Chemical process operation fault diagnosis method and system

Abstract

The invention relates to a fault diagnosis method and system for chemical process operation. The system comprises a space-time distribution matrix generation module, a collaborative anomaly detection module, a dynamic incidence matrix and network mapping module, an anomaly propagation path identification module and a root cause repair measure module. According to the scheme, a dynamic and intelligent fault diagnosis system is constructed, and the stability and the fault processing efficiency of the chemical process are remarkably improved. According to the method, on the basis of a space-time distribution matrix, reaction path data and multiphase flow data are uniformly mapped, dynamic fusion of variables in time and space is achieved, and the running state in the reactor is comprehensively reflected. Precise identification is realized through collaborative anomaly and isolated anomaly classification analysis, and the source and the property of the anomaly are locked in combination with space-time matching of reaction path data and multiphase flow data. Through tracking of a propagation path and an abnormal node sequence, an abnormal evolution process is clearly displayed, and a foundation is laid for formulating effective repair measures.

Description

Technical Field

[0001] The present invention relates to the field of chemical process fault supervision, and specifically to a fault diagnosis method and system for chemical process operation. Background Art

[0002] Chemical fault diagnosis technology is a technology for detecting, identifying, analyzing, and locating possible equipment or process abnormalities in chemical production processes, aiming to quickly and accurately find the root causes of faults and propose effective repair measures to ensure the safety, stability, and efficiency of chemical production. This field is an important part of chemical process safety management and plays an important role in preventing safety accidents, reducing economic losses, and improving production efficiency.

[0003] Chemical fault diagnosis technology involves complex multivariable relationships, multi-scale dynamic changes, and the non-linear characteristics of chemical systems. Common fault types include equipment failures (such as the failure of pumps, valves, and heat exchangers), process abnormalities (such as temperature, pressure, and flow rate deviating from the normal range), and control system failures (such as sensor malfunctions or controller abnormalities). Faults may be caused by equipment aging, operation errors, external interference, or process design defects. The difficulty in diagnosis lies in quickly identifying the source of the problem and determining its propagation and impact.

[0004] Due to the complexity of chemical processes involving multi-dimensional data (such as reaction paths, flow patterns, flow rates, etc.) and dynamic changes, existing methods often lack the ability to comprehensively integrate time and space information and are difficult to truly reflect the overall dynamic characteristics of system operation. Moreover, traditional methods usually cannot effectively distinguish between systematic abnormalities and local abnormalities, making the source location of the problem ambiguous and prone to misjudgment or missed judgment. In addition, the ability to analyze the correlation relationships between variables is weak, unable to quantify the abnormal propagation process, and unable to identify the importance of key nodes in the propagation chain, resulting in difficulty in sorting out the abnormal diffusion and evolution mechanism. More seriously, existing technologies mostly rely on local information or single indicators, and this single way is prone to ignoring key factors in complex systems, resulting in inaccurate or even missed root cause identification. Summary of the Invention

[0005] The present invention provides a fault diagnosis method and system for chemical process operation in view of the technical problems existing in the prior art.

[0006] The technical solution of the present invention to solve the above technical problems is as follows:

[0007] A fault diagnosis method for chemical process operation, characterized in that the method includes:

[0008] Monitoring reaction path data, and collecting multiphase flow data in real time. Combining the spatial distribution of the reactor, mapping the reaction path data and the multiphase flow data into the same spatial coordinate system to generate a space-time distribution matrix;

[0009] Perform collaborative anomaly detection using reaction path data and multiphase flow data, compare with the expected values, record the anomaly time and the position on the reaction path, monitor multiphase flow data anomalies simultaneously, match the data at the anomaly position on the reaction path with the multiphase flow anomaly data in space and time, and mark according to the matching results;

[0010] Associate the correlation strength between every two data in the space-time distribution matrix, generate a time-varying correlation matrix in each time window to form a sequence of dynamic correlation matrices, and map it into a dynamic correlation network;

[0011] Identify the nodes with marks, track the propagation path of the anomaly of the node, identify the starting point of the anomaly path and record the anomaly evolution order, calculate the anomaly propagation centrality of each node in the dynamic correlation network simultaneously, and identify the root cause nodes of the marks according to the anomaly propagation centrality;

[0012] According to the root cause nodes in the analyzed anomaly propagation path, retrieve corresponding repair measures for different root causes, generate control strategies matching the repair measures for the non-root cause nodes in the anomaly propagation path, and re-evaluate the dynamic correlation network after the repair measures and control strategies are executed to determine whether the anomaly is eliminated.

