A hydrocracking production process data preprocessing method and system

Abstract

The present invention belongs to, but is not limited to, the field of data preprocessing technology. A method and system for preprocessing data in a hydrocracking production process are disclosed. The method collects production data with missing data in the hydrocracking production process. A missing data completion method using a masked denoising autoencoder with multiple interpolation based on a self-attention mechanism is used to complete key missing data in the production data. The missing data completion method using a masked denoising autoencoder with multiple interpolation based on a self-attention mechanism accurately fills in some missing data using a masked autoencoder based on a self-attention mechanism, and then completes the data as a whole using multiple interpolation. The missing data completion technology using a masked autoencoder with a self-attention mechanism filters similar data using a self-attention mechanism, and then completes missing data with similar data using a masked autoencoder. The method of the present invention improves data completion accuracy while reducing computational difficulty.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the technical field of data preprocessing, and in particular relates to a method and system for preprocessing data in a hydrocracking production process. Background Art

[0002] In industrial production, the reliability and accuracy of data preprocessing are crucial to the efficient, profitable, and safe operation of the plant. As a continuous and complex industrial production process, the hydrocracking process is affected by fluctuations in the taste and flow of inlet raw materials, changes in product specification requirements, and degradation of equipment operating performance. The production process usually includes multiple operating conditions. Under complex operating conditions, during the collection of industrial data, hardware or network-related problems may lead to missing collected data. The lack of input data has a significant impact on the subsequent hydrocracking data modeling and optimization adjustments. At this time, it is necessary to develop management technology to complete the missing data collected and make it suitable for the complex operating conditions in the hydrocracking process.

[0003] Domestic and international scholars have conducted extensive research on data preprocessing for industrial data collection. Existing missing data imputation techniques include the historical average (HA) method and the linear interpolation (LI) method. The HA method imputes missing values ​​by calculating the average of highly correlated time slots. This method is effective for aggregated energy data, but its performance is poor under high volatility patterns. The linear interpolation (LI) method estimates missing values ​​in the input data by connecting the two measurements before and after the missing data. This method works well for short-term missing values, but its effectiveness decreases significantly when the missing value period is prolonged. Furthermore, domestic and international academics have designed a variety of methods to improve the accuracy of data preprocessing in industrial production processes, such as correlation coefficients, mutual information, Euclidean distance, and Bayesian inference, among other common data preprocessing methods. However, these methods are generally applicable to simple processes rather than complex industrial production processes.

[0004] To address the shortcomings of HA and LI, researchers have proposed methods such as self-attention, masked autoencoders (MAEs), and multiple imputation (MI) to complete missing data. Self-attention calculates the weight between each element in the sequence data and all other elements. This weight reflects the relationship between elements, thereby displaying similar, related data and facilitating subsequent processing of missing data. However, data completion using the self-attention mechanism can lead to data uniformity and poor data fitting. The basic idea of ​​the masked autoencoder (MAE) is to train the original missing data values ​​to be recovered from the remaining similar input data. The MAE is trained to complete the missing data by learning the similar structure of the remaining collected data. However, the MAE suffers from the problem of partial data omission during data completion. Using multiple imputation (MI) to complete data can fill all missing values, but it suffers from low completion accuracy and high computational complexity.

[0005] Therefore, a missing data completion technology that combines the advantages of self-attention mechanism (Self-Attention), masked autoencoder (MAE) and multiple interpolation (MI) is needed to improve the processing of missing data in restrictive settings. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a data preprocessing method and system for a hydrocracking production process. The method first screens similar data in the input data and then learns the similar data to accurately complete some missing data, thereby reducing the computational difficulty of subsequent overall data completion and improving the accuracy of missing data completion.

[0007] The present invention is achieved by a method for preprocessing data in a hydrocracking production process, comprising the following steps:

[0008] Step 1: collecting production data with missing data in the hydrocracking production process;

[0009] Step 2: Use the missing data completion method of masked denoising autoencoder multiple interpolation based on self-attention mechanism (MI-SAMAE) to complete the key missing data in the production data;

[0010] The missing data completion method of the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) accurately fills in some missing data through the masked autoencoder based on the self-attention mechanism, and then improves the whole through the multiple interpolation method; the missing value estimated by the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) can be expressed as:

[0011]

[0012] Among them, Z in Indicates an incomplete sample set;

[0013] The missing data completion technology of the masked autoencoder based on the self-attention mechanism screens similar data through the self-attention mechanism, and then completes the missing data with similar data through the masked autoencoder; the minimization total objective function of the missing data completion technology of the masked autoencoder based on the self-attention mechanism (SAMAE) can be expressed as:

[0014]

[0015] Furthermore, the production data includes production data with missing data output by each controller in the hydrocracking production process.

