Generative adversarial network-based catalytic cracking process anomaly determination method and device

By using the historical and real-time data of the catalytic cracking device to perform timing information and multi-parameter correlation analysis based on the method of generating an adversarial network, the accuracy and applicability of abnormal diagnosis of catalytic cracking process in the prior art are solved, and catalytic cracking is realized Real-time, accurate abnormality detection and multi-dimensional diagnosis of the process.

CN118940599APending Publication Date: 2024-11-12CHINA PETROLEUM & CHEMICAL CORP +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310531840.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-11
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The prior art relies on individual experience and historical data in the diagnosis of abnormalities of catalytic cracking processes, and cannot effectively utilize timing information and multi-parameter correlation, resulting in low recognition accuracy and unable to cover most abnormal situations. The existing algorithms for catalytic cracking processes Poor applicability.

Method used

Using a method based on the generative adversarial network, the generative adversarial network model is trained and generated by obtaining the history and real-time data of the catalytic cracking device, and timing information and multi-parameter correlation analysis are performed to provide real-time and multi-dimensional abnormal diagnosis information.

Benefits of technology

Real-time and accurate abnormal detection of catalytic cracking processes is achieved, and it can cover known and unknown working conditions, improve the accuracy and response speed of diagnosis, and reduce the occurrence of safety accidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940599A_ABST
    Figure CN118940599A_ABST
Patent Text Reader

Abstract

The invention provides a generative adversarial network-based catalytic cracking process anomaly determination method and device. The method for determining the abnormality of the catalytic cracking process comprises the following steps: acquiring typical process data of at least one set of catalytic cracking device after running for a predetermined period, and dividing the typical process data into a training data set and a test data set; preprocessing the training data set, and training a generative adversarial network model based on the preprocessed training data set; the preprocessing comprises the following steps of: performing threshold division on data with production characteristics, and performing characteristic combination on discrete production data; and inputting the test data set into the trained generative adversarial network model to obtain a parameter anomaly detection result. According to the method provided by the invention, the process fluctuation of the catalytic cracking device can be detected and analyzed, and the detected process fluctuation can cover various known or unknown working conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent diagnosis and analysis of process anomalies in petrochemical enterprises, and in particular to a method for determining catalytic cracking process anomalies based on a generative adversarial network, a device for determining catalytic cracking process anomalies based on a generative adversarial network, and electronic equipment for implementing the method for determining catalytic cracking process anomalies. Background Art

[0002] While the petrochemical industry is developing rapidly, it is also facing various severe tests. Due to fluctuations in the chemical process, process personnel are unable to quickly and effectively adjust the process, resulting in process anomalies turning into accidents. Most of the accidents are caused by the complex process of petrochemical equipment, which requires the staff to have high professional capabilities and process experience. In fact, the level of process personnel is uneven. Some employees lack the ability or experience to deal with abnormal process fluctuations, which leads to their inability to effectively identify, diagnose, and analyze abnormal problems in a timely manner, and thus they cannot deal with them correctly and in a timely manner, resulting in process fluctuations turning into accidents. Therefore, providing process personnel with a real-time, fast, and accurate process parameter fluctuation data analysis method is an important reference and means for abnormal cause analysis, and is also an effective measure to adjust process abnormal fluctuations.

[0003] At present, the diagnosis method for abnormal process fluctuations is mainly based on threshold alarms, and the cause of process fluctuations and process adjustments are determined based on the historical trend of a single piece of data through the personal experience of process personnel. However, this traditional process abnormality analysis method is highly dependent on individual experience.

[0004] In addition, there are also some methods that use mathematical analysis models to diagnose process abnormal fluctuations. They are mainly based on algorithm analysis of historical data, summarizing the abnormal causes under different data fluctuations, and making quick judgments when similar abnormalities occur again. However, the existing process abnormality diagnosis and analysis methods have the following defects:

[0005] (1) Such methods are limited by the sample size of industrial historical data, the algorithm recognition accuracy, and the targeted analysis of a certain type of special process. When the sample size is too small, the algorithm model training effect is poor and cannot cover most abnormal situations; (2) The core algorithm of the model affects the accuracy of the overall analysis and recognition, but some existing algorithms only analyze a single piece of data and cannot use time series information to automatically associate other related process parameters affected by fluctuations, thereby more accurately using multi-data analysis to locate the cause; (3) Models trained using other process data cannot effectively analyze specific processes, which makes it difficult to apply various existing analysis methods to catalytic cracking processes.

