Working condition monitoring method and system for petrochemical industrial process and medium

Abstract

The invention provides a working condition monitoring method for a petrochemical industrial process, a working condition monitoring system for the petrochemical industrial process and a computer readable storage medium. The working condition monitoring method for the petrochemical industrial process comprises the following steps: acquiring monitoring data and extracting time sequence data; the time series data are input to a monitoring model, the monitoring model comprises a gating circulation unit network and a Gaussian mixture variational auto-encoder with a Copula function introduced, the gating circulation unit network is used for extracting potential representation of the time series data and determining distribution parameters of potential variables, and the Copula function is used for inputting the potential representation of the time series data into the Gaussian mixture variational auto-encoder; a Gaussian mixture variational auto-encoder introducing a Copula function realizes potential variable modeling according to distribution parameters of potential variables; based on the monitoring model, using a kernel density estimation method to calculate monitoring statistics and setting a control limit of the monitoring statistics; and performing anomaly detection according to the monitoring statistics and the control limit of the monitoring statistics.

Description

Technical Field

[0001] This invention relates to the petrochemical industry, and more particularly to a method for monitoring operating conditions in petrochemical processes, a system for monitoring operating conditions in petrochemical processes, and a computer-readable storage medium. Background Technology

[0002] As one of the core units in the oil refining process, the catalytic cracking (FCC) process faces numerous challenges in actual operation, including complex and variable operating conditions, numerous and coupled process variables, and strict control of performance indicators. Similar to the crude oil distillation unit (CDU), the feedstock properties processed by the catalytic cracking unit fluctuate greatly, and the production conditions change frequently. This makes it difficult to accurately monitor the operating conditions based on performance indicators such as energy consumption (EC), and it is also impossible to effectively distinguish between different operating states (such as high-load conditions, low-load conditions, and transitional conditions).

[0003] In the actual data modeling process of FCC, due to the long operating cycle and complex and variable operating conditions of the equipment, there are complex temporal correlations and nonlinear relationships among the data obtained from a large number of measurement points. This not only makes it difficult to classify the data into detailed operating conditions, but also increases the difficulty of anomaly detection and operating condition identification.

[0004] Furthermore, the data collected during the FCC process exhibits non-uniform time-delay dynamics. Non-uniform time-delay dynamics refer to the unpredictable random disturbances that occur in modern industrial production as raw materials are transformed into final products through multiple process units. Simultaneously, due to the varying reaction times at each process stage, the variables collected by different sensors exhibit varying degrees of autocorrelation. Autocorrelation, or the degree of correlation between a variable at different times, reflects the relationship between the variable's current value and its past values. If a variable exhibits strong autocorrelation, meaning high dynamics, and this characteristic is not considered when constructing the monitoring model, it will lead to a decline in the model's performance, thereby affecting the accuracy of process monitoring.

[0005] In order to overcome the above-mentioned defects of the existing technology, there is an urgent need in the field for a method for monitoring operating conditions in petrochemical industrial processes that can effectively extract the dynamic characteristics of FCC process data, accurately learn the complex operating condition distribution in FCC process, significantly improve the accuracy of operating condition monitoring and anomaly detection, and effectively reduce the false alarm rate. Summary of the Invention

[0006] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form to prepare for the more detailed descriptions that follow.

[0007] To overcome the aforementioned deficiencies in the existing technology, this invention provides a method for monitoring operating conditions in petrochemical industrial processes, a system for monitoring operating conditions in petrochemical industrial processes, and a computer-readable storage medium. These methods can effectively extract the dynamic characteristics of data from FCC processes, accurately learn the complex operating condition distribution in FCC processes, significantly improve the accuracy of operating condition monitoring and anomaly detection, and effectively reduce the false alarm rate.

[0008] Specifically, the above-described method for monitoring operating conditions in a petrochemical process according to a first aspect of the present invention includes the following steps: acquiring monitoring data and extracting time-series data; inputting the time-series data into a monitoring model, the monitoring model including a gated recurrent unit network and a Gaussian mixture variational autoencoder incorporating a Copula function, wherein the gated recurrent unit network is used to extract the latent representation of the time-series data and determine the distribution parameters of the latent variables, and the Gaussian mixture variational autoencoder incorporating a Copula function models the latent variables based on the distribution parameters of the latent variables; calculating a monitoring statistic based on the monitoring model using a kernel density estimation method and setting control limits for the monitoring statistic; and performing anomaly detection based on the monitoring statistic and the control limits of the monitoring statistic.

[0009] Preferably, in one embodiment of the present invention, the step of acquiring monitoring data and extracting time series data further includes: distinguishing the operating conditions of the acquired monitoring data based on a clustering analysis method, wherein the clustering analysis method includes the K-means clustering method; and using a sliding window method to perform alignment preprocessing on the monitoring data under different operating conditions to extract time series data under different operating conditions.

[0010] Preferably, in one embodiment of the present invention, the step of the gated recurrent unit network extracting the latent representation of the time series data and determining the distribution parameters of the latent variables includes:

[0011] The time series data is x = {x1, x2, ..., x} T The gated recurrent unit network encodes the time series data information into the final hidden state h by recursively calculating the hidden state. T As a potential representation of the time series data:

[0012] h T =GRU(x T ,h T-1 ),

[0013] Among them, h T-1 It is the final hidden state determined based on the time series data input at time T-1, and GRU is the gated recurrent unit network;

[0014] The hidden state h output by the gated recurrent unit network is transformed by linear transformation. T Mapped to the distribution parameters of the latent variables:

[0015] μ g ,σ g =f GRU (h T ),

[0016] Among them, f GRU (·) is a linear transformation, μ g σ is the mean that maps to the Gaussian mixture model. g The standard deviation is mapped to the Gaussian mixture model, where g represents different operating conditions.

[0017] Preferably, in one embodiment of the present invention, the latent variable modeling includes the step of: the distribution of the latent variables is:

[0018]

[0019] Among them, Q φ It is the conditional probability distribution under encoder parameters φ, z is the latent variable output by the encoder, D is the dimension of the latent variable, and C is the conditional probability distribution under encoder parameters φ. ψ It is the Copula function, The latent variable z is calculated from the Gaussian cumulative distribution function. i marginal distribution The mean is μ g variance is The Gaussian distribution.

[0020] Preferably, in one embodiment of the present invention, the loss function of the monitoring model takes into account data reconstruction error, latent variable regularization, and correlation between variables.

[0021] Preferably, in one embodiment of the present invention, the loss function is:

[0022]

[0023] Where λ1 and λ2 are weight parameters, It is the data reconstruction error term, D KL (Qφ (z|x)‖P(z)) is the KL divergence term, ∑ i,j log C ψ (z i ,z j ) is a Copula related term.