[0016] Furthermore, the self-attention mechanism calculates the weights between elements in the sequence data, understands the contextual information in the sequence, and marks data with similar change trends;

[0017] The self-attention mechanism calculates the attention weight of the elements in the sequence through the softmax function:

[0018] w ij =softmax(score(Q i ,K j ))

[0019] Among them, score(Q i ,K j ) represents the attention score, Q i represents the query vector, K i Represents a key vector.

[0020] Furthermore, the masked autoencoder (MAE) recovers the missing values ​​of the original data from the input data with similar change trends by additionally setting a mask on the original data;

[0021] In a high missing rate environment, the input vector of masked autoencoders (MAEs) can be a partial observation vector. By learning from similar input data, the missing values ​​of the input data can be completed.

[0022] Furthermore, during the MAE learning process, the masked autoencoder (MAEs) defines a new error term And only consider the contribution of the observations, as follows:

[0023]

[0024] Among them, i represents the set of observations on day i, 1 represents the indicator function, when x ij If ≠0, it is equal to 1, otherwise it is 0.

[0025] Furthermore, the multiple imputation (MI) method uses a single imputation method of mean imputation, linear regression imputation, deductive imputation, nearest distance imputation, hot card imputation and cold card imputation to repeatedly carry out several rounds of imputation to obtain multiple different complete data sets, and imputes the missing data that does not exist for similar data that cannot be filled by the mask denoising autoencoder in a high missing rate environment, and finally integrates them to obtain complete data.

[0026] Another object of the present invention is to provide a hydrocracking production process data preprocessing method and a hydrocracking production process data preprocessing system, comprising:

[0027] Production data acquisition module, which collects production data with missing data in the hydrocracking production process;

[0028] The missing data completion module uses the missing data completion method of multiple interpolation with masked denoising autoencoders based on self-attention mechanism (MI-SAMAE) to complete the key missing data in the production data;

[0029] The missing data completion method of the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) accurately fills in some missing data through the masked autoencoder based on the self-attention mechanism, and then improves the whole through the multiple interpolation method; the missing value estimated by the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) can be expressed as:

[0030]

[0031] Among them, Z in Indicates an incomplete sample set;

[0032] The missing data completion technology of the masked autoencoder based on the self-attention mechanism screens similar data through the self-attention mechanism, and then completes the missing data with similar data through the masked autoencoder; the minimization total objective function of the missing data completion technology of the masked autoencoder based on the self-attention mechanism (SAMAE) can be expressed as:

[0033]

[0034] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the hydrocracking production process data preprocessing method.

[0035] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the hydrocracking production process data preprocessing method.

[0036] Another object of the present invention is to provide an information data processing terminal, which includes the hydrocracking production process data preprocessing system.

[0037] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0038] First, the present invention collects a set of industrial production data with missing values ​​output by each controller in the hydrocracking production process; Generate additional missing values ​​for the industrial production data, with the original data as output, Through the self-attention mechanism, the correlation between the elements of the original sequence is calculated, the relationship between the data context is understood, and the model is helped to more accurately handle missing data in the sequence. The masked autoencoder (MAE) method is then used to train the original missing data values ​​to recover them from the remaining similar input data. The two methods are combined to form a masked denoising autoencoder (SAMAE) based on the self-attention mechanism for preliminary filling of missing data in industrial processes. The multiple interpolation method (MI) uses a single filling method to further fill in the missing values ​​omitted in the preliminary filling, obtaining more complete industrial temperature input data. The masked autoencoder multiple interpolation method based on the self-attention mechanism (MI-SAMAE) is applied to data preprocessing of the hydrocracking industrial process, ensuring that data filling can accurately and completely fill in missing data in the case of high missing data rates. Compared with existing technologies, the MI-SAMAE method combines the advantages of the self-attention mechanism, MAE, and multiple interpolation methods, and can improve the handling of missing data even in restrictive settings. It is applicable to hydrocracking process flows that are more complex than simple processes.