[0006] For example, a Chinese patent document with a publication date of November 23, 2016 and a publication number of CN106156515A proposed a method and system for dynamic simulation of faults in a two-stage catalytic cracking regeneration system, which analyzes the dynamic characteristics of faults of the entire system under abnormal disturbances by reasonably setting up dynamic simulation of abnormal operating conditions. This method is based on DCS data and process models, and simulates the changes in data parameters under different operating conditions in advance, so as to analyze real-time data. However, the catalytic cracking process is complex, and often slight changes in influencing factors will lead to various abnormal operating conditions. The abnormalities simulated by the preset process parameter changes cannot cover all abnormal catalytic cracking conditions, so this method cannot monitor and analyze the abnormalities that are not included.

[0007] A Chinese patent document with a publication date of October 20, 2020 and publication number of CN111797533A proposes a method and system for detecting anomalies in operating parameters of a nuclear power plant, which can detect anomalies in real-time operating parameters and improve the accuracy of anomaly detection. This method uses the principal component analysis algorithm (PCA), which has a large amount of calculation and weak ability to analyze associated data, and is not suitable for multi-parameter diagnostic analysis; the algorithm has a good analysis effect on point anomalies, but it is difficult to analyze contextual anomalies and overall anomalies of time series data; the algorithm assumes that the variables follow a Gaussian distribution, and cannot obtain good detection and analysis results when the data does not follow a Gaussian distribution and has nonlinear dependencies.

[0008] A Chinese patent document with a publication date of September 22, 2020 and publication number of CN111695598A proposes a monitoring data anomaly diagnosis method based on a generative adversarial network, which is suitable for the anomaly diagnosis of long-term measured data of a structural health monitoring system. This method uses health monitoring data as a training set for training, so the trained model cannot be applied to anomaly detection and analysis of catalytic cracking process data; this method performs detection and analysis on a single indicator, and the combination of fluctuations in different parameter indicators often corresponds to different process anomalies, which is more complex, so the trained model cannot be applied; the catalytic cracking process parameters are much larger in magnitude than the health monitoring indicators, so there are higher requirements for the real-time monitoring efficiency of the model, and this method is difficult to apply in practice. Summary of the invention

[0009] In view of the technical problem in the prior art that there is a lack of effective training models for data anomaly analysis in specific processes, the present invention provides a method for determining catalytic cracking process anomalies based on a generative adversarial network. This method can analyze and summarize time series information, data fluctuation changes and related process parameters based on historical and real-time process data, thereby providing real-time, multi-dimensional information reference for process personnel to adjust catalytic cracking process fluctuations.

[0010] To achieve the above-mentioned purpose, the first aspect of the present invention provides a method for determining a catalytic cracking process anomaly based on a generative adversarial network, the method comprising: obtaining typical process data of at least one set of catalytic cracking units after a predetermined operation period, and dividing the data into a training data set and a test data set; preprocessing the training data set, and training a generative adversarial network model based on the preprocessed training data set; the preprocessing comprises: performing threshold division on data with production characteristics, and performing feature combination on discrete production data; and inputting the test data set into the trained generative adversarial network model to obtain parameter anomaly detection results.

[0011] In an exemplary embodiment of the present invention, the training of the generative adversarial network model based on the preprocessed training data set may include: initializing the discriminator D and the generator G; based on the loss function, training the discriminator D and the generator G using the preprocessed training data set until the model converges and a local optimal solution is obtained; wherein the discriminator D is used to judge the authenticity of the generated sample data compared to the real sample data in the training data set, and the generator G is used to generate generated sample data with a distribution similar to the real sample data, and the loss function is:

[0012]

[0013] In an exemplary embodiment of the present invention, the training of the discriminator D and the generator G using the preprocessed training data set based on the loss function may include: first fixing the discriminator D, inputting the random noise Z into the generator G, and optimizing the generator G based on the first part of the loss function; then fixing the generator G, inputting both the generated sample data and the real sample data into the discriminator D, and optimizing the discriminator D based on the second part of the loss function; updating the parameters of the discriminator D and the generator G, and repeating the above operations to carry out the next round of adversarial training; wherein,

[0014] The first part of the loss function is:

[0015] The second part of the loss function is:

[0016]

[0017] In an exemplary embodiment of the present invention, the preprocessing may further include: sorting the time series data by time to remove duplicate data; performing normalization processing on all data; and converting the processed data into an n×n two-dimensional matrix.

[0018] In an exemplary embodiment of the present invention, initializing the discriminator D may include: randomly initializing parameters in the discriminator D and the generator G using a random seed.

[0019] In an exemplary embodiment of the present invention, the predetermined period may be at least one maintenance period.

[0020] In an exemplary embodiment of the present invention, typical processes of the catalytic cracking unit may include a MIP process, a RFCC process, a VRFCC process, and a FCC process.