[0024] Preferably, in one embodiment of the present invention, the monitoring statistics include the SPE statistic and the T statistic. 2 The SPE statistic is calculated using the following formula:

[0025]

[0026] Where, x i Represents the original input data. This represents the generated reconstructed data, where m represents the dimension of the data.

[0027] The T 2 The formula for calculating the statistic is:

[0028] T 2 =(z-μ) z ) T ∑ -1 (z-μ z ),

[0029] Where z is the latent variable output by the variational autoencoder, and μ z Let be the mean vector of the latent variables, and ∑ be the covariance matrix of the latent variables.

[0030] Preferably, in one embodiment of the present invention, the formula for calculating the control limit of the SPE statistic is:

[0031]

[0032] in, Let σ be the mean of the SPE statistic on the training dataset. SPE The standard deviation of the SPE statistic is given by , and k represents the set threshold parameter.

[0033] The T 2 The formula for calculating the control limits of a statistic is:

[0034]

[0035] Where p is the dimension of the latent variable. The value is the 95th quantile of a chi-square distribution with p degrees of freedom.

[0036] Preferably, in one embodiment of the present invention, the step of anomaly detection based on the monitoring statistic and the control limit of the monitoring statistic includes: responding to the T 2 The statistic is less than or equal to the stated T 2 If the SPE statistic is less than or equal to the control limit of the SPE statistic, the data detection result is considered normal; in response to the T 2 The statistic is less than or equal to the stated T 2 If the SPE statistic exceeds the control limit of the SPE statistic and the SPE statistic is greater than the control limit of the SPE statistic, the data detection result is a first abnormal state; or in response to the T 2 The statistic is greater than the stated T 2 The control limits of the statistics are defined as follows: the data detection result is a second abnormal state, and the degree of abnormality of the first abnormal state is less than that of the second abnormal state.

[0037] Furthermore, the above-described operating condition monitoring system for petrochemical industrial processes provided according to a second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the operating condition monitoring method for petrochemical industrial processes provided in any of the above embodiments.

[0038] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions. When the computer instructions are executed by a processor, the operating condition monitoring method for petrochemical industrial processes provided in any of the above embodiments is implemented. Attached Figure Description

[0039] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related properties or features may have the same or similar reference numerals.

[0040] Figure 1 A schematic diagram of a condition monitoring system for petrochemical processes according to some embodiments of the present invention is shown.

[0041] Figure 2 A flowchart of a method for monitoring operating conditions in a petrochemical process, according to some embodiments of the present invention, is shown.

[0042] Figure 3 A schematic diagram of the monitoring model GRU-GMCVAE provided according to some embodiments of the present invention is shown;

[0043] Figure 4 A schematic diagram of the catalytic cracking process is shown;

[0044] Figure 5 A schematic diagram illustrating the monitoring results of energy consumption under global modeling using various neural network models provided according to some embodiments of the present invention is shown.

[0045] Figure 6 A schematic diagram is shown illustrating the monitoring results of the monitoring model GRU-GMCVAE provided according to some embodiments of the present invention for technical and economic indicators (TEI) and gasoline production (YG);

[0046] Figure 7 A comparative graph is shown showing the monitoring results of various neural network models and multivariate statistical models provided according to some embodiments of the present invention for technical and economic indicators (TEI) and gasoline production (YG) under global modeling and local modeling.

[0047] Figure 8 The process flow diagram of the Eastman process in Tennessee is shown; and

[0048] Figure 9 A graph showing monitoring results for failure mode 1 of the Tennessee Eastman process, provided according to some embodiments of the present invention, is shown.

[0049] Figure label:

[0050] 100: Operating condition monitoring systems used in petrochemical processes;

[0051] 110: Memory;

[0052] 111: Computer-readable storage medium;

[0053] 120: Processor;

[0054] 200: Methods for monitoring operating conditions in petrochemical processes; and

[0055] S210~S240: Steps. Detailed Implementation

[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be noted that the aspects described below with reference to the accompanying drawings and specific embodiments are merely exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0057] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0058] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood as the orientations shown in the relevant paragraphs and accompanying drawings. These relative terms are for illustrative purposes only and do not imply that the described apparatus must be manufactured or operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0059] It is understood that although terms such as "first," "second," and "third" may be used herein to describe various components, regions, layers, and / or parts, these components, regions, layers, and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers, and / or parts. Therefore, the first components, regions, layers, and / or parts discussed below may be referred to as second components, regions, layers, and / or parts without departing from some embodiments of the present invention.

[0060] As mentioned above, in the actual data modeling process of FCC, due to the long operating cycle and complex and variable operating conditions of the equipment, there are complex temporal correlations and nonlinear relationships among the data obtained from a large number of measurement points. This not only makes it difficult to finely classify the operating conditions of the data, but also increases the difficulty of anomaly detection and operating condition identification.

[0061] Furthermore, the data collected during the FCC process exhibits non-uniform time-delay dynamics. Non-uniform time-delay dynamics refer to the unpredictable random disturbances that occur in modern industrial production as raw materials are transformed into final products through multiple process units. Simultaneously, due to the varying reaction times at each process stage, the variables collected by different sensors exhibit varying degrees of autocorrelation. Autocorrelation, or the degree of correlation between a variable at different times, reflects the relationship between the variable's current value and its past values. If a variable exhibits strong autocorrelation, meaning high dynamics, and this characteristic is not considered when constructing the monitoring model, it will lead to a decline in the model's performance, thereby affecting the accuracy of process monitoring.

[0062] To overcome the aforementioned deficiencies in the existing technology, this invention provides a method for monitoring operating conditions in petrochemical industrial processes, a system for monitoring operating conditions in petrochemical industrial processes, and a computer-readable storage medium. These methods can effectively extract the dynamic characteristics of data from FCC processes, accurately learn the complex operating condition distribution in FCC processes, significantly improve the accuracy of operating condition monitoring and anomaly detection, and effectively reduce the false alarm rate.

[0063] In some non-limiting embodiments, the above-described method for monitoring operating conditions in petrochemical processes provided in the first aspect of the present invention can be implemented via the above-described system for monitoring operating conditions in petrochemical processes provided in the second aspect of the present invention.

[0064] Please refer to Figure 1 , Figure 1 A schematic diagram of a condition monitoring system for petrochemical processes, provided according to some embodiments of the present invention, is shown.

[0065] like Figure 1 As shown, the operating condition monitoring system 100 for petrochemical industrial processes may be configured with a memory 110 and a processor 120. The memory 110 includes, but is not limited to, the computer-readable storage medium 111 described in the third aspect of the present invention, which stores computer instructions thereon. The processor 120 is connected to the memory 110 and configured to execute the computer instructions stored in the memory 110 to implement the operating condition monitoring method for petrochemical industrial processes provided in the first aspect of the present invention.