[0039] Second, the technical solution of the present invention contributes to improving the production efficiency of the hydrocracking industry after transformation. The technical solution of the present invention performs data preprocessing on the industrial production data collected by the sensors of the hydrocracking industrial production process, and accurately completes the missing industrial production data, which is conducive to the smooth advancement of the subsequent data modeling optimization process, improves the accurate analysis of industrial data, strengthens the online control of the hydrocracking industrial process, is conducive to subsequent industrial manual intervention, improves the hydrocracking industrial production standards and is conducive to the greening of the industrial production process.

[0040] Furthermore, the technical solution of the present invention utilizes the self-attention mechanism (SA) to improve the data completion capability of the traditional masked autoencoder (MAE) data completion method, while also using multiple interpolation (MI) to complete the production data that cannot be completed. The technical solution of the present invention can further improve and enhance the data completion process based on traditional methods, and can also achieve more complete and accurate completion of missing data in production data, providing convenience for subsequent industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a method for preprocessing data in a hydrocracking production process provided by an embodiment of the present invention;

[0042] Figure 2 This is a structural diagram of a data preprocessing system for a hydrocracking production process provided by an embodiment of the present invention;

[0043] Figure 3 This is a diagram showing the MI-SAMAE simulation training effect provided by an embodiment of the present invention;

[0044] Figure 4 This is a diagram showing the results of a partial data simulation training of MI-SAMAE provided by an embodiment of the present invention;

[0045] Figure 5 This is a diagram showing the effect of the MAE simulation training results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] like Figure 1 As shown, an embodiment of the present invention provides a method for preprocessing data of a hydrocracking production process, which is applied to complete missing data of the hydrocracking production process and solve the problem of accuracy of data collected in the hydrocracking production process. Specifically, the method includes:

[0048] S101, collecting production data with missing data in a hydrocracking production process, wherein the production data includes production data with missing data output by each controller in the hydrocracking production process;

[0049] S102, using the self-attention mechanism to calculate the weights between elements in the sequence data, understand the contextual information in the sequence, and mark data with similar change trends to facilitate more accurate processing of the sequence data;

[0050] S103, since the collected data has missing data values, a mask is additionally set on the original data through a masked autoencoder (MAE) to recover the missing values ​​of the original data from the output data with similar change trends;

[0051] S104, by combining the two processes in S102 and S103 to form a missing data completion technology based on the self-attention mechanism masked autoencoder (SAMAE), the missing data of key variables in the hydrocracking process are completed, and the key missing data in the production data are completed;

[0052] S105, because the mask denoising autoencoder cannot fill in the missing data when there is no similar data in the high missing rate environment, the multiple interpolation (MI) method is used to repeatedly expand the single filling method for several rounds to obtain multiple different complete data sets, and finally integrate them to obtain the complete data;

[0053] S106, by combining the two methods in S104 and S105 to form a missing data completion method based on multiple interpolation of masked denoising autoencoders (MI-SAMAE), which improves the accuracy of data completion while reducing the computational difficulty.

[0054] Specifically, S101 includes:

[0055] In the embodiment of the present invention, industrial production data with missing data outputted from the hydrocracking production process is randomly collected at a sampling interval of 1 minute. A set of industrial production data consists of production data with missing data outputted from various controllers in the hydrocracking production process.

[0056] Specifically, S102 includes:

[0057] The self-attention mechanism calculates the weight between each element and all other elements in the sequence data. The purpose is to reflect the relationship between elements through the weight, thereby displaying related data with similarities, which facilitates the subsequent processing of missing data.