[0021] The second aspect of the present invention provides a catalytic cracking process anomaly determination device based on a generative adversarial network, the determination device comprising a parameter acquisition unit, a preprocessing unit, a training unit and an anomaly detection unit; wherein the parameter acquisition unit comprises an acquisition module and a division module, the acquisition module is used to acquire typical process data of at least one set of catalytic cracking units after a predetermined operation cycle, the division module is used to divide the acquired typical process data into a training data set and a test data set; the preprocessing unit comprises a feature combination module, the feature combination module is used to perform threshold division for data with production characteristics, and perform feature combination on discrete production data; the training unit is used to train a generative adversarial network model based on the preprocessed training data set; the anomaly detection unit is used to input the test data set into the trained generative adversarial network model, and output the parameter anomaly detection result.

[0022] In another exemplary embodiment of the present invention, the training unit may include a discriminator, a generator, an initialization module and an optimization update module; wherein the discriminator is used to judge the authenticity of generated sample data compared with real sample data in a training data set; the generator is used to generate generated sample data with a distribution similar to that of real sample data based on input random noise; the initialization module is used to initialize the parameters of the discriminator and the generator; the optimization update module is used to update the training discriminator and generator based on the loss function using the preprocessed training data set until the model converges and a local optimal solution is obtained.

[0023] In another exemplary embodiment of the present invention, the preprocessing unit may also include a sorting module, a normalization module and a two-dimensional conversion module; wherein the sorting module is used to sort the time series data by time to remove duplicate data; the normalization module is used to perform normalization processing on all data; and the two-dimensional conversion module is used to convert the processed data into an n×n two-dimensional matrix.

[0024] A third aspect of the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by one or more of the above processors so that the computer implements the catalytic cracking process abnormality determination method as described above.

[0025] A fourth aspect of the present invention provides a computer-readable storage medium, in which at least one program code is stored. The program code is loaded and executed by a processor to enable a computer to implement the catalytic cracking process abnormality determination method as described above.

[0026] Through the technical solution provided by the present invention, the present invention has at least the following technical effects:

[0027] (1) The present invention provides a method for determining process parameter anomalies based on a generative adversarial network algorithm for a catalytic cracking unit. The generative adversarial network algorithm model is trained using historical process data of the catalytic cracking unit. The method can detect and analyze fluctuations of various processes (e.g., catalytic cracking MIP process, RFCC process, VRFCC process, and FCC process, etc.). The detected process fluctuations cover various known or unknown operating conditions.

[0028] (2) The present invention uses a generative adversarial network training model to perform real-time diagnostic analysis on the process data of the catalytic cracking unit. In addition to single-parameter detection and analysis, it is also effective for multi-parameter data association analysis, and can quickly and effectively perform abnormality detection and analysis on a large amount of real-time data of the catalytic cracking process unit;

[0029] (3) The generative adversarial network training model used in the present invention is trained based on historical process data. In addition to being effective for point anomaly detection, it also has good effects on contextual anomaly and overall anomaly detection of time series data.

[0030] (4) The generative adversarial network training model used in the present invention can detect and analyze the real-time data of nonlinear dependencies, and can accurately determine whether the discrete analysis results of the feature vector after dimensionality reduction are abnormal for the catalytic cracking unit process;

[0031] (5) The present invention can diagnose and analyze all abnormal fluctuation conditions of catalytic cracking processes (e.g., MIP process, RFCC process, VRFCC process and FCC process), and provide real-time, multi-dimensional information reference for process personnel to adjust catalytic cracking process fluctuations, thereby advancing the abnormal handling response time and reducing the occurrence of production safety accidents;

[0032] (6) Compared with the existing training algorithms, the training model and training method of the present invention are specially designed for the catalytic cracking process, and have excellent effect on the diagnosis of parameter abnormalities in the catalytic cracking process, and the accuracy of the diagnostic results is also higher.

[0033] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0035] Figure 1 A flow chart of a method for determining abnormality in a catalytic cracking process provided by an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of the structure of a catalytic cracking process abnormality determination device provided by an embodiment of the present invention;

[0037] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0038] Description of Reference Numerals

[0039] 101- parameter acquisition unit, 102- preprocessing unit, 103- training unit, 104- anomaly detection unit, 1011- acquisition module, 1012- division module, 1021- sorting module, 1022- normalization module, 1023- feature combination module, 1024- two-dimensional conversion module, 1031- discriminator, 1032- generator, 1033- initialization module, 1034- optimization update module, 201- processor, 202 memory. DETAILED DESCRIPTION

[0040] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0041] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0042] In the present invention, “first”, “second”, etc. are merely used for convenience of description and distinction and cannot be understood as indicating or implying relative importance.

[0043] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installation", "connection", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integrated connection; it can be a direct connection, or an indirect connection; it can be a wired connection, or a wireless connection. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0044] At present, existing model algorithms are all based on historical data to perform abnormal analysis of process parameter fluctuation data. This type of method is not only limited by the number of samples of industrial historical data, but also has poor training effects when the number of samples is too small and cannot cover most abnormal situations; and the trained model cannot effectively analyze specific processes. In addition, some algorithms only analyze a single piece of data and cannot use time series information to automatically associate other related process parameters affected by fluctuations, making it impossible to more accurately locate the cause using multi-data analysis.