[0066] The working principle of the above-mentioned operating condition monitoring system for petrochemical processes will first be described with reference to some embodiments of operating condition monitoring methods for petrochemical processes. Those skilled in the art will understand that these embodiments of operating condition monitoring methods for petrochemical processes are merely non-limiting implementations provided by the present invention, intended to clearly demonstrate the main concepts of the invention and provide specific solutions convenient for public implementation, rather than limiting all functions or operating methods of the operating condition monitoring system for petrochemical processes. Similarly, the operating condition monitoring system for petrochemical processes is also only one non-limiting implementation provided by the present invention, and does not constitute a limitation on the executing entity and execution order of the steps in these operating condition monitoring methods for petrochemical processes.

[0067] Please refer to Figure 2 , Figure 2 A flowchart of a method for monitoring operating conditions in a petrochemical process, according to some embodiments of the present invention, is shown.

[0068] like Figure 2 As shown, the operating condition monitoring method 200 for petrochemical industrial processes may include step S210: acquiring monitoring data and extracting time series data.

[0069] In some embodiments, the operating condition monitoring system for petrochemical industrial processes can first determine the variables for global modeling from the acquired monitoring data, extract key variables that are highly correlated with each indicator, and improve the robustness and accuracy of the monitoring model subsequently constructed.

[0070] Specifically, the operating condition monitoring system can identify key variables from the acquired monitoring data based on expert experience. Then, based on the monitored performance indicators, it uses the XGBOOST ensemble learning algorithm and mechanistic experience to determine the feature variables among the key variables as variables for global modeling.

[0071] Furthermore, the operating condition monitoring system can differentiate the operating conditions of the acquired monitoring data to achieve model partitioning under condition-driven variables. Condition-driven local modeling ensures that the constructed monitoring model can be optimized for different operating conditions, thereby improving the accuracy of fault detection.

[0072] Specifically, the operating condition monitoring system can distinguish the operating condition status of the acquired monitoring data based on cluster analysis methods, including K-means clustering.

[0073] In some embodiments, the operating condition monitoring system for petrochemical industrial processes can use the K-means clustering method to perform cluster analysis on the acquired monitoring data to distinguish different operating conditions and classify the monitoring data into the closest operating condition category to assist in the anomaly detection and process monitoring of the subsequent monitoring model.

[0074] Specifically, the given input data is X = {x1, x2, ..., x...} n The clustering result derived from the input data X is C. f ={c1,c2,……,c K} where n is the number of samples, K is the number of clusters, and f is the number of iterations. First, initialize the cluster centers. The operating condition monitoring system can randomly select a sample from the input data X as the initial cluster center μ1 for the first cluster.

[0075] The operating condition monitoring system can measure the μ value between each sample and the cluster center. p Similarity between them:

[0076]

[0077] Based on the above formula, the operating condition monitoring system can measure the similarity between each sample and the initial cluster center μ1. Based on the similarity value, a cluster c1 is formed centered on the initial cluster center μ1. The selection principle for the next cluster center is to choose the point that is furthest from the previous cluster center μ1; for example, the second cluster center can be the point with the largest distance from the initial cluster center μ1. This process continues until all cluster centers have been selected, and the set of cluster centers is determined as Φ. f ={μ1,μ2,……,μ K}

[0078] Then, based on each cluster, update the cluster centers of each cluster:

[0079]

[0080] Where, μ q For cluster c q The new cluster centers are determined, and corresponding clusters are formed. The set of cluster centers is then updated, and the next set of cluster centers is Φ. f+1 When the updated set of cluster centers Φ f+1 The set of cluster centers Φ before the update f When the change between them is less than ε, output the updated clustering result C. f+1 As a result of cluster analysis.

[0081] Based on the clustering results output by cluster analysis, the operating condition monitoring system can perform local modeling for different operating conditions.

[0082] Those skilled in the art will understand that the operating condition monitoring system can choose to use global modeling to filter characteristic variables of the data and use a unified model to process all operating condition data, or it can choose local modeling to construct multiple sub-models based on different operating condition data. Furthermore, it can use cluster analysis to classify the operating conditions based on the variables selected by global modeling. In the operating condition monitoring method for petrochemical industrial processes provided by this invention, a combination of global and local modeling is used to achieve local modeling of different operating conditions, improving the targeting and accuracy of the operating condition monitoring system.

[0083] Then, the operating condition monitoring system can extract time-series data. Preferably, the operating condition monitoring system can extract time-series data using a sliding window method.

[0084] Industrial processes typically exhibit time-series dependence, making it difficult for single-point monitoring to effectively capture the dynamic changes in chemical systems. By employing the sliding window method, the operating condition monitoring system can extract appropriate time series segments from the acquired monitoring data, enabling subsequent monitoring models to fully learn the dynamic characteristics of the chemical system and improve their ability to identify abnormal states.

[0085] Preferably, in the case of local modeling, the operating condition monitoring system can utilize a sliding window method to perform alignment preprocessing on monitoring data under different operating conditions to extract time series data under different operating conditions. The monitoring data under different operating conditions can be determined based on the updated clustering results output by the clustering analysis method in the above embodiments.

[0086] Please continue to refer to this. Figure 2 The operating condition monitoring system can execute step S220: inputting time series data into the monitoring model, which includes a gated cyclic unit network and a Gaussian mixture variational autoencoder with a Copula function.

[0087] Gated Recurrent Units (GRUs) are used to extract latent representations from time series data and determine the distribution parameters of latent variables. GRUs possess strong time series modeling capabilities and can effectively extract the dynamic characteristics of FCC processes.

[0088] The Gaussian Mixture Variational Autoencoder (GMCVAE), which incorporates a Copula function, is an extension of the Gaussian Mixture Variational Autoencoder (GMVAE). GMCVAE can more flexibly capture the distribution of latent variables in process conditions and, combined with a model partitioning strategy under conditional driving variables (i.e., using K-means clustering analysis to partition conditions), achieves adaptive classification of different operating states. Furthermore, due to the introduction of a Copula function into the GMCVAE structure, it can better characterize the prevalent relationships between dependent variables in industrial processes, further improving the model's ability to model the correlations between latent variables.

[0089] Compared to the traditional GMVAE method, the monitoring model GRU-GMCVAE proposed in this invention can more accurately learn the complex operating condition distribution in the FCC process, significantly improving the accuracy of operating condition monitoring and anomaly detection, while effectively reducing the false alarm rate.

[0090] Please refer to Figure 3 , Figure 3 A schematic diagram of the monitoring model GRU-GMCVAE provided according to some embodiments of the present invention is shown.