[0058] The self-attention mechanism calculates the attention weight of the elements in the sequence through the softmax function:

[0059] w ij =softmax(score(Q i ,K j ))

[0060] Among them, score(Q i ,K j ) represents the attention score, and the specific calculation process is as follows:

[0061]

[0062] in, represents the scaling factor, Q i represents the query vector, K i Represents the key vector. The query vector Q is obtained by learning the weight matrix and linear transformation of the input elements in the self-attention mechanism. i , key vector K i and the value vector V i Specifically, it can be expressed as:

[0063] Q i =W Q ·x i

[0064] K i =W K ·x i

[0065] V i =W V ·x i

[0066] Among them, W Q , W K and W V There are three trainable parameter matrices, which are represented by the value vector V j Multiplying and summing the corresponding attention weights gives the output:

[0067]

[0068] Specifically, S103 includes:

[0069] The masked autoencoder (MAE) can fill in the missing values ​​of the data by learning the rest of the data similar groups; however, when the missing rate is high, the masked autoencoder is difficult to learn similar data and thus difficult to completely fill in the missing values ​​of the data.

[0070] In a high-missing rate environment, the input vector of the masked autoencoder (MAEs) can be a partial observation vector. By learning similar input data, the missing values ​​of the input data are filled. In the MAE learning process, the missing values ​​replaced by zero values ​​do not participate in the reconstruction process. In this process, a new error term is defined. Only the contribution of the observed value is considered. As follows:

[0071]

[0072] Among them, x i represents the input vector, x ij represents the jth element of the input vector, i represents the set of observations on day i, g(f(x i )) represents the reconstructed output vector, g(f(x i )) j Represents the jth element of the reconstructed output vector, 1 represents the indicator function, when x ij When ≠0, it is equivalent to 1, otherwise it is 0. This type of loss is called masked reconstruction loss because missing values ​​are masked during the reconstruction phase. Simply restoring the observed values ​​still results in missing data, and MAE allows for finding similar data during the reconstruction process to learn and fill in the missing data.

[0073] Specifically, S104 includes:

[0074] Combining the advantages of the self-attention mechanism (Self-Attention) and the masked autoencoder (MAE), a masked autoencoder missing data completion technology (SAMAE) based on the self-attention mechanism is formed. The self-attention mechanism is used to filter similar data, and then the masked autoencoder is used to complete the missing data with similar data.

[0075] The minimization of the total objective function of the masked autoencoder missing data completion technology (SAMAE) based on the self-attention mechanism can be expressed as:

[0076]

[0077] Among them, z i represents the input vector, z ij represents the jth element of the input vector, i represents the set of observations on day i, g(f(z i )) represents the reconstructed output vector, g(f(z i )) j Represents the jth element of the reconstructed output vector, 1 represents the indicator function, when z ij ≠0 is equivalent to 1 otherwise 0. This type of loss is called masked reconstruction loss because missing values ​​will be masked during the reconstruction phase.

[0078] Specifically, S105 includes:

[0079] Through the multiple imputation method (MI), several rounds of imputation are repeated using single imputation methods such as mean imputation, linear regression imputation, deductive imputation, nearest distance imputation, hot card imputation and cold card imputation to obtain multiple different complete data sets. The missing data that cannot be filled by the mask denoising autoencoder in a high missing rate environment and do not have similar data are filled, and finally the complete data are obtained by integration.

[0080] For an incomplete data set, a single imputation method is used to perform m imputation operations to obtain m groups of imputation values, generating m complete data sets. The same analysis algorithm is then used to independently analyze all complete data sets, and m analysis results are obtained. A classification model is built based on each data set, and the model parameters are obtained. Finally, all model parameters are merged into a set of parameters through the idea of ​​​​demeaning.

[0081] Taking the linear regression imputation method as an example, first, a linear regression model is established:

[0082]

[0083] Among them, X co represents the complete sample set, x ij represents the sample in the complete sample set, y i represents the dependent variable, β j represents the model parameters, β0 represents the constant term, also known as the intercept, and ε j Represents the error. The linear model used to estimate the missing value can be expressed as:

[0084]

[0085] in, represents the estimated model parameters based on the complete sample, represents the least squares estimate of β0, X in Indicates an incomplete sample set.

[0086] Specifically, S106 includes:

[0087] Combining the advantages of the masked autoencoder (SAMAE) based on the self-attention mechanism and the multiple interpolation method (MI), a missing data completion method based on the masked denoising autoencoder multiple interpolation (MI-SAMAE) of the self-attention mechanism is formed. The missing data is accurately filled by the masked autoencoder based on the self-attention mechanism, and then the overall improvement is carried out through the multiple interpolation method, which improves the accuracy of data completion while reducing the computational difficulty. It is used to solve the accuracy problem of data collected in the hydrocracking production process.

