Divide the preprocessed and feature-selected data D(X,Y) into training set D. train (X train ,Y train ) and test set D test (X test ,Y test ). Where X train and X test Y represents the factor state quantities in the training and test sets, respectively. train and Y test These are the target state variables in the training set and the test set, respectively.

[0069] 2-2: Choosing the Support Vector Machine Regression Algorithm: Y = W T φ(X)+B, where W and B are the parameters of the algorithm. A slack variable ξ is introduced. i and Construct the following loss functions and solve for the optimal solutions for W and B:

[0070]

[0071] stf(X train )-Y train ≤ε+ξ i

[0072]

[0073] Where e is the penalty factor, ε is the allowable deviation, and ξ i and These are slack variables;

[0074] By using optimization theory, the quadratic programming problem is transformed into a dual problem, and the optimal solutions for W and B are obtained, ultimately leading to a preliminary prediction model for NOx concentration state variables.

[0075]

[0076] in, and α i W is a Lagrange multiplier. * and B * These are the model parameters.

[0077] Step S3 specifically includes the following steps:

[0078] 3-1: The trend of NOx concentration state quantity curve is divided into three types: increasing, unchanged, and decreasing. These are represented by label1, label2, and label3, respectively.

[0079] 3-2: The dataset U(X,label) is obtained by combining the trend of NOx concentration state variables with the state variables of factors.

[0080] 3-3: Divide U(X,label) into training set U train (X train ,label train ) and test set U test (X test ,label test ).

[0081] 3-4: Select the Support Vector Machine classification algorithm: label = C T β(X) + V, where C and V are the parameters of the algorithm. A slack variable ξ is introduced. i Construct the following loss functions and solve for the optimal solutions for C and V:

[0082]

[0083] stlabel(C T β(X train )+V≥1-ξ i )

[0084] ξ i ≥0

[0085] i = 1, 2, ..., 9

[0086] Where e is the penalty factor.

[0087] By using optimization theory, the quadratic programming problem is transformed into a dual problem, and the optimal solutions for C and V are found, ultimately yielding a NOx concentration state variable trend classifier.

[0088]

[0089] in, and α i For Lagrange multipliers, C * and V * These are the model parameters.

[0090] Step S4 specifically includes the following steps:

[0091] 4-1: Preliminary Prediction Model for Existing NOx Concentration State Quantities NOx concentration state trend classifier

[0092] 4-2: Suppose that at time t there is a state variable x of NOx factor. t The preliminary prediction model for NOx concentration state variables yielded the following results: Where y t+1 This represents the predicted value of the NOx concentration target state quantity at time t+1.

[0093] 4-3: The NOx concentration state state trend classifier yields the following results: label t+1 The trend of NOx concentration at time t+1.

[0094] 4-4: Based on y t and y t+1 NOx concentration state quantity trend state

[0095] y t+1 -y t >0 definition At this point, the NOx concentration state is in an increasing state;

[0096] y t+1 -y t =0 definition At this point, the NOx concentration state quantity remains unchanged;

[0097] y t+1 -y t <0 definition At this point, the NOx concentration state is decreasing.

[0098] 4-5: If The NOx concentration output by the industrial learning machine is y. est =y t+1 ;if Then proceed to the next step.

[0099] Step S5 specifically includes the following steps:

[0100] 5-1: Set a compensation factor as Δy, and define Δy = y t -y t-1 .

[0101] 5-2: If In an ascending state, label t+1 If the state is unchanged or decreasing, then output y. est =y t+1 +Δy; if In a decreasing state, label t+1If the state is unchanged or increasing, then output y. est =y t+1 -Δy; if For an invariant state, label t+1 If it is in an ascending state, then output y. est =y t+1 -Δy; if For an invariant state, label t+1 If it is in a decreasing state, then output y. est =y t+1 +Δy.