[0091] like Figure 3As shown, the monitoring model GRU-GMCVAE includes gated recurrent unit networks (GRU) in the encoder and decoder layers. In addition, the monitoring model GRU-GMCVAE also introduces Copula functions.

[0092] Gated recurrent unit networks (GRUs) can be designed to solve the vanishing gradient problem and improve the ability to model long sequence data.

[0093] In some embodiments, the gated recurrent unit network (GRU) can adopt a simpler structure, using only update gates and reset gates to control the information flow, thereby reducing computation and improving the training efficiency of the monitoring model GRU-GMCVAE training process.

[0094] Specifically, for a given time series data x = {x1, x2, ..., x...} T The gated recurrent unit network (GRU) can recursively compute the hidden state h. t :

[0095] h t =GRU(x t ,h t-1 ),

[0096] Among them, h t It is the hidden state at time t, x t The input data at time t, h t-1 It represents the hidden state at time t-1. GRU is a gated recurrent unit network model.

[0097] During the calculation at each time step t, the gated recurrent unit network (GRU) updates the hidden state h of the previous time step by controlling the gates. t-1 The retention rate at the current moment determines how much past information the model remembers:

[0098] z t =σ(W z x t +U z h t-1 +b z ),

[0099] Among them, z t It updates the gate value, σ is the activation function, and W... z It is the weight matrix, U z It is the weight matrix corresponding to the previous time step, b z It is the bias term in the update gate calculation. When the value z of the update gate... t As the value approaches 1, the gated recurrent unit network (GRU) retains more of the hidden state h from the previous time step. t-1 When updating the value z of the gatet When the value is close to 0, the gated recurrent unit network (GRU) tends to use the current input x more. t To calculate the new state.

[0100] Calculate candidate hidden states At that time, the gated loop unit network (GRU) resets the hidden state h of the gate control from the previous moment. t-1 To control the impact of information from the previous time step on the candidate hidden state:

[0101] r t =σ(W r x t +U r h t-1 +b r ),

[0102] Where, r t It resets the gate value, σ is the activation function, and W... r It is the weight matrix, U r It is the weight matrix corresponding to the previous time step, b r This is the bias term in the reset gate calculation. When the reset gate value r t When the value approaches 1, the gated cyclic unit network (GRU) allows the previously hidden state h to be hidden. t-1 Affects the current state; when the gate value r is reset t When the value approaches 0, the gated recurrent unit network (GRU) ignores the hidden state h from the previous time step. t-1 This makes the model tend to rely only on the current input x. t To calculate the new state.

[0103] Candidate hidden state at the current moment Calculation:

[0104]

[0105] in, It is the candidate hidden state at the current time step, tanh(·) is the hyperbolic tangent activation function, and W h and U h It is the weight matrix, b h This is the bias term used to calculate the candidate hidden state. When the gate value r is reset... t When it is close to 1, then the candidate state Simultaneously depends on the current input x t and the hidden state h from the previous moment t-1 When the value of the gate is reset r t When it is close to 0, then the candidate state Mainly determined by the current input x t Decide.

[0106] The final hidden state h at each time step t t By update gate z t Control determines the current candidate hidden state. Compared to the hidden state h in the previous moment t-1 Fusion:

[0107]

[0108] Where, when updating the value z of the gate t As the value approaches 1, the Gated Recurrent Unit (GRU) network relies more heavily on candidate states. When updating the value z of the gate t When the value is close to 0, more of the hidden state h from the previous moment is retained. t-1 Information.

[0109] like Figure 3 As shown, for a time series data x = {x1, x2, ..., x...} T The gated recurrent unit network (GRU) can recursively compute the hidden state h. t Encode the time series data information into the final hidden state h. T As a potential representation of time series data:

[0110] h T =GRU(x T ,h T-1 ),

[0111] Among them, h T-1 It is the final hidden state determined based on the time series data input at time T-1, and GRU is the gated recurrent unit network.

[0112] The final hidden state h T As a potential representation of the entire time series data, it contains information about historical time steps.

[0113] The hidden state h output by the gated recurrent unit network is transformed by linear transformation. T Mapped to the distribution parameters of the latent variables:

[0114] μ g ,σ g =f GRU (h T ),

[0115] Among them, f GRU (·) is a linear transformation, μ g σ is the mean that maps to the Gaussian mixture model. g The standard deviation is mapped to the Gaussian mixture model, where g represents different operating conditions.

[0116] The operating condition monitoring system can use the gated recurrent unit network (GRU) trained and optimized in the deployed monitoring model GRU-GMCVAE to obtain the latent representation of the extracted time series data and determine the distribution parameters of the latent variables.

[0117] The monitoring model GRU-GMCVAE also includes a Gaussian mixture variational autoencoder (GMCVAE) incorporating a copula function. The GMCVAE, with its copula function, can model latent variables based on their distribution parameters. The introduced copula structure can describe the nonlinear dependencies between latent variables.

[0118] Here, the distribution of the latent variables can be:

[0119]

[0120] Among them, Q φ This is the conditional probability distribution under encoder parameters φ, where z is the latent variable of the encoder output, D is the dimension of the latent variable, and C is the value of C. ψ It is a Copula function. The latent variable z is calculated using the Gaussian cumulative distribution function (CDF). d marginal distribution The mean is μ g variance is The Gaussian distribution.

[0121] Specifically, the operating condition monitoring system can use the Gaussian cumulative distribution function to transform and obtain the latent variable z. d marginal distribution

[0122]

[0123] Where Φ(·) is the cumulative distribution function (CDF) of the standard normal distribution.

[0124] Each latent variable z d Through its marginal cumulative distribution function Standardize the variables so that all variables can be mapped to the [0,1] interval, and then perform joint modeling using the Copula function:

[0125]

[0126] Furthermore, the operating condition monitoring system can optimize the distribution of latent variables in the GRU-GMCVAE monitoring model by maximizing the ELBO (Evidence Lower Bound) loss function. Specifically, during the optimization process, a Copula constraint term can be added to the ELBO loss function. That is, the ELBO loss function can include a reconstruction error term, a KL divergence term, and a Copula correlation modeling term, which helps improve the GRU-GMCVAE monitoring model's representation ability and anomaly identification accuracy in multi-operating condition dynamic processes. Here, the ELBO loss function... It can be:

[0127]

[0128] Wherein, the prior distribution P(z|g) refers to the distribution law of the latent variable z when the system is in the g-th operating condition; D KL (Q φ (z|x,g)||P(z|g)) is used to calculate the KL divergence between the posterior distribution containing Copula and the prior distribution P(z|g); in some embodiments, the prior distribution P(z|g) can be λ is a hyperparameter used to control the strength of the Copula constraint; P θ (x|z′,g) is the distribution of the reconstructed data generated during the decoding process. This distribution can also be modified by the Copula function to make the decoder output more consistent with the statistical dependency structure of the real data. Preferably, the distribution P of the generated reconstructed data is... θ (x|z′,g) can be:

[0129]

[0130] Among them, P θ Let be the conditional probability distribution under the decoder parameters θ, x be the original input data, z′ be the sample obtained from the posterior distribution of the latent variables, and μ be the intermediate variable used to connect the encoder and decoder. x and σ x These are the mean and standard deviation of the reconstructed data generated by the decoder based on z′ and the operating condition g, respectively.