[0088] Estimating missing values ​​through multiple interpolation of masked denoising autoencoders based on self-attention mechanism (MI-SAMAE) can be expressed as:

[0089]

[0090] Among them, Z in Indicates an incomplete sample set.

[0091] like Figure 2 As shown, an embodiment of the present invention provides a hydrocracking production process data preprocessing method and a hydrocracking production process data preprocessing system, comprising:

[0092] Production data acquisition module, which collects production data with missing data in the hydrocracking production process;

[0093] The missing data completion module uses the missing data completion method of multiple interpolation with masked denoising autoencoders based on self-attention mechanism (MI-SAMAE) to complete the key missing data in the production data;

[0094] The missing data completion method of the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) accurately fills in some missing data through the masked autoencoder based on the self-attention mechanism, and then improves the whole through the multiple interpolation method; the missing value estimated by the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) can be expressed as:

[0095]

[0096] Among them, Z in Indicates an incomplete sample set;

[0097] The missing data completion technology of the masked autoencoder based on the self-attention mechanism screens similar data through the self-attention mechanism, and then completes the missing data with similar data through the masked autoencoder; the minimization total objective function of the missing data completion technology of the masked autoencoder based on the self-attention mechanism (SAMAE) can be expressed as:

[0098]

[0099] 1. Specific application fields or related products of the present invention.

[0100] This invention provides a software platform for preprocessing data from the hydrocracking industrial production process. The software uses a self-attention mechanism (SA) to perform contextual analysis on industrial production data and tag similar data. It then uses a masked autoencoder (MAE) to accurately complete some missing data, achieving good data fitting results. Finally, multiple interpolation (MI) is used to complete the remaining missing data, ensuring data integrity. Based on this approach, the software constructs a self-attention-based masked denoising autoencoder multiple interpolation (MI-SAMAE) training model to complete missing data in the hydrocracking industrial production process.

[0101] Through simulation training and comparative experiments, it is proved that the software platform can be effectively used to complete the missing data in the initial distillation point data of diesel in the hydrocracking industrial production process; through comparative experiments with masked autoencoders (MAE), it is verified that it has a good data fitting effect; and through the calculation of root mean square error (RMSE), it is verified that its completion error is small and more consistent with the original data.

[0102] 2. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0103] This paper constructs a model and simulates the proposed masked denoising autoencoder multiple interpolation (MI-SAMAE) method based on the self-attention mechanism, and applies the trained model to complete the missing data in the diesel initial boiling point data in the hydrocracking industrial production process. The simulation results are as follows Figure 3 、 Figure 4 :

[0104] By observing the simulation training effect diagram, it can be seen that the data fitting effect and data filling effect of the model are good. In order to further verify its effectiveness and feasibility, the masked denoising autoencoder multiple interpolation based on the self-attention mechanism (MI-SAMAE) and the masked autoencoder (MAE) method are compared. The results of the comparison test are as follows: Figure 5 :

[0105] method RMSE MI-SAMAE 3.73 MAE 10.47

[0106] Table 1 Root mean square error of MI-SAMAE and MAE

[0107] The comparative test results show that the masked denoising autoencoder multiple interpolation (MI-SAMAE) method based on the self-attention mechanism has stronger data fitting and smaller error than the traditional masked autoencoder (MAE) method, and is more comprehensive and accurate in completing missing data.