[0013] As a further solution of the present invention, monitor the reaction path data, collect multiphase flow data in real time, combine with the spatial distribution of the reactor, map the reaction path data and multiphase flow data into the same space coordinate system to generate a space-time distribution matrix, specifically including:

[0014] Monitor the reaction path data in the reactor in real time, including: the concentrations of reactants and intermediate products;

[0015] Collect multiphase flow data in real time, including: the flow pattern distribution of the multiphase flow in the reactor, the flow velocity data of the multiphase flow in the reactor, and the real-time data of the gas, liquid, and solid three-phase distribution;

[0016] Establish the space coordinate system of the reactor, divide the reactor into multiple grid units, each unit is used as the basic unit of the unified coordinate, and map the positions of the acquisition sensors into the space coordinate system to define the spatial positions of each sampling point;

[0017] According to the dynamic characteristics of the reaction process, divide the entire operation time into several time windows, take the average value of the data values in each time window, and organize them into a time series in the space coordinate system;

[0018] Integrate the time-serialized data into a space-time distribution matrix.

[0019] As a further solution of the present invention, the integration of the time - serialized data into a spatio - temporal distribution matrix is specifically as follows:

[0020] Synthesize the spatio - temporal distribution matrix X ij (t k ), where i represents the spatial coordinates of the reactor, j represents the real - time collected data, and k is the time dimension, which is directly related to the time point:

[0021]

[0022] In the spatio - temporal distribution matrix X ijk (t):

[0023] The first dimension is the three - dimensional spatial distribution of the reactor, the second dimension is the data set, and the third dimension is the time series.

[0024] As a further solution of the present invention, the matching of the data at the abnormal positions of the reaction path and the multiphase flow abnormal data in space and time, and the marking according to the matching result specifically include:

[0025] Calculate the concentration expectation values, normal flow pattern distributions, normal flow velocity data, and typical distribution ranges of reactants and intermediate products under normal operating conditions;

[0026] Dynamically update the concentration expectation values, normal flow pattern distributions, normal flow velocity data, and typical flow pattern distributions in the normal state in the form of a sliding time window, and calculate the expectation value interval of the variable;

[0027] Compare the concentrations of reactants and intermediate products and multiphase flow data collected in real time with the expectation value interval. Within the sliding time window, calculate the moving average of the deviation. If the moving average continuously exceeds the expectation value interval, it is determined as a persistent abnormality, and the detected abnormal points are recorded in terms of time and space, including the time point of the abnormality occurrence and the abnormal spatial position on the reaction path;

[0028] Match the abnormal spatial position of the reaction path with the multiphase flow abnormal data, and compare their spatial positions and time dimensions;

[0029] If the abnormal positions of the reaction path and the multiphase flow coincide in space and time, it is marked as a collaborative abnormality, otherwise it is marked as an isolated abnormality.

[0030] As a further solution of the present invention, generating an association matrix that changes with time in each time window, forming a dynamic association matrix sequence, and mapping it into a dynamic association network specifically includes:

[0031] Select from the spatio - temporal distribution matrix X ij (t k)Extract the real-time collected data j, where j = {j 1 , j 2 , j 3 ,..., j m};

[0032] Calculate the correlation strength between the data j at the spatial position i at different time points t k , and generate a correlation matrix R(t x , j y ), and generate a correlation matrix R(t k );

[0033] Generate a sequence of dynamic correlation matrices that change over time within the entire time range;

[0034] Map the sequence of dynamic correlation matrices to a dynamic correlation network.

[0035] As a further solution of the present invention, the calculation of the correlation strength between the data j at the spatial position i at different time points t k , and generate a correlation matrix, specifically: x , j y , and generate a correlation matrix, specifically:

[0036]

[0037] where is the correlation strength value, is the covariance of the data j x and j y , and are the standard deviations of the data j x and j y respectively;

[0038] Construct the correlation matrix R(t k ) at the time point t k to represent the correlation strength between different data:

[0039]

[0040] The generation of the sequence of dynamic correlation matrices that change over time within the entire time range is specifically:

[0041] Combine the correlation matrices at all time points into a sequence of dynamic correlation matrices:

[0042] {R(t 1 ), R(t 2 ),..., R(t K )}, where K is the total number of time points;

[0043] Use a sliding time window for the matrix elements Perform smoothing:

[0044]

[0045] where n is the width of the sliding window.