[0131] Thus, the Gaussian mixture variational autoencoder (GMCVAE), by introducing the Copula function, not only considers the marginal distribution of each variable when reconstructing data, but also uses the Copula function for joint modeling to determine the joint distribution among variables, thereby improving the realism of the generated data. By introducing Copula, GMCVAE can better handle the nonlinear dependencies between variables when modeling complex data, making the latent space more consistent with the real data distribution, thereby enhancing the anomaly detection capability, generation capability, and interpretability of the monitoring model.

[0132] Preferably, the operating condition monitoring system can propose a loss function that comprehensively considers data reconstruction error, latent variable regularization, and inter-variable correlation as the loss function of the monitoring model. Specifically, the loss function can be:

[0133]

[0134] Where λ1 and λ2 are weight parameters, It is the data reconstruction error term. For the generated reconstructed data, D KL (Q φ (z|x)‖P(z)) is the KL divergence term, ∑ i,j log C ψ (z i ,z j ) is a Copula related term.

[0135] The first term in the loss function It can be used to measure the accuracy of data reconstruction, the second term D KL (Q φ (z|x)‖P(z)) is used to ensure the stability of the distribution of latent variables, the third term ∑ i,j log C ψ (z i ,z j Then, the effectiveness of anomaly detection can be improved by modeling the dependencies between variables.

[0136] Please continue to refer to this. Figure 2 The operating condition monitoring system can perform step S230: based on the monitoring model, calculate the monitoring statistics using the kernel density estimation method and set the control limits for the monitoring statistics.

[0137] In some embodiments, monitoring statistics may include the SPE statistic and the T statistic. 2 Statistics.

[0138] The formula for calculating the SPE statistic is:

[0139]

[0140] Where, xi Represents the original input data. This represents the generated reconstructed data, where m represents the dimension of the data.

[0141] The magnitude of the SPE statistic directly reflects the error level of the input data during the reconstruction process. The larger the error, the greater the deviation of the current data point from the normal pattern, thus indicating that an abnormal situation may have occurred during the operation.

[0142] The operating condition monitoring system can use kernel density estimation (KDE) to calculate the distribution of SPE (Special Purpose Parameter) and set control limits, thereby further improving the adaptability of the SPE statistic. The set control limits for the SPE statistic can be:

[0143]

[0144] in, Let σ be the mean of the SPE statistic on the training dataset. SPE Here, SPE represents the standard deviation of the SPE statistic, and k represents the set threshold parameter. When the SPE statistic of the monitoring data exceeds the set control limit of the SPE statistic, the monitoring data can be considered abnormal.

[0145] T 2 The T statistic is used to assess the degree of abnormality of latent variables. 2 The formula for calculating the statistic is:

[0146] T 2 =(z-μ) z ) T ∑ -1 (z-μ z ),

[0147] Where z is the latent variable output by the variational autoencoder, and μ z Let be the mean vector of the latent variables, and ∑ be the covariance matrix of the latent variables.

[0148] T 2 The calculation of the statistic depends on the distribution information of the latent variables, when the T of the monitored data... 2 The statistic exceeds the set T 2 The control limits of the statistics indicate that the projection location of the monitoring data in the potential space deviates from the normal distribution, and there may be anomalies.

[0149] T 2 The formula for calculating the control limits of a statistic is as follows:

[0150]

[0151] Where p is the dimension of the latent variable. The value is the 95th quantile of a chi-square distribution with p degrees of freedom.

[0152] Please continue to refer to this. Figure 2 The operating condition monitoring system can execute step S240: perform anomaly detection based on the monitoring statistics and the control limits of the monitoring statistics.

[0153] In the above embodiments, the operating condition monitoring system can combine SPE statistics and T 2 The logic for anomaly detection using statistical measures can be as follows:

[0154] Response to T 2 The statistic is less than or equal to T 2 The control limits of the SPE statistic are less than or equal to the control limits of the SPE statistic, that is, when And SPE≤SPE limit At that time, the output data detection result is in a normal state;

[0155] Response to T 2 The statistic is less than or equal to T 2 The control limits of the SPE statistic and the SPE statistic is greater than the control limits of the SPE statistic, that is, when And SPE > SPE limit When the output data detection result is the first abnormal state; or

[0156] Response to T 2 The statistic is greater than T 2 Control limits for a statistic, i.e., when At that time, the output data detection result is the second abnormal state.

[0157] Here, the degree of abnormality in the first abnormal state is less than that in the second abnormal state. When the data detection result is in the first abnormal state, the data can be considered to be in a mild abnormal state, which will not significantly affect the variables of the chemical system and is within an acceptable range. When the data detection result is in the second abnormal state, it is a serious abnormality that requires immediate attention from technical personnel.

[0158] The following are some non-limiting embodiments and comparative examples, which illustrate the feasibility and effectiveness of the proposed method for monitoring operating conditions in petrochemical industrial processes.

[0159] Please refer to Figure 4 , Figure 4 A schematic diagram of the catalytic cracking process is shown.

[0160] like Figure 4 As shown, the catalytic cracking process can include a reaction-regeneration process, a fractionation process, and an absorption-stabilization process.

[0161] The enlarged view on the left is a partial view of abnormal operating conditions of flue gas emission of pollutants (FEP) in the catalytic cracking unit. When the regulator PDICA105 is in positive-acting mode, its output is usually large. Therefore, the output value of the temperature regulator TICA101 is generally small. The low selector ORST101 mainly controls the regeneration slide valve SV101 based on the output value of the temperature regulator TICA101 to maintain appropriate temperature control. When the riser temperature decreases, the output of the temperature regulator TICA101 increases, the opening of the regeneration slide valve SV101 increases, the catalyst circulation rate increases, and the temperature recovers. The adjustment of the opening of the regeneration slide valve SV101 has a significant impact on energy consumption (EC). Furthermore, the control logic of the regeneration slide valve SV101 is closely related to the temperature control and the flow rate of the regenerator C102 shown in the enlarged diagram on the left. This affects the overall thermal energy utilization efficiency in the catalytic cracking unit and the fuel consumption of the regenerator C102. An imbalance in the ratio of catalyst to fuel in the regenerator C102 may lead to incomplete combustion or over-combustion, causing the flue gas pollutant emission FEP to deviate from the normal range, thus forming an abnormal operating condition of the flue gas pollutant emission FEP.

