A continuous catalytic reforming reaction process working condition division method

Abstract

The application discloses a continuous catalytic reforming reaction process working condition division method, which selects time series data of raw material components, adopts empirical mode decomposition for denoising, then normalizes the time series data into Gaussian distribution, divides the time series data into equal probability intervals according to the ordinate, and respectively performs symbolic representation to measure the similarity between different segments, and finally uses hierarchical clustering to divide the symbolic time series data into different working conditions. The method can judge the change of the working condition according to the change of the raw material properties, and establish a corresponding model for different working conditions, which has important value for improving the real-time optimization precision of the subsequent continuous catalytic reforming reaction process, and promoting efficient production and improving economic benefits of enterprises.

Description

Technical Field

[0001] This invention relates to the field of operating condition monitoring in the production process of refining and chemical enterprises, specifically a method for classifying operating conditions in a continuous catalytic reforming reaction process to improve modeling accuracy. Background Technology

[0002] In the continuous catalytic reforming process of refining and chemical enterprises, the reaction is affected by many factors such as reactor inlet temperature, reaction pressure, space velocity and feed properties. The combined effect of multiple factors can lead to different reaction modes, which can have an adverse impact on modeling accuracy and subsequent real-time optimization applications.

[0003] On the one hand, current modeling of continuous reforming processes still treats the entire reaction process as a single operating condition, failing to consider the impact of various factors on the reaction modes, leading to deviations in model accuracy under different reaction conditions. On the other hand, although some studies have considered the impact of different operating conditions on modeling accuracy and divided the reaction process into different operating conditions for separate modeling, these studies mainly consider the impact of reactor load changes on the reaction, dividing the reaction into different operating conditions based on different loads. However, in actual production, although load changes have a significant impact on the entire production process, they are not frequently adjusted; production is generally carried out continuously for a relatively long period under the same load. Currently, there are few studies that investigate the impact of multiple factors such as feedstock composition under the same production load and use this to divide production operating conditions. Dividing the continuous reforming reaction process into operating conditions is expected to further improve modeling accuracy and more accurately serve the real-time optimization of catalytic reforming. Summary of the Invention

[0004] To address the aforementioned problems, this invention discloses a method for classifying operating conditions in a continuous catalytic reforming reaction process. Under the same production load, key raw material components are screened based on the production process. Noise is removed through Empirical Mode Decomposition (EMD). The data is then normalized to a Gaussian distribution and equally divided to reduce dimensionality. Based on the segmentation points, the data is divided into equally probable intervals, and character representations are used to measure the similarity between data points. Finally, hierarchical clustering is used to classify the operating conditions. This method includes the following steps:

[0005] First, time series data of raw material components were selected, and empirical mode decomposition was used for noise reduction, including:

[0006] 1) Collect feedstock composition data Y∈R at n time points during the continuous catalytic reforming process in a refining and chemical enterprise. n×o R represents a real number, and o represents the number of raw material components;

[0007] 2) Based on the process, select b key raw material components X from Y:

[0008] X∈Rn×b , X∈Y, b≤o

[0009] 3) Empirical mode decomposition is used to decompose X into the sum of e eigenmode functions and residual signals, where the raw material component X at time t is... t Represented as:

[0010]

[0011] In the formula imf i Let r represent the i-th intrinsic mode function, and r represent the residual signal;

[0012] 4) For i < k, use imf i Treating it as noise, signal reconstruction is performed to obtain the denoised key raw material components. Where k represents the threshold, and the key raw material component at time t. Represented as:

[0013]

[0014] Then, the time series data is normalized to a Gaussian distribution, divided into equal probability intervals according to the ordinate, and symbolically represented to measure the similarity between different segments, including:

[0015] 5) Normalization yields the Gaussian distribution of the key raw material component X', where the normalized key raw material component X' at time t is... t Represented as:

[0016]

[0017] In the formula, μ and δ represent respectively The mean and standard deviation;

[0018] 6) Divide X' into w equal segments, and represent each segment using the mean to obtain the dimensionality-reduced raw material component C, where the p-th segment C... p Represented as:

[0019]

[0020] 7) Divide C into α intervals and symbolize them, where C is in the q-th interval. p Symbolization Represented as:

[0021]

[0022] In the formula, alpha represents the character set, and β is the delimiter.