[0046] As a further solution of the present invention, mapping the dynamic association matrix sequence into a dynamic association network specifically includes:

[0047] Node set of the dynamic association network:

[0048] V = {v 1 , v 2 , v 3 ,..., v m}, where each node v j corresponds to a data j;

[0049] Define the edge weights through the in the association matrix:

[0050]

[0051] And set a threshold τ, and only when keep the corresponding edge;

[0052] For each time point t k , generate a static network G(t k ) = (V, E(t k )), where E(t k ) is the edge set at time point t k ;

[0053] Combine the static networks generated at all time points into a dynamic network sequence:

[0054] {G(t 1 ), G(t 2 ), G(t 3 ),..., G(t K )}.

[0055] As a further solution of the present invention, identifying the starting point of the abnormal path and recording the abnormal evolution order, and at the same time calculating the abnormal propagation centrality of each node in the dynamic association network, and identifying and marking the root cause node according to the abnormal propagation centrality specifically includes:

[0056] Obtain the abnormal data with co-abnormal or isolated abnormal marks and their corresponding time points, and mark the data nodes where the abnormal data is located as abnormal nodes in the dynamic association network;

[0057] Starting from each abnormal node, read the association strength between the abnormal node and all surrounding nodes, and search along the edge with the highest association strength, recording the path order and propagation chain of the nodes;

[0058] For each abnormal node, determine all the nodes on its propagation path, and generate an abnormal propagation sequence of the nodes:

[0059]

[0060] Analyze each propagation path, use the timestamp and association strength information to identify the earliest abnormal node on the propagation chain, record it as the starting point of the path, and generate a set of starting abnormal nodes;

[0061] Calculate the abnormal propagation centrality of each node in the dynamic association network, and sort the nodes according to the abnormal propagation centrality to identify the node with the highest centrality;

[0062] Integrate the starting point of the propagation path and the propagation centrality, and set a centrality threshold to determine the root cause node. If the centrality of the starting point node is higher than the centrality threshold, directly mark it as the root cause node; otherwise, select the node with the highest centrality in the propagation path as the root cause node.

[0063] As a further solution of the present invention, the calculation of the abnormal propagation centrality of each node in the dynamic association network is specifically as follows:

[0064] For each time point t k , calculate the centrality of all nodes in the current network G(t k ), including:

[0065] Degree centrality C D (v jx ):

[0066]

[0067] Among them, deg(v) is the number of edges connected to the node, and N is the total number of nodes in the network;

[0068] Betweenness centrality C B (v jx ):

[0069]

[0070] Is the number of paths passing through node v jy to node v jz in the shortest path from node v jx , Is the number of paths from node v jy to node v jzThe total number of shortest paths;

[0071] Closeness centrality C C (v jx ):

[0072]

[0073] d(v jx , v jz ) is the length of the shortest path from node v jx to node v jz ;

[0074] Calculate the comprehensive abnormal propagation centrality C S (v jx ):

[0075] C S (v jx ) = β 1 C D (v jx ) + β 2 C B (v jx ) + β 3 C C (v jx );

[0076] Among them, β 1 , β 2 , β 3 are weight coefficients;

[0077] Sort all nodes according to the comprehensive abnormal propagation centrality to obtain the node importance ranking.

[0078] Another object of the present invention is to provide a fault diagnosis system for chemical process operation, and the system includes:

[0079] A space-time distribution matrix generation module, which is used to monitor reaction path data, collect multiphase flow data in real time, and combine the reactor space distribution to map the reaction path data and multiphase flow data into the same space coordinate system to generate a space-time distribution matrix;

[0080] A collaborative anomaly detection module, which is used to perform collaborative anomaly detection using reaction path data and multiphase flow data, compare with the expected value, record the anomaly time and the position in the reaction path, monitor multiphase flow data anomalies at the same time, match the data at the anomaly position of the reaction path with the multiphase flow anomaly data in space and time, and mark according to the matching result;

[0081] The dynamic correlation matrix and network mapping module is used to correlate the correlation strength between every two data in the space-time distribution matrix, generate a time-varying correlation matrix in each time window to form a sequence of dynamic correlation matrices, and map them into a dynamic correlation network, where nodes represent process variