A method for detecting steady-state conditions during continuous catalytic reforming reaction

Abstract

The present invention discloses a method for detecting steady-state conditions in a reaction process of a continuous catalytic reforming device. The method automatically selects process parameters that are highly correlated with key quality parameters of the product from the continuous catalytic reforming reaction process, calculates the regression coefficient and weights it, and then uses PLS to calculate the residual space, and finally uses PCA to detect steady-state conditions in the residual space. Compared with the detection method that only uses PCA, this method takes into account the relationship between the steady-state condition and the key quality parameters of the product, avoiding false alarms caused by changes in normal operating conditions; compared with the detection method that uses PLS plus PCA, this method screens and weights the process parameters, removes the influence of irrelevant parameters, and further reduces the false alarm rate. This method can help enterprises accurately identify steady-state processes, and is of great value to stabilize and optimize continuous reforming production and improve the economic benefits of enterprises.

Description

Technical Field

[0001] The present invention relates to the field of working condition monitoring in the production process of a refinery, and specifically to a method for determining whether a continuous catalytic reforming reaction process is in a steady state and providing a timely warning before instability occurs. Background Art

[0002] In the production process of refining and chemical enterprises, the calculation of material balance, heat balance, estimation of model parameters, and real-time optimization are all based on the premise that the production process is in a steady state. Therefore, detecting whether the production process is currently in a steady state is crucial to the efficient production of refining and chemical enterprises.

[0003] However, the current steady-state condition detection method for industrial processes mainly uses the principal component analysis (PCA) algorithm to calculate the SPE and T of process parameters. 2 Statistics do not take into account the impact of process parameter changes on product key quality parameters under the control of the system. Since there is no distinction between operating condition switching or changes in normal operating conditions, when process parameters change but product key quality parameters remain unchanged, the process is mistakenly judged to be in an unstable state. In recent years, researchers have considered the relationship between product key quality parameters and steady state, divided product key quality parameters into parts that can be explained by process parameters and parts that cannot be explained by process parameters, used partial least squares (PLS) to establish a model of process parameters and product key quality parameters, and then calculated the residuals as the part of product key quality parameter changes that cannot be explained by process parameters, and then used PCA to detect and judge the residuals. However, due to the extremely complex reaction of the continuous reforming process and the large number of process parameters and product key quality parameters, the indiscriminate use of all process parameters and product key quality parameters to establish a model introduced a lot of irrelevant data, which reduced the accuracy of the model established by PLS, resulting in certain false positives in steady-state judgment. Therefore, the steady-state condition detection effect of the continuous reforming reaction process needs to be further improved. Summary of the invention

[0004] In view of the above problems, the present invention discloses a steady-state condition detection technology that automatically screens out process parameters with strong correlation with product key quality parameters based on the correlation between data, and considers the relationship between product key quality parameters and steady-state conditions. By calculating the regression coefficient, the process parameters with strong correlation with product key quality parameters are automatically screened out and the process parameters are weighted before establishing a model, so that variables irrelevant to product key quality parameters are removed as much as possible, the accuracy of PLS ​​modeling is improved, and the false alarm rate is reduced when PCA is used to detect the residual space of product key quality parameters later. The method includes the following steps:

[0005] 1) Collect the process parameters X∈R at m stable moments during the continuous catalytic reforming reaction of the refinery m×aAnd product key quality parameters Y∈R m×b , a represents the number of process parameters, b represents the number of key quality parameters of the product;

[0006] 2) Establish a linear regression model Y = XW between process parameters and product key quality parameters, and use the following optimization method to solve it

[0007] The coefficients of the process parameters in the model are W∈R a×b :

[0008]

[0009] Where || || Fro represents the Frobenius norm. For matrix A, || || 21 Indicates L 21 Norm, for matrix A, λ is the penalty coefficient, remember a ij Represents the element located in the i-th row and j-th column of matrix A;

[0010] 3) Filter to meet w k >d, k=1,2,...,c process parameters are the process parameters with strong correlation with the key quality parameters of the product d represents the threshold value, and c represents the number of process parameters with strong correlation;

[0011] 4) Use the following method to process parameters Perform weighted processing to obtain weighted process parameters

[0012]

[0013] 5) Use PLS to extract q principal components to establish And the model of Y:

[0014]

[0015] In the formula Represents the model's estimated value of the key quality parameters of the product, V∈R c×b represents the regression coefficient of the model;

[0016] 6) Calculate the residual space of key quality parameters of the product in the model

[0017] 7) Use PCA to calculate Y * SPE and T 2 Statistics;

[0018] 8) Based on SPE and T 2 Statistics, calculate the BIC statistic:

[0019]

[0020] Where p(T 2 |F) indicates the case of a fault T 2 The probability of the statistic crossing the boundary, p(F|T 2 ) indicates T 2 The probability of a failure when the statistic is out of bounds, p(SPE|F) represents the probability of the SPE statistic being out of bounds when a failure occurs, and p(F|SPE) represents the probability of a failure when the SPE statistic is out of bounds;

[0021] 9) Online monitoring of the continuous catalytic reforming reaction process and calculation of the instantaneous statistics BIC * ;

[0022] 10) Based on instantaneous statistics BIC * Make a steady-state judgment: If the BIC statistic * <1-α, α represents the confidence level, then the system is in a steady state and goes to step (9); otherwise, an early warning is issued and goes to step (9).

[0023] Beneficial effects:

[0024] The present invention discloses a method for detecting steady-state conditions in a continuous catalytic reforming reaction process. Compared with the detection method using only PCA, this method takes into account the relationship between steady-state conditions and key product quality parameters, avoiding false alarms caused by changes in normal operating conditions; compared with the detection method using PLS plus PCA, this method screens and weights process parameters, removes the influence of irrelevant parameters, and further reduces the false alarm rate. This method can detect steady-state conditions in real time and accurately, thereby helping enterprises to promptly discover instability, which is of great value to enterprises' safe and efficient production and improved economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for detecting steady-state conditions during a continuous catalytic reforming reaction process of the present invention;

[0026] Figure 2 A coefficient thermodynamic diagram of a method for detecting steady-state conditions of a continuous catalytic reforming reaction process according to the present invention;

[0027] Figure 3 The test results of the method of the present invention are:

[0028] Figure 4 The test results of PCA fault detection method. DETAILED DESCRIPTION

[0029] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, and the implementation effect of the method in the detection of steady-state conditions in the continuous catalytic reforming reaction process is described by a specific operation flow. This embodiment is implemented on the premise of the technical solution of the present invention, but the protection scope of the present invention is not limited to the following embodiments.

[0030] The present invention takes the continuous catalytic reforming unit of a certain oil refinery as an example. The detection data of the refinery includes a=9 process parameters, namely, the reforming feed amount, the reforming circulating hydrogen flow rate, the reforming gas-liquid separator pressure, the inlet temperatures of 4 reactors, the proportion of raw material cycloalkanes and the proportion of raw material aromatics. A total of 7 key product quality parameters are selected, namely b=7, including the total product amount, the outlet temperatures of 4 reactors, the proportion of product cycloalkanes and the proportion of product aromatics. Taking the historical data of the enterprise from May 11, 2020 to May 21, 2020, a total of 2450 samples, as an example, the role of this method in the detection of steady-state conditions in the continuous catalytic reforming reaction process is introduced.

[0031] The implementation process of this method is as follows Figure 1 As shown, the implementation steps are as follows:

[0032] 1) Collect the process parameters X∈R of m=2000 stable moments in the continuous catalytic reforming reaction process of the refinery m×a And product key quality parameters Y∈R m×b , where a=9, b=7;

[0033] 2) Establish a linear regression model Y = XW between process parameters and product key quality parameters, and use the following optimization method to solve the coefficient W∈R of the process parameters in the model: a×b :

[0034]

[0035] The absolute value of the W matrix is ​​represented by a heat map to indicate the correlation. The darker the color, the larger the coefficient. White represents 0. Figure 2 As shown, in this example, λ=0.18;

[0036] 3) According to Figure 2 Correlation using w k >d is used to screen the process variables. In this example, d = 0. The reforming gas-liquid separator pressure (3), the first reactor inlet temperature (4), the third reactor inlet temperature (6), the fourth reactor inlet temperature (7), the raw material cycloalkane ratio (8), and the raw material aromatic ratio (9) are selected as process parameters with strong correlation with the key quality parameters of the product, that is, c = 6;