[0162] The enlarged view on the right is a partial diagram of abnormal operating conditions in the catalytic cracking unit's energy consumption (EC). Data sampling for the operating condition monitoring system used in petrochemical processes can be performed at 10-minute intervals. When abnormal fluctuations occur in the reaction temperature output by the temperature controller TICA101, the excessive opening of the regeneration slide valve SV101 leads to a decrease in the valve differential pressure, causing a rapid reduction in the output of the controller PDICA105, thus affecting energy consumption (EC). In this situation, timely adjustments to the regeneration slide valve SV101 can restore the catalytic cracking unit to normal operating conditions.

[0163] Figure 4 In the illustrated embodiment, the catalytic cracking unit data used comes from a catalytic cracking unit with an annual processing capacity of 1.8 million tons in a certain enterprise. The collected data covers time-series process data and experimental analysis system data during the entire operation of the catalytic cracking unit in 2024. After expert screening, 48 representative operating variables and performance indicators were identified from the time-series process data and experimental analysis system data. Furthermore, the raw data needs to be preprocessed before constructing and optimizing the monitoring model. This avoids problems such as decreased model fitting accuracy, underfitting, or overfitting caused by a large number of variables and strong correlations between them. By performing necessary screening and processing on the data before modeling, key variables highly correlated with each indicator are extracted, improving the robustness and accuracy of the monitoring model.

[0164] In the FCC process, four important performance indicators are: energy consumption (EC), flue gas emissions (FEP), technical and economic indicator (TEI), and yield of gas (YG). Data for EC, FEP, TEI, and YG can all be obtained directly from the plant. The formula for calculating the technical and economic indicator (TEI) ξ is as follows:

[0165] ξ=(1.2×Y G +Y D +Y LPG )×(ρ in / 1000)×100 / (100-W CCR ×0.8)+η / 20,

[0166] In the formula, Y G Y D Y LPG ρ represents the yield of gasoline, diesel, and liquefied petroleum gas, respectively. in W CCR η and η represent the raw material density, residual carbon content, and slag blending ratio, respectively.

[0167] If a load condition monitoring system directly uses all variables for unified modeling, the monitoring accuracy may decrease due to variable redundancy and complex variable coupling relationships. To improve the targeting and accuracy of the load condition monitoring system, it can select the most relevant variables for each performance indicator for modeling, thereby avoiding the influence of redundancy and building a more accurate monitoring model.

[0168] Specifically, the operating condition monitoring system can use the XGBOOST ensemble learning algorithm and combine it with the plant's mechanistic experience to select the most representative feature variables from the 48 representative operational variables and performance indicators. These feature variables are then selected based on four performance indicators to be used in the construction of the subsequent monitoring model, thereby improving the monitoring capability and prediction accuracy of the monitoring model.

[0169] The characteristic variables selected based on energy consumption (EC) are shown in Table 1.

[0170]

[0171] The characteristic variables selected based on flue gas pollutant emissions (FEP) in Table 1 are shown in Table 2.

[0172]

[0173] The characteristic variables selected based on the technical and economic indicators (TEI) in Table 2 can be shown in Table 3.

[0174]

[0175] The characteristic variables selected based on gasoline production (YG) in Table 3 can be shown in Table 4.

[0176]

[0177] Table 4

[0178] In industrial process monitoring, different operating conditions may correspond to different operating modes, and directly applying a uniform fault monitoring model may lead to false alarms. Therefore, the operating condition monitoring system can use K-means clustering analysis to classify the acquired monitoring data according to operating conditions, ensuring that the monitoring model GRU-GMCVAE can be optimized for different operating conditions, thereby improving the accuracy of fault detection.

[0179] Please refer to Figure 5 , Figure 5 The diagram illustrates the monitoring results of various neural network models provided according to some embodiments of the present invention for energy consumption under global modeling.

[0180] like Figure 5 As shown, when the statistic exceeds the control limit, it indicates that there is an anomaly in the sampled data points. Under global modeling, the monitoring model GRU-GMCVAE can detect the anomaly the fastest (at the 103rd sampling point), with a response speed faster than other neural network models.

[0181] In this embodiment, the operating condition monitoring system can use the Fault Detection Rate (FDR) and False Alarm Rate (FAR) as evaluation metrics to assess the performance of the monitoring model. The definitions of FDR and FAR are as follows:

[0182]

[0183] Where tp refers to the number of fault samples correctly identified as faults by the monitoring model, and fp refers to the number of normal samples misidentified as faults by the monitoring model.

[0184] Tables 5 and 6 comprehensively demonstrate the fault monitoring capabilities and false alarm control effects of various neural network models and multivariate statistical models under global modeling and further local modeling achieved through condition-driven (i.e., using K-means clustering analysis to classify operating conditions) approaches for energy consumption (EC) and flue gas pollutant emissions (FEP).

[0185]

[0186] Table 5

[0187] As can be seen from Table 5, the monitoring model GRU-GMCVAE in the operating condition monitoring method proposed in this invention has the lowest false alarm rate (FAR) in the monitoring of energy consumption (EC). 2 Both the statistical value and the SPE statistical value are 0%. However, in the operational condition monitoring of flue gas pollutant emissions (FEP), the monitoring model GRU-GMCVAE has the highest failure detection rate (FDR) (T). 2 With a statistical rate of 90.00% and an SPE statistical rate of 86.67%, it performs best among various neural network models. In comparison, other neural network models have varying degrees of shortcomings. For example, GRU-VAE (Gated Recurrent Unit-Variational Autoencoder) and LSTM-VAE (Long Short-Term Memory Network-Variational Autoencoder) perform poorly in both global and local modeling, exhibiting high false alarm rates in monitoring operating conditions for energy consumption (EC) and poor accuracy in monitoring operating conditions for flue gas pollutant emissions (FEP).

[0188]

[0189] Table 6

[0190] As can be seen from Table 6, the multivariate statistical model also exhibits significant limitations. Traditional PCA (Principal Component Analysis) and ICA (Independent Component Analysis) methods show a false alarm rate (FAR) exceeding 50% under global modeling, despite some statistical indicators such as... While the false alarm rate (FAR) decreased in SFA (Slow Feature Analysis) and ESFA (Enhanced Slow Feature Analysis) methods, the fault detection rate (FDR) remained lower than that of the monitoring model GRU-GMCVAE in monitoring flue gas pollutant emissions (FEP), particularly in the identification of complex nonlinear operating conditions, where the performance of multivariate statistical models was significantly insufficient. Even with further local modeling driven by K-means conditions, multivariate statistical models such as ICA, ESFA, and IESFA (Improved Enhanced Slow Feature Analysis) achieved improved fault detection rates (FDR) (some indicators reaching over 70%), but their control over the false alarm rate (FAR) still lagged behind that of the monitoring model GRU-GMCVAE. For example, in the FAR of ICA, T... 2 Both the statistical value and the SPE statistical value are higher than 65%, indicating that the ICA model has poor adaptability under nonlinear conditions.