[0045] In order to solve the above problems, the present invention proposes a process parameter abnormality diagnosis and analysis model based on a generative adversarial network algorithm for typical processes of catalytic cracking units (for example, RFCC, VRFCC, FCC, MIP and other types of catalytic cracking processes), and uses a large amount of industrial process data of long-term operation of existing domestic catalytic cracking units as a basis to train the analysis model, thereby establishing an automatic analysis algorithm based on historical and real-time process data, analyzing and summarizing time series information, fluctuation changes and related associated process parameters, and providing real-time, multi-dimensional information reference for process personnel to adjust process fluctuations, thereby advancing the response time of abnormal handling and reducing the occurrence of production safety accidents.

[0046] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0047] Embodiment 1

[0048] A first embodiment of the present invention provides a method for determining abnormality in a catalytic cracking process based on a generative adversarial network, the method comprising the following steps:

[0049] Step S101: Acquire typical process data of at least one set of catalytic cracking unit after running for a predetermined period, and divide the data into a training data set and a test data set.

[0050] Here, the predetermined period may be at least one maintenance period to ensure that the catalytic cracking unit operates under all operating conditions within the period, so that various process parameters covering the entire process flow of the catalytic cracking unit can be obtained.

[0051] There are many types of typical processes for catalytic cracking units, such as RFCC (heavy oil catalytic cracking process of American Shi Wei Company), VRFCC (decompression residue catalytic cracking process of Sinopec), FCC (heavy oil catalytic cracking process of Universal Oil Products), MIP (catalytic cracking process with high production of isoparaffins), etc. A typical process data can be selected for model training to obtain a parameter anomaly analysis model for a specific process of the catalytic cracking unit. For example, various process parameters of multiple sets of catalytic cracking MIP process units that have been running for many years and for a long period of time can be collected as sample training. The analysis capability of the model trained in this way can almost cover all abnormal fluctuation conditions of the catalytic cracking MIP process, and can perform more accurate diagnosis and analysis.

[0052] The acquired process data can be divided into training data sets and test data sets according to the operating time. For example, the historical operating data of the catalytic cracking unit can be divided into the training data set and used as training samples to complete the training of the generative adversarial network model. The real-time operating data of the catalytic cracking unit can be divided into the test data set for input into the trained generative adversarial network model to complete the abnormal diagnosis and detection of real-time data.

[0053] Step S102: training a generative adversarial network model based on the preprocessed training data set.

[0054] Exemplarily, based on the preprocessed training data set, the process of training the generative adversarial network model includes but is not limited to the following sub-steps S1021 to S1023.

[0055] Sub-step S1021, pre-processing the training data set.

[0056] For example, the process of preprocessing the training data set includes but is not limited to the following sub-steps S1021a to S1021d.

[0057] Sub-step S1021a, sort the time series data by time to remove duplicate data.

[0058] Sub-step S1021b: normalize all data.

[0059] Sub-step S1021c: threshold division is performed on the data with production characteristics, and feature combination is performed on the discrete production data. By performing feature combination on the discrete data in the training sample, model training for data association analysis of multi-parameter combination can be enhanced.

[0060] For example, the standard range of some parameters at certain points in the process, such as the temperature and pressure of a certain device, has a standard range of values ​​in the normal operation of the catalytic cracking unit. If the parameters exceed this threshold range, it can be considered that an abnormality has occurred. Therefore, a threshold can be set for some points according to the normal operating values ​​during process design, so as to find discrete points in the training data.

[0061] In addition, when an abnormality occurs in the catalytic cracking unit, multiple parameters often change at the same time, that is, the change of one parameter is often associated with the change of parameters at different points. For example, when both A and B abnormalities occur, the data at point a will become smaller; but when A abnormality occurs, in addition to a becoming smaller, b and c will become larger; and when B abnormality occurs, in addition to a becoming smaller, d will also become smaller. Therefore, by combining and correlating multiple parameters as a preprocessing step, the training effect of the model can also be enhanced.

[0062] Sub-step S1021d: convert the training data set into a two-dimensional matrix of n×n. By converting the training samples into two dimensions, sample features can be better extracted and learned.

[0063] Sub-step S1022: Initialize the discriminator D and the generator G. For example, the parameters in the discriminator D and the generator G may be randomly initialized using a random seed.

[0064] Sub-step S1023: Based on the loss function, the discriminator D and the generator G are continuously trained using the preprocessed training data set until the model converges and a local optimal solution is obtained.

[0065] Exemplarily, the loss function for a catalytic cracking unit may be set as:

[0066]

[0067] The goal of the generative adversarial network is to make the generator G generate samples that are sufficient to deceive the discriminator D. From a mathematical point of view, it is hoped that the generated samples and the real samples have the same probability distribution, or it can be said that the generated samples and the real samples have the same probability density function, that is, p G (x) = p data (x). As for how to judge whether the generated samples satisfy the above relationship, we can use the loss function to optimize G and obtain the optimal generator G*.

[0068] In the loss function, G is a generator network that receives a random noise z (random number) and generates typical process parameters through the noise. G(z) represents the process parameters generated by the G network, and D(G(z)) represents the probability that the D network determines whether the process parameters generated by G are true. G* refers to the optimal generator.

[0069] D is a network that determines whether the process parameter data is "real". Its input parameter is x, which represents a process parameter, and its output parameter is D(x). D(x) represents the probability that the D network determines whether the process parameter is real. If D(x) is 1, it represents a 100% real process parameter. If D(x) is 0, it represents an impossible process parameter.

[0070] E represents expectation, Ex~p data (x) represents the expectation of the discriminator for the judgment result of the real sample; Ez~p z (z) represents the expectation of the discriminator for the false sample judgment result; Ex,y~p data represents the expectation of the distribution of real samples and false samples; D s (x) represents the standard discriminator function; λ s Represents a hyperparameter, which is a coefficient used to measure the weight of each part of the loss function.

[0071] For the optimization of the discriminator D, it is required that the discriminator D can satisfy p data (x) The sample with the distribution is identified as a positive sample. That is, the max() formula means that the generator G is fixed and the discriminator D is asked to distinguish the real samples from the fake samples as much as possible. That is, it is hoped that the discriminator D can distinguish the real samples from the generated samples when the generator G does not move.

[0072] For the optimization of the generator G, it is required that the data generated by the generator G is as similar as possible to the data in the training data set, that is, the same data distribution. The min() formula means that the discriminator D is fixed, and by adjusting the generator G, it is hoped that the discriminator D will make mistakes and try not to let the discriminator D distinguish between real samples and generated samples.

[0073] Exemplarily, based on the loss function, the process of continuously training the discriminator D and the generator G using the preprocessed training data set includes but is not limited to the following sub-steps S1023a to S1023c.

[0074] Sub-step S1023a, first fix the discriminator D, input the random noise Z into the generator G, and then optimize the generator G based on the first part of the loss function.

[0075] Among them, the calculation formula of the first part of the loss function is:

[0076]

[0077] Sub-step S1023b, fix the generator G again, send both the generated sample data and the real sample data to the discriminator D, and then optimize the discriminator D based on the second part of the loss function.

[0078] Among them, the second part of the loss function is:

[0079]

[0080] Sub-step S1023c, update the parameters of the discriminator D and the generator G, and repeat the above operations to carry out the next round of adversarial training.

[0081] Step S103: input the test data set into the trained generative adversarial network model to obtain parameter anomaly detection results.

[0082] The implementation environment of the embodiment of the present application includes at least one terminal and a server, and the method is executed on the terminal or the server respectively. The terminal and the server can be connected in communication to realize the interactive transmission of information. The terminal and the server perform model training by acquiring typical process data of the catalytic cracking unit to determine the parameter anomaly detection result.

[0083] Among them, the terminal can be any electronic product that can interact with the user through one or more methods such as keyboard, touchpad, touch screen, voice interaction, etc., such as PC (Personal Computer), PPC (Pocket Personal Computer), tablet computer, etc.

[0084] The server can be a single server or a server cluster composed of multiple servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0085] It should be noted that the training model and training method of this embodiment can not only be used to detect abnormal operating conditions of catalytic cracking units, but are also effective for various chemical process flows of other petrochemical units. However, the training model is more suitable for abnormal diagnosis of catalytic cracking processes and has the best diagnostic effect.

[0086] Here, petrochemical units mainly refer to the process equipment used in the operation of various chemical process flows in the process of crude oil splitting and separation and processing of various petroleum fractions in oil refining enterprises. Common process flows in petroleum refining include atmospheric and vacuum distillation, catalytic cracking, delayed coking, hydrocracking, solvent deasphalting, hydrofining and catalytic reforming. Petrochemical units can be atmospheric and vacuum units, catalytic cracking units, hydrocracking units, delayed coking units, etc.

[0087] Embodiment 2

[0088] The second embodiment of the present invention takes the MIP process of a catalytic cracking unit as an example to further explain in detail how to perform abnormal fluctuation detection and analysis of the catalytic cracking process by training a generative adversarial network model.

[0089] Please refer to Figure 1 , a method for determining abnormality in catalytic cracking process based on generative adversarial network can be implemented through the following steps.