[0023] Finally, hierarchical clustering was used to divide the symbolic time series data into different operating conditions, specifically:

[0024] 8) Calculate the minimum distance MINDIST(Q,C) between two time series Q and C using the following formula, assuming that time series Q and C are symbolized as follows: and but:

[0025]

[0026] The formula for calculating the dist function is:

[0027]

[0028] In the formula, r and c represent any two characters in the character set. Hierarchical clustering is used to cluster the obtained character set, dividing the data into l types of working conditions.

[0029] Beneficial effects:

[0030] This invention discloses a method for classifying operating conditions in a continuous catalytic reforming reaction process. This method considers the influence of feedstock properties on the reaction process and avoids the adverse effects of changes in operating conditions caused by different feedstocks on model accuracy. This method can determine changes in operating conditions based on variations in feedstock properties and establish corresponding models for different operating conditions. This is of great value for improving the real-time optimization accuracy of subsequent continuous catalytic reforming reaction processes, thereby promoting efficient production and improving economic benefits for enterprises. Attached Figure Description

[0031] Figure 1 This is a flowchart of a method for classifying operating conditions in a continuous catalytic reforming reaction process according to the present invention;

[0032] Figure 2 This is an EMD decomposition diagram of the raw material components in an embodiment of the present invention;

[0033] Figure 3 The percentage of cycloalkanes and aromatics after the EMD reconstruction of this invention;

[0034] Figure 4 This is the data symbolization result of an embodiment of the present invention;

[0035] Figure 5 This is the result of the working condition division in this invention;

[0036] Figure 6 The modeling and prediction results for different working conditions of this invention are shown. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, illustrating the effectiveness of this method in classifying operating conditions during continuous catalytic reforming reactions through specific operational procedures. This embodiment is implemented based on the technical solution of the present invention, but the scope of protection of the present invention is not limited to the following embodiments.

[0038] This invention takes a continuous catalytic reforming unit in an oil refinery as an example. The feedstock components of this unit include alkanes, cycloalkanes, and aromatics, i.e., o=3. Since the continuous reforming process first converts alkanes to cycloalkanes, and then cycloalkanes to aromatics, cycloalkanes and aromatics have the greatest impact on the product according to process analysis. Therefore, data on cycloalkanes and aromatics are selected for analysis, i.e., b=2. Samples from June 8th to June 16th, 2020, totaling n=2303 time points, are selected for analysis to illustrate the role of this method in the division of operating conditions in the continuous catalytic reforming reaction process.

[0039] The implementation process of this method is as follows: Figure 1 As shown, the implementation steps are as follows:

[0040] 1) Collect the feedstock component Y at n=2303 time points during the continuous catalytic reforming process of the refining and chemical enterprise;

[0041] 2) Based on the process, select b key raw material components X from Y:

[0042] X∈R n×b , X∈Y, b≤o

[0043] 3) Empirical mode decomposition is used to decompose X into the sum of e eigenmode functions and residual signals, where the raw material component X at time t is... t Represented as:

[0044]

[0045] In the formula imf i Let represent the i-th intrinsic mode function, and r represent the residual signal. The time series EMD decomposition and EMD decomposition frequency of cycloalkanes and aromatics are as follows: Figure 2 As shown;

[0046] 4) According to Figure 2 The frequency of the empirical mode function is defined for i ≤ 3. i Treating it as noise, the signal is reconstructed to obtain the denoised key raw material components. The key raw material components at time t Represented as:

[0047]

[0048] Figure 3 Comparison images of noise reduction before and after are provided from June 11th to June 13th;

[0049] 5) Normalize X to obtain the Gaussian distribution of the key raw material component X', where the normalized key raw material component X' at time t is... t Represented as:

[0050]

[0051] In the formula, μ and δ represent respectively The mean and standard deviation;

[0052] 6) In this embodiment, X' is divided into w = 96 segments, and each segment is represented by its mean to obtain the dimension-reduced raw material component C, where the p-th segment C p Represented as:

[0053]

[0054] 7) Divide C into α = 8 equally probable intervals and symbolize them, where C in the q-th interval... p Symbolization For ease of calculation, the dividing points are obtained by looking up Table 1. Each column in Table 1 indicates the number of intervals. In this embodiment, the column with 8 intervals is selected, resulting in a total of 7 dividing points:

[0055]

[0056] In the formula, alpha = [a,b,c,d,e,f,g,h], and the segmentation result is as follows: Figure 4 As shown:

[0057] Table 1. Equal Probability Partition Points

[0058]

[0059] 8) Use the following formula to calculate the minimum distance MINDIST(Q,C) between two time series Q and C to determine the similarity between the two time series. Assume that time series Q and C are symbolized as follows: and but:

[0060]

[0061] The formula for calculating the dist function is:

[0062]

[0063] In the formula, r and c represent any two characters in the character set. Hierarchical clustering is used to cluster the obtained character set, dividing the data into l=2 categories. The classification results are as follows: Figure 5 As shown.

[0064] To demonstrate the effectiveness of the above-mentioned work condition classification, two models were developed for each of the two work conditions, and these two models were used to predict the two work conditions respectively. 300 data points were selected for each work condition, and the predictions were performed using the two models. The sum of squared errors between the model outputs and the normalized true values ​​was used as the prediction result. Figure 6 As shown in Table 2, the average prediction error is as follows:

[0065] Table 2 Modeling and Prediction Errors under Different Working Conditions

[0066] Model 1 <![CDATA[3.67×10 3 ]]> <![CDATA[3.80×10 3 ]]> Model 2 <![CDATA[5.37×10 3 ]]> <![CDATA[3.46×10 3 ]]>

[0067] pass Figure 6 Comparing with Table 2, it can be seen that the model for operating condition 1 has a smaller prediction deviation for operating condition 1; similarly, the model for operating condition 2 has a smaller prediction deviation for operating condition 2, meaning that each operating condition model describes the operating condition more accurately. Therefore, the method of this invention can improve modeling accuracy, thus laying the foundation for subsequent real-time optimization of continuous catalytic reforming units.

Claims

1. A method for dividing the operating conditions of a continuous catalytic reforming reaction process, characterized in that... The raw material component time series data were selected, and empirical mode decomposition was used for noise reduction. Then, the time series data was normalized to a Gaussian distribution, divided into equal probability intervals according to the ordinate, and symbolically represented to measure the similarity between different segments. Finally, hierarchical clustering was used to divide the symbolically represented time series data into different working conditions, including the following steps: 1) Collect data from the continuous catalytic reforming process in refining and chemical enterprises. Raw material composition data at each moment , Represent real numbers, Indicates the number of raw material components; 2) According to the process from Selected from Key raw material components : 3) Employing empirical mode decomposition to... Decomposed into The sum of the intrinsic mode functions and residual signals, where raw material components at time Represented as: In the formula express One eigenmode function Represents the residual signal; 4) The following Treating it as noise, the signal is reconstructed to obtain the denoised key raw material components. ,in Indicates the threshold. Key raw material components at the moment Represented as: 5) Normalization yields the key raw material components with a Gaussian distribution. ,in Key raw material components after time normalization Represented as: In the formula and They represent The mean and standard deviation; 6) Divided into equal parts Each segment is represented by its mean, resulting in the dimensionality-reduced time series of raw material components. , of which part Represented as: 7) Divided into The intervals are symbolized, where the first interval is... In each interval Symbolization Represented as: In the formula Represents a character set. The dividing point; 8) Use hierarchical clustering to cluster the obtained character set, dividing the data into... Type of working condition.

2. The method for dividing the operating conditions of a continuous catalytic reforming reaction process according to claim 1, characterized in that... Step (8) Calculate the two time series using the following formula. and minimum distance between : in The formula for calculating the function is: In the formula This represents any two characters in the character set. Hierarchical clustering is used to cluster the resulting character set, dividing the data into groups. Type of working condition.

Images