[0191] Additionally, please refer to Figure 6 , Figure 6 A schematic diagram is shown illustrating the monitoring results of the monitoring model GRU-GMCVAE provided according to some embodiments of the present invention for technical and economic indicators (TEI) and gasoline production (YG).

[0192] Figure 6 The left-hand figure shows the monitoring results of the GRU-GMCVAE monitoring model for the Technical and Economic Indicator (TEI). Figure 6 The right-hand figure shows the monitoring results of the GRU-GMCVAE monitoring model for gasoline production (YG). From... Figure 6 It can be seen that the monitoring model GRU-GMCVAE performs well.

[0193] exist Figure 6 In the left image, T 2 The statistical index showed significant fluctuations at the 36th sample point, indicating an anomaly in the catalytic cracking unit system at that moment, which the monitoring model GRU-GMCVAE successfully detected. During subsequent operation, T... 2 The statistics gradually stabilized, indicating that the system has adapted to the new operating conditions. Meanwhile, an anomaly was detected in the SPE statistic at the 36th sample point, and the subsequent SPE statistic continued to fluctuate to some extent, demonstrating that the monitoring model GRU-GMCVAE is highly sensitive to anomalies and can continuously monitor changes in operating conditions.

[0194] exist Figure 6 In the image on the right, T 2 The statistic showed a significant abrupt change at 185 sample points. The monitoring model GRU-GMCVAE was able to respond promptly and detect this anomaly, followed by T... 2 The statistic quickly fell back below the control limit, indicating that the operating conditions had returned to normal. Furthermore, the SPE statistic also identified an anomaly at the 184th sample point and recovered to near the control limit within a short time, validating the monitoring model GRU-GMCVAE's ability to monitor sudden changes in operating conditions.

[0195] Further, please refer to Figure 7 , Figure 7 The diagram shows a comparison of the monitoring results of various neural network models and multivariate statistical models provided according to some embodiments of the present invention for technical and economic indicators (TEI) and gasoline production (YG) under global modeling and local modeling.

[0196] like Figure 7 As shown, Figure 7 (a) A comparison chart shows the monitoring results of various neural network models for the Technical and Economic Indicator (TEI) under global and local modeling. It can be seen that the false alarm rate (FAR) of the global GRU-GMCVAE and global LSTM-GMCVAE performs best under global modeling. Under local modeling, the false alarm rate (FAR) of the local GRU-GMCVAE is better than that of the local LSTM-GMCVAE. Figure 7 (b) A comparison chart shows the monitoring results of the multivariate statistical model for the technical and economic indicators (TEI) under global and local modeling. It can be seen that the false alarm rate (FAR) of the global GRU-GMCVAE and the local GRU-GMCVAE is much lower than that of other statistical models, demonstrating superior performance and the ability to accurately identify anomalies under complex operating conditions.

[0197] Figure 7 (c) A comparison chart shows the monitoring results of various neural network models for gasoline production (YG) under global and local modeling. It can be seen that the fault detection rate (FDR) of the global GRU-GMCVAE is much higher than that of the global LSTM-GMCVAE, indicating that while the global LSTM-GMCVAE performs well in monitoring technical and economic indicators (TEI), it performs poorly in monitoring gasoline production (YG). Furthermore, the fault detection rate (FDR) of the global GRU-GMCVAE is also significantly higher than that of other neural network models under global modeling. Figure 7(d) shows a comparison of the monitoring results of the multivariate statistical model for gasoline production (YG) under global and local modeling. It can be seen that the fault detection rate (FDR) of the global GRU-GMCVAE is significantly higher than that of other statistical models under global modeling, and the FDR of the local GRU-GMCVAE is also significantly higher than that of other statistical models under local modeling. This indicates that the GRU-GMCVAE has stronger anomaly identification capabilities and monitoring robustness during the FCC process, further validating the superiority of the monitoring model.

[0198] Please refer to Figure 8 , Figure 8 The process flow diagram of the Eastman process in Tennessee is shown.

[0199] The operational monitoring method for petrochemical processes proposed in this invention can also be applied to the Tennessee Eastman process. The Tennessee Eastman (TE) process is a complex dynamic process widely used to simulate real chemical production processes to support research in the field of process monitoring. The control system of the Tennessee Eastman process mainly consists of key equipment such as reactors, product condensers, gas-liquid separators, circulating compressors, and stripping towers, comprehensively reflecting common physical changes and chemical reaction processes in industrial production.

[0200] The TE unit can simulate various complex chemical reaction processes and process mixed raw materials with multiple chemical components. The mixed raw materials are initially heated in a preheater through heat exchange with the reaction products, and then mixed with specific reaction gases in the reactor. Under the high temperature and pressure of the reactor and in the environment created by the catalyst, a series of complex chemical reactions occur, while simultaneously removing some impurities contained in the raw materials. The resulting product mixture then passes through a series of separation and fractionation systems, including a condenser, compressor, separator, and stripper, to produce various chemical products. Figure 8 The diagram shows the products G and H generated after the feeds A and C react with reactants D and E, respectively. The control system of this Tennessee Eastman process mainly consists of five process units: a two-phase reactor where the exothermic reaction occurs, a separator, a stripping tower, a compressor, and a condenser. In the diagram, FI is the flow indicator, PI is the pressure indicator, LI is the level indicator, and TI is the temperature indicator.

[0201] The Tennessee Eastman process encompasses 12 manipulated variables, 22 process measurement variables, and 19 component variables. To further achieve local modeling, the TE process defines 21 failure modes, each representing a different type of process anomaly, thereby enabling in-depth analysis and evaluation of the performance of the monitoring model GRU-GMCVAE.

[0202] In the experimental dataset, each failure scenario for the Tennessee Eastman process contains 960 sample data points. In each failure scenario, the failure event is introduced starting from the 160th sample point to simulate sudden abnormal situations that may occur in industrial production.

[0203] In this embodiment, to evaluate the generalization ability and adaptability of the monitoring model GRU-GMCVAE, fault mode 1 and fault mode 13 were selected as specific research cases. Fault mode 1 describes an abnormal change in reactor cooling water temperature, while fault mode 13 describes an abnormal change in condenser cooling water flow rate. These two faults involve abnormal temperature control and cooling system performance, which are key issues in the thermal management process of chemical production.