[0090] Step 1: Collect historical data of various parameters of four large-scale MIP process catalytic cracking units in China over a long period (one maintenance cycle) as the sample set for subsequent model training.

[0091] During the data collection process, we chose to collect data from four MIP process catalytic cracking units within one maintenance cycle because we needed to ensure that the sample cycle used was long and the number was large, so as to improve the accuracy of model diagnosis and identification.

[0092] After the collection is completed, the collected data can be divided into a training data set and a test data set.

[0093] The use of specific data (i.e., typical process data) for training ensures the targeted nature of data identification and enhances the applicability of the model in the scenario of process fluctuation diagnosis and analysis in catalytic cracking units.

[0094] Step 2: Use the long-term operation historical data training set of the catalytic cracking MIP process to train the generative adversarial network algorithm model.

[0095] Sub-step 2-1: training data preprocessing, which may specifically include the following four sub-steps.

[0096] (1) Sort the time series data by time and remove duplicate data.

[0097] (2) Normalize the data to improve the data processing efficiency of the subsequent generative adversarial network learning model.

[0098] (3) Some data with production characteristics can be divided by thresholds, and discrete production data can be combined with features to enhance model training for data association analysis of multi-parameter combinations.

[0099] (4) The training samples are converted into a two-dimensional matrix of n×n to better extract and learn the sample features.

[0100] Step 2-2: Initialize the parameters of the discriminator D and the generator G. For example, use a random seed to randomly initialize the parameters of the discriminator D and the generator G.

[0101] Steps 2-3: Use adversarial training to conduct adversarial games.

[0102] (1) Random noise is fed into the generator G as input, with the goal of generating pseudo data samples that are difficult for the discriminator to determine whether they are true or false.

[0103]

[0104] (2) Both generated samples and real samples are sent to the discriminator D. The goal of the discriminator is to determine which samples come from the generator and which samples are real samples.

[0105]

[0106] Steps 2-4: Continuously update the parameters of the discriminator D and the generator G to complete a round of training.

[0107]

[0108] Sub-step 2-5: Repeat the above sub-steps 2-3 and 2-4 until the model converges, that is, the local optimal solution is found and the training is completed.

[0109] Step 3: Real-time data anomaly diagnosis and detection: Based on the collected real-time MIP process parameters of the catalytic cracking unit, the training model is used to perform process abnormal fluctuation diagnosis and analysis.

[0110] Step 3-1: Collect various real-time MIP process parameters of the catalytic cracking unit as a test data set.

[0111] Step 3-2: Use the trained generative adversarial network model to perform abnormal diagnosis and analysis on real-time process parameters.

[0112] Step 3-3: Output anomaly detection results through model calculation.

[0113] Steps 3-4: The process operator handles the process anomaly based on the abnormality diagnosis and analysis results.

[0114] The embodiment of the present application is based on the model training of various process parameters of many domestic catalytic cracking MIP process units for many years of long-term operation, and improves the traditional generative adversarial network algorithm, which improves the multivariate anomaly detection capability for time series data, and can detect and analyze the real-time data of data association, context anomaly and overall anomaly of multi-parameter combination, as well as nonlinear dependency, making it more suitable for real-time anomaly analysis of massive process data of catalytic cracking. This method is not only for the 4 catalytic cracking units collected in the embodiment of the present application, but can also be applied to the real-time diagnosis and analysis of process abnormality fluctuations of other MIP process catalytic cracking units.

[0115] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0116] (1) The generative adversarial network algorithm model is trained using historical data from the catalytic cracking unit, which can detect and analyze various process fluctuations. Compared with the anomalies simulated by changes in preset process parameters, the process fluctuations detected by this method can cover various known or unknown operating conditions.

[0117] (2) The generative adversarial network algorithm is used to train the model to perform real-time diagnostic analysis on the device data. In addition to single-parameter detection and analysis, it is also effective for abnormal fluctuation diagnosis of multi-parameter combination data correlation analysis.

[0118] (3) This patented algorithm improves and upgrades the traditional generative adversarial network algorithm, enhancing its ability to detect multivariate anomalies in time series data, making it more suitable for anomaly analysis of catalytic cracking process data.

[0119] (4) After the model is trained based on historical process data, in addition to point anomaly detection, it also has good results in detecting contextual anomalies and overall anomalies of time series data.

[0120] (5) It can detect and analyze real-time data with nonlinear dependencies, and compared with other analysis methods, it can accurately analyze highly complex catalytic cracking process unit data based on historical data model training results.

[0121] (6) Various process parameters of multiple domestic catalytic cracking MIP process units that have been in operation for many years and for a long period of time are collected as sample training sets. The trained model has the ability to analyze almost all abnormal fluctuation conditions of the catalytic cracking MIP process. Compared with the method models trained with other training sets, it can perform diagnostic analysis on catalytic cracking process data more accurately.

[0122] (7) The generative adversarial network algorithm used has been trained to perform fast and effective anomaly detection and analysis on large amounts of real-time data from catalytic cracking process units.

[0123] It should be noted that the training model of this embodiment is trained using data from the MIP process, so the model has better results, more targetedness and applicability for abnormal diagnosis and analysis of the MIP process of catalytic cracking. In addition, the training model of this embodiment can also be used to complete abnormal diagnosis and analysis of other types of catalytic cracking processes (e.g., VRFCC processes). Although its diagnostic effect is slightly lower than the accuracy of predicting the MIP process, it is still slightly higher than the accuracy of other training models in the prior art.

[0124] In other words, the historical data of a specific process can be used as a training sample, and after model training, the abnormal fluctuation condition detection of the specific process can be completed; the historical data of a specific process can also be used as a training sample, and after model training, the abnormal fluctuation condition detection of all catalytic cracking processes can be completed. For example, when training a model, its input data can be specific process data (for example, MIP data), and when using the training model to perform process abnormal fluctuation diagnosis and analysis, its input real-time detection data can be any type of catalytic cracking process data, that is, it can be real-time data of the MIP process, or real-time data of processes such as VRFCC, RFCC, or FCC.

[0125] Embodiment 3

[0126] A third embodiment of the present invention provides a catalytic cracking process abnormality determination device based on a generative adversarial network. The determination device may include a parameter acquisition unit, a preprocessing unit, a training unit, and an abnormality detection unit.

[0127] The parameter acquisition unit is used to acquire historical operation data of at least one set of catalytic cracking units. For example, the parameter acquisition unit includes an acquisition module and a division module; the acquisition module is used to acquire typical process data of at least one set of catalytic cracking units after a predetermined operation period; the division module is used to divide the acquired typical process data into a training data set and a test data set.

[0128] The preprocessing unit is connected to the parameter acquisition unit and is used to preprocess the training data set. The preprocessing at least includes: combining the discrete production data with multiple parameter features.

[0129] The training unit is connected to the preprocessing unit and is used to train a generative adversarial network model based on the preprocessed training data set.

[0130] The anomaly detection unit is connected to the training unit, and is used to input the test data set into the trained generative adversarial network model and output the parameter anomaly detection result.

[0131] Embodiment 4

[0132] The fourth embodiment of the present invention provides a catalytic cracking process abnormality determination device based on a generative adversarial network, see Figure 2 The determination device may include a parameter acquisition unit 101, a preprocessing unit 102, a training unit 103 and an anomaly detection unit 104.

[0133] The parameter acquisition unit 101 is composed of an acquisition module 1011 and a division module 1012. The acquisition module 1011 is used to acquire typical process data of at least one catalytic cracking unit after a predetermined operation period; the division module 1012 is used to divide the acquired typical process data into a training data set and a test data set.

[0134] The preprocessing unit 102 is connected to the parameter acquisition unit 101, and includes a sorting module 1021, a normalization module 1022, a feature combination module 1023 and a two-dimensional conversion module 1024. The sorting module 1021 is used to sort the time series data by time to remove duplicate data. The normalization module 1022 is used to perform normalization processing on all data. The feature combination module 1023 is used to perform threshold division on data with production characteristics and perform feature combination on discrete production data. The two-dimensional conversion module 1024 is used to convert the processed data into an n×n two-dimensional matrix.

[0135] The training unit 103 is connected to the preprocessing unit 102, and includes a discriminator 1031, a generator 1032, an initialization module 1033 and an optimization update module 1034. Among them, the discriminator 1031 is connected to the generator 1032 and the preprocessing unit 102 respectively, and is used to judge the authenticity of the generated sample data compared with the real sample data in the training data set. The generator 1032 is used to generate generated sample data with a distribution similar to the real sample data based on the input random noise. The initialization module 1033 is connected to the discriminator 1031 and the generator 1032 respectively, and is used to initialize the parameters of the discriminator and the generator. The optimization update module 1034 is connected to the discriminator 1031 and the generator 1032 respectively, and is used to continuously update the training discriminator and the generator based on the loss function using the preprocessed training data set until the model converges and obtains a local optimal solution.

[0136] The anomaly detection unit 104 is connected to the parameter acquisition unit 101 and the training unit 103 respectively, and is used to input the test data set into the trained generative adversarial network model and output the parameter anomaly detection result.

[0137] Above Figure 2 When the device provided realizes its functions, only the division of the above-mentioned functional modules is used as an example. In practical applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0138] Embodiment 5

[0139] A fifth embodiment of the present invention further provides an electronic device, see Figure 3 The electronic device includes a processor 201 and a memory 202, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by one or more of the above-mentioned processors to enable the computer to implement the catalytic cracking process abnormality determination method as described above.