Claims

1. A method for state quantity prediction in an industrial control process incorporating a trend judgment mechanism, characterized by, The method specifically comprises the following steps: Step S1: preprocessing and correlation analysis are performed on the collected industrial process state quantity, and the state quantity is divided into factor state quantity and target state quantity according to the industrial control requirement; Step S2: based on an artificial intelligence algorithm, an association equation of the target state quantity and the factor state quantity is created, so as to build a preliminary intelligent prediction model of the target state quantity in the industrial control process; Step S3: trend characteristic information of the target state quantity is extracted, and a trend classifier of the target state quantity in the industrial control process is built based on the factor state quantity and the trend characteristic information; Step S4: a trend judgment mechanism is established, if the change trend of the target state quantity predicted by the preliminary intelligent prediction model is consistent with the classification result of the trend classifier, the predicted target state quantity is taken as the output, otherwise, the next step is performed; Step S5: a trend error compensation mechanism is created, the target state quantity is corrected by using the change quantity information of the industrial process state quantity and the trend error direction, and the correction result is taken as the output; Step S4 comprises the following steps: 4-1: preliminary prediction model of existing target state quantity Y = φ(W * ,B * ,X) and target state quantity trend classifier label = β(C * ,V * ,X); wherein W * and B * are model parameters obtained after training; X = (X1, X2, … X j ), j is the number of factor state quantities after feature selection; C * and V * are model parameters obtained after training; 4-2: suppose that at time t there is a factor state variable x t , the target state variable is predicted by the preliminary prediction model as follows: y t+1 = φ(W * , B * , x t ), where y t+1 is the predicted value of the target state variable at time t+1; 4-3: The target state quantity trend classifier gives: label t+1 = β(C * , V * , x t ), where label t+1 is the target state quantity trend state at time t+1. 4-4: in accordance with y t and y t+1 get target state quantity trend state y t+1 -y t > 0 definition At this time, the target state quantity is the rising state; y t+1 -y t = 0 defines At this time the target state quantity is an invariant state; y t+1 -y t Definition At this time the target state quantity is the falling state; 4-5: If then the output y of the industrial learning machine est = y t+1 ; if then go to next step; Step S5 comprises the following steps: 5-1: Set a compensation factor as Ay, define Ay=y t -y t-1 ; 5-2： If is an up state, label t+1 is unchanged or a down state, then output y est = y t+1 + Δy; If label is down t+1 output y est = y t+1 - Δy; If is the constant state, label t+1 is the up state, then output y est = y t+1 - Δy; If is the constant state, label t+1 is the falling state, then output y est = y t+1 + Δy.

2. The method of claim 1, wherein the trend determination mechanism is introduced in the industrial control process. Step S1 comprises the following steps: 1-1: Collecting industrial process operation data D * (X * ,Y * ), wherein X * is a factor state quantity, k is the number of factor state quantities; Y * is a target state quantity; 1-2: to industrial process operation data D * (X * , Y * ) performs missing value imputation, outlier treatment operations; 1-3: correlation analysis is performed on the preprocessed data D * (X * , Y * ) and feature selection is performed on the factor state quantities according to the correlation analysis result, and finally data D(X, Y) is obtained, wherein X=(X1, X2, … X j j is the number of factor state quantities after feature selection, and j < k.

3. The method of claim 2, wherein the trend determination mechanism is introduced in the industrial control process state quantity prediction method, characterized in that, Step S2 comprises the following steps: 2-1: The data D(X, Y) after preprocessing and feature selection is divided into a training set D train (X train ,Y train ) and a test set D test (X test ,Y test ); wherein X train and X test are the factor state quantities in the training set and the test set respectively, Y train and Y test are the target state quantities in the training set and the test set respectively; 2-2: Select a suitable artificial intelligence regression algorithm equation Y = φ(W, B, X); wherein W and B are parameters of the algorithm; is the loss function of the algorithm, wherein loss is the loss value; 2-3: Factor state quantities X of the training set train and target state quantities Y train Training the intelligent algorithm: Build Y train = φ(W, B, X train ) with the objective of minimizing a loss function: where W * and B * are the model parameters obtained after training; Y = φ(W * ,B * ,X) 2-4: Factor state quantities X using test set test and target state quantities Y test Verify validity of resulting model; end up with preliminary prediction model for target state quantities Y = φ(W * , B * , X).

4. The method of claim 3, wherein the trend determination mechanism is introduced in the industrial control process. Step S3 comprises the following steps: 3-1: The target state quantity curve change trend is divided into three types: rising, unchanged, and falling; respectively represented by ​ 3-2: data set U(X, label) is obtained according to the change trend of the target state quantity combined with the factor state quantity; 3-3: Divide U(X,label) into training set U train (X train ,label train ) and test set U test (X test ,label test ) 3-4: a suitable artificial intelligence classification algorithm is selected, label = β(C, V, X), wherein C and V are parameters of the algorithm; loss = η(C, V, X) is a loss function of the algorithm, wherein loss is a loss value; 3-5: X using the training set train and lable train Training the intelligent algorithm: Build lab train = β(C, V, X train ) Minimize the loss function: min(loss) = η(C * ,V * ,X train ); where C * and V * are the model parameters obtained after training; Obtain model label = β(C * * X);​ 3-6: Use test set X test and label test Verify the resulting model; end up with a target state quantity trend classifier label = β(C * , V * , X).

Images