[0108] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0109] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for preprocessing data in a hydrocracking production process, characterized in that: The following steps are involved: Step 1: collecting production data with missing data in the hydrocracking production process, wherein the production data includes production data with missing data output by each controller in the hydrocracking production process; Step 2: Calculate the weights between elements in the sequence data through the self-attention mechanism, understand the contextual information in the sequence, and mark data with similar change trends to facilitate more accurate processing of the sequence data; Step 3: Since there are missing data values ​​in the collected data, the original data is additionally masked by the masked autoencoder to recover the missing values ​​of the original data from the output data with similar change trends; Step 4: Combining the two processes in step 2 and step 3 to form a missing data completion technology based on a masked autoencoder with a self-attention mechanism to complete the missing data of key variables in the hydrocracking process and complete the key missing data in the production data; Step 5: Since the mask denoising autoencoder cannot fill in the missing data when there is no similar data in the high missing rate environment, a multiple interpolation method is used to repeatedly carry out several rounds of filling using a single filling method to obtain multiple different complete data sets, and finally integrate them to obtain the complete data; Step 6: Combining the two methods in step 4 and step 5 to form a missing data completion method based on multiple interpolation of mask denoising autoencoders, thereby improving the accuracy of data completion while reducing the computational difficulty; Use the missing data completion method of mask denoising autoencoder multiple interpolation based on self-attention mechanism to complete the key missing data in the production data; The missing data completion method of the masked denoising autoencoder multiple interpolation based on the self-attention mechanism accurately fills part of the missing data through the masked autoencoder based on the self-attention mechanism, and then improves the whole by the multiple interpolation method; wherein, the missing value estimated by the masked denoising autoencoder multiple interpolation based on the self-attention mechanism is expressed as: Where Z in Indicates an incomplete sample set; The missing data completion technology of the masked autoencoder based on the self-attention mechanism filters similar data through the self-attention mechanism, and then completes the missing data with similar data through the masked autoencoder; wherein, the minimization total objective function of the missing data completion technology of the masked autoencoder based on the self-attention mechanism is expressed as:

2. The method for preprocessing data of a hydrocracking production process according to claim 1, wherein: The production data includes production data with missing data output by each controller in the hydrocracking production process.

3. The method for preprocessing data of a hydrocracking production process according to claim 1, wherein: The self-attention mechanism calculates the weights between elements in the sequence data, understands the contextual information in the sequence, and marks data with similar change trends; The self-attention mechanism calculates the attention weight of the elements in the sequence through the softmax function: w ij =softmax(score(Q i ,K j )) Among them, score(Q i ,K j ) represents the attention score, Q i represents the query vector, K j Represents a key vector.

4. The method for preprocessing data of a hydrocracking production process according to claim 1, wherein: The masked autoencoder recovers the missing values ​​of the original data from the output data with similar change trends by additionally setting a mask on the original data; In a high missing rate environment, the input vector of the masked autoencoder is a partial observation vector. By learning from similar input data, the missing values ​​of the input data are supplemented.

5. The method for preprocessing data of a hydrocracking production process according to claim 4, wherein: During the MAE learning process, the masked autoencoder defines a new error term And only consider the contribution of the observations, as follows: Among them, i represents the set of observations on day i, 1 represents the indicator function, when x ij If ≠0, it is equal to 1, otherwise it is 0.

6. The method for preprocessing data of a hydrocracking production process according to claim 1, wherein: The multiple interpolation method uses a single interpolation method of mean interpolation, linear regression interpolation, deductive interpolation, nearest distance interpolation, hot card interpolation and cold card interpolation to repeatedly carry out several rounds of interpolation to obtain multiple different complete data sets, and interpolates the missing data that cannot be filled by the mask denoising autoencoder in a high missing rate environment and does not have similar data, and finally integrates them to obtain complete data.

7. A hydrocracking production process data preprocessing system according to any one of claims 1 to 6, characterized in that: include: Production data acquisition module, which collects production data with missing data in the hydrocracking production process; The missing data completion module uses a masked denoising autoencoder multiple interpolation method based on the self-attention mechanism to complete the key missing data in the production data; The missing data completion method of the masked denoising autoencoder multiple interpolation based on the self-attention mechanism accurately fills part of the missing data through the masked autoencoder based on the self-attention mechanism, and then improves the whole by the multiple interpolation method; wherein, the missing value estimated by the masked denoising autoencoder multiple interpolation based on the self-attention mechanism is expressed as: Where Z in Indicates an incomplete sample set; The missing data completion technology of the masked autoencoder based on the self-attention mechanism filters similar data through the self-attention mechanism, and then completes the missing data with similar data through the masked autoencoder; wherein, the minimization total objective function of the missing data completion technology of the masked autoencoder based on the self-attention mechanism is expressed as:

8. A computer device, characterized in that: The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the hydrocracking production process data preprocessing method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for preprocessing data of a hydrocracking production process according to any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal includes the hydrocracking production process data preprocessing system as described in claim 7.

Images