[0204] By analyzing the data of fault mode 1 and fault mode 13, the operating condition monitoring system can systematically evaluate the fault detection performance of the operating condition monitoring method provided by the present invention under abnormal conditions of different natures of the process, and further verify the practical application value and robustness of the operating condition monitoring method.

[0205] Please refer to Figure 9 , Figure 9 A graph showing monitoring results for failure mode 1 of the Tennessee Eastman process, provided according to some embodiments of the present invention, is shown.

[0206] like Figure 9 As shown, the monitoring model GRU-GMCVAE can observe T at sampling points 163-164. 2 The statistics and SPE statistics exceeded the control limits, indicating that a fault was detected. This demonstrates that the monitoring model GRU-GMCVAE has excellent monitoring performance.

[0207] In summary, the operating condition monitoring method for petrochemical industrial processes provided by this invention combines the advantages of GRU networks (strong sensitivity to short-term fluctuations in sequences and more flexible gating mechanisms) with GMVAE's superior fitting ability to complex data distributions. Furthermore, the introduction of a Copula structure more effectively captures the nonlinear dependencies between multiple variables, making it outstanding in complex operating condition switching scenarios. It can detect and adapt to changes in operating conditions faster and more accurately. This results in a operating condition monitoring method with not only superior false alarm control and a higher fault identification rate, but also significantly faster anomaly response speed. It exhibits exceptional performance, particularly in handling complex, multivariate, and nonlinear sequence data, fully validating the practicality and robustness of the GRU-GMCVAE monitoring model structure in fault detection scenarios.

[0208] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0209] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different techniques and arts. For example, the data, instructions, commands, information, signals, bits, symbols, and chips described throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, light fields or optical particles, or any combination thereof.

[0210] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0211] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0212] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0213] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0214] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring operating conditions in petrochemical industrial processes, characterized in that, Including the following steps: Acquire monitoring data and extract time series data; The time series data is input into a monitoring model, which includes a gated recurrent unit network and a Gaussian mixture variational autoencoder with a Copula function. The gated recurrent unit network is used to extract the latent representation of the time series data and determine the distribution parameters of the latent variables. The Gaussian mixture variational autoencoder with a Copula function models the latent variables based on the distribution parameters of the latent variables. Based on the monitoring model, the monitoring statistic is calculated using the kernel density estimation method, and control limits for the monitoring statistic are set; and Anomaly detection is performed based on the monitoring statistics and the control limits of the monitoring statistics.

2. The operating condition monitoring method as described in claim 1, characterized in that, The steps of acquiring monitoring data and extracting time series data further include: Based on cluster analysis methods, the operating conditions of the acquired monitoring data are distinguished. These cluster analysis methods include K-means clustering. Using the sliding window method, the monitoring data under different operating conditions are aligned and preprocessed to extract time series data under different operating conditions.

3. The operating condition monitoring method as described in claim 1, characterized in that, The steps of the gated recurrent unit network to extract the latent representation of the time series data and determine the distribution parameters of the latent variables include: The time series data is x = {x1, x2, ..., x} T The gated recurrent unit network encodes the time series data information into the final hidden state h by recursively calculating the hidden state. T As a potential representation of the time series data: h T =GRU(x T ,h T-1 ), Among them, h T-1 It is the final hidden state determined based on the time series data input at time T-1, and GRU is the gated recurrent unit network; Through linear transformation, the hidden state hT output by the gated recurrent unit network is mapped to the distribution parameters of the latent variables: m g ,s g =f GRU (h T ), Among them, f GRU (·) is a linear transformation, μ g σ is the mean that maps to the Gaussian mixture model. g The standard deviation is mapped to the Gaussian mixture model, where g represents different operating conditions.

4. The operating condition monitoring method as described in claim 3, characterized in that, The latent variable modeling includes the following steps: The distribution of the latent variables is as follows: Among them, Q φ It is the conditional probability distribution under encoder parameters φ, z is the latent variable output by the encoder, D is the dimension of the latent variable, and C is the conditional probability distribution under encoder parameters φ. ψ It is the Copula function, The latent variable z is calculated from the Gaussian cumulative distribution function. i marginal distribution The mean is μ g variance is The Gaussian distribution.

5. The operating condition monitoring method as described in claim 1, characterized in that, The loss function of the monitoring model takes into account data reconstruction error, latent variable regularization, and correlation between variables.

6. The operating condition monitoring method as described in claim 5, characterized in that, The loss function is: Where λ1 and λ2 are weight parameters, It is the data reconstruction error term, D KL (Q φ (z|x)‖P(z)) is the KL divergence term, ∑ i,j log C ψ (z i ,z j ) is a Copula related term.

7. The operating condition monitoring method as described in claim 1, characterized in that, The monitoring statistics include SPE statistics and T. 2 The SPE statistic is calculated using the following formula: Where, x i Represents the original input data. This represents the generated reconstructed data, where m represents the dimension of the data. The T 2 The formula for calculating the statistic is: T 2 =(z-μ z ) T ∑ -1 (z-μ z ), Where z is the latent variable output by the variational autoencoder, and μ z Let be the mean vector of the latent variables, and ∑ be the covariance matrix of the latent variables.

8. The operating condition monitoring method as described in claim 7, characterized in that, The formula for calculating the control limits of the SPE statistic is as follows: in, Let σ be the mean of the SPE statistic on the training dataset. SPE The standard deviation of the SPE statistic is given by , and k represents the set threshold parameter. The T 2 The formula for calculating the control limits of a statistic is: Where p is the dimension of the latent variable. The value is the 95th quantile of a chi-square distribution with p degrees of freedom.

9. The operating condition monitoring method as described in claim 8, characterized in that, The step of detecting anomalies based on the monitoring statistics and the control limits of the monitoring statistics includes: In response to the T 2 The statistic is less than or equal to the stated T 2 If the SPE statistic is less than or equal to the control limit of the SPE statistic, the data detection result is considered normal. In response to the T 2 The statistic is less than or equal to the stated T 2 If the SPE statistic exceeds the control limit of the SPE statistic, the data detection result is a first abnormal state; or In response to the T 2 The statistic is greater than the stated T 2 The control limits of the statistics are defined as follows: the data detection result is a second abnormal state, and the degree of abnormality of the first abnormal state is less than that of the second abnormal state.

10. A condition monitoring system for petrochemical industrial processes, characterized in that, include: Memory, on which computer instructions are stored; as well as A processor, connected to the memory, and configured to execute computer instructions stored in the memory to implement the operating condition monitoring method for petrochemical industrial processes as described in any one of claims 1 to 9.

11. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, the operating condition monitoring method for petrochemical industrial processes as described in any one of claims 1 to 9 is implemented.

Images