[0037] 4) Use the model regression coefficient to calculate the weighted coefficient of the process parameter and weight the screened process parameters:

[0038]

[0039] 5) Use PLS to extract the principal components to establish and Y model, when the principal component is 4, The cumulative contribution rate has reached 0.95, so q = 4 principal components are selected to establish Model with Y

[0040] 6) Calculate the residual space of key quality parameters of the product in the model

[0041] 7) Use PCA to calculate Y * SPE and T 2 Statistics;

[0042] 8) Based on SPE and T 2 Statistics, calculate the BIC statistic:

[0043]

[0044] 9) Use the above method to test the last 450 samples and calculate the instantaneous statistic BIC * , the results are as follows Figure 3 As shown;

[0045] 10) Take confidence level α = 0.99, Figure 3 The dotted horizontal line in the middle is 1-α=0.01. When the instantaneous statistic BIC * If this line is exceeded, it will be unstable.

[0046] For comparison, only PCA is used for detection. Figure 4 As shown. Figure 3 and Figure 4 It can be seen that in the test data set, the method used in this patent significantly reduces the false alarm rate compared to PCA, and the method used in this patent improves the accuracy of instability warning compared to PCA.

Claims

1. A method for detecting steady-state conditions in a continuous catalytic reforming reaction process, characterized in that: Automatically screen the process parameters related to the key quality parameters of the product, and weight the process parameters. Then use partial least squares regression PLS to establish a model between the process parameters and the key quality parameters of the product, obtain the residual space of the key quality parameters of the product in the model, and then use principal component analysis PCA to perform steady-state condition detection in the residual space, including the following steps: 1) Collect the process parameters X∈R at m stable moments during the continuous catalytic reforming reaction of the refinery m×a And product key quality parameters Y∈R m×b , a represents the number of process parameters, b represents the number of key quality parameters of the product; 2) Establish a linear regression model between process parameters and product key quality parameters The coefficient W∈R of the process parameter in the model is obtained by the following optimization method: a×b : In the formula|||| Fro represents the Frobenius norm. For matrix A, |||| 21 Indicates L 21 Norm, for matrix A, λ is the penalty coefficient; remember a ij represents the element located in the i-th row and j-th column of matrix A; 3) Screening process parameters that are highly correlated with product key quality parameters c represents the number of process parameters with strong correlation; 4) Weighted processing is performed on the process parameters with strong correlation to obtain the weighted process parameters 5) Use PLS to extract q principal components to establish and Model: In the formula Represents the model's estimated value of the key quality parameters of the product, V∈R c×b is the regression coefficient of the model; 6) Calculate the residual space Y of the key quality parameters of the product in the model * ; 7) Use PCA to calculate Y * SPE and T 2 Statistics; 8) Based on SPE and T 2 , calculate the BIC statistic; 9) Online monitoring of the continuous catalytic reforming reaction process and calculation of the instantaneous statistics BIC * ; The BIC statistic is calculated using the following method: Where p(T 2 |F) indicates the case of a fault T 2 The probability of the statistic crossing the boundary, p(F|T 2 ) indicates T 2 The probability of a failure when the statistic is out of bounds, p(SPE|F) represents the probability of the SPE statistic being out of bounds when a failure occurs, and p(F|SPE) represents the probability of a failure when the SPE statistic is out of bounds; 10) Based on instantaneous statistics BIC * Make steady-state judgment: If the instantaneous statistic BIC * <1-α, α represents the confidence level, the system is in a steady state and go to step 9), otherwise, issue an early warning and go to step 9).

2. A method for detecting steady-state conditions of a continuous catalytic reforming reaction process according to claim 1, characterized in that: The screening process of process parameters with strong correlation with product key quality parameters is as follows: t >d, t = 1, 2, ..., c are process parameters with strong correlation with product key quality parameters, and d is the threshold value.

3. The method for detecting steady-state conditions of a continuous catalytic reforming reaction process according to claim 1, characterized in that: Use the following method to To perform weighting:

4. The method for detecting steady-state conditions of a continuous catalytic reforming reaction process according to claim 1, characterized in that: use Calculate the residual space of the key quality parameters of the product in the model.

Images