[0140] Of course, the electronic device may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The electronic device may also include other components for realizing various functions of the device, which will not be described in detail here.

[0141] Embodiment 6

[0142] The sixth embodiment of the present invention further provides a computer-readable storage medium, in which at least one program code is stored. The program code is loaded and executed by a processor to enable a computer to implement the catalytic cracking process abnormality determination method as described above.

[0143] Optionally, the computer readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, and an optical disc data storage device. Those skilled in the art may understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc, and other media that can store program codes.

[0144] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0145] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0146] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for determining abnormality in catalytic cracking process based on generative adversarial networks, It is characterized in that The method comprises: Acquire typical process data of at least one set of catalytic cracking unit after a predetermined operation period, and divide the data into a training data set and a test data set; Preprocessing the training data set, and training a generative adversarial network model based on the preprocessed training data set; the preprocessing includes: performing threshold division on data with production characteristics, and combining features of discrete production data; The test dataset is input into the trained generative adversarial network model to obtain parameter anomaly detection results.

2. The method for determining abnormality in a catalytic cracking process based on a generative adversarial network according to claim 1, It is characterized in that The method of training a generative adversarial network model based on the preprocessed training data set includes: Initialize the discriminator D and generator G; Based on the loss function, the discriminator D and the generator G are trained using the preprocessed training data set until the model converges and a local optimal solution is obtained; Among them, the discriminator D is used to judge the authenticity of the generated sample data compared with the real sample data in the training data set, and the generator G is used to generate generated sample data with a distribution similar to the real sample data. The loss function is:

3. The method for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 2, It is characterized in that The method of training the discriminator D and the generator G based on the loss function and using the preprocessed training data set includes: First, fix the discriminator D, input the random noise Z into the generator G, and then optimize the generator G based on the first part of the loss function; Then fix the generator G, send both the generated sample data and the real sample data to the discriminator D, and optimize the discriminator D based on the second part of the loss function; Update the parameters of the discriminator D and the generator G, and repeat the above operations to carry out the next round of adversarial training; Among them, the first part of the loss function is: The second part of the loss function is:

4. The method for determining abnormality in a catalytic cracking process based on a generative adversarial network according to claim 2, It is characterized in that The pre-processing further comprises: Sort the time series data by time to remove duplicate data; All data are normalized; The processed data is converted into an n×n two-dimensional matrix.

5. The method for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 2, It is characterized in that The initializing the discriminator D includes: using a random seed to randomly initialize parameters in the discriminator D and the generator G.

6. The method for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 1, It is characterized in that The predetermined period is at least one maintenance period.

7. The method for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 2, It is characterized in that Typical processes of the catalytic cracking unit include MIP process, RFCC process, VRFCC process and FCC process.

8. A device for determining abnormality in catalytic cracking process based on generative adversarial network, It is characterized in that The determination device includes a parameter acquisition unit, a preprocessing unit, a training unit and an anomaly detection unit; wherein, The parameter acquisition unit includes an acquisition module and a division module, wherein the acquisition module is used to acquire typical process data of at least one set of catalytic cracking unit after a predetermined operation period, and the division module is used to divide the acquired typical process data into a training data set and a test data set; The preprocessing unit includes a feature combination module, which is used to perform threshold division on data with production characteristics and perform feature combination on discrete production data; The training unit is used to train a generative adversarial network model based on the preprocessed training data set; The anomaly detection unit is used to input the test data set into the trained generative adversarial network model and output the parameter anomaly detection result.

9. The device for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 8, It is characterized in that The training unit includes a discriminator, a generator, an initialization module and an optimization update module; wherein, The discriminator is used to judge whether the generated sample data is true or false compared to the real sample data in the training data set; The generator is used to generate generated sample data similar to the distribution of real sample data according to the input random noise; The initialization module is used to initialize the parameters of the discriminator and the generator; The optimization update module is used to update the training discriminator and generator based on the loss function using the preprocessed training data set until the model converges and obtains a local optimal solution.

10. The device for determining abnormality in catalytic cracking process based on generative adversarial network according to claim 8, It is characterized in that The preprocessing unit also includes a sorting module, a normalization module and a two-dimensional conversion module; Wherein, the sorting module is used to sort the time series data by time to remove duplicate data; The normalization module is used to perform normalization processing on all data; The two-dimensional conversion module is used to convert the processed data into an n×n two-dimensional matrix.

11. An electronic device, It is characterized in that The electronic device includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by one or more of the above processors so that the computer implements the catalytic cracking process abnormality determination method described in any one of claims 1 to 7.

12. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores at least one program code, and the program code is loaded and executed by a processor so that the computer implements the catalytic cracking process abnormality determination method according to any one of claims 1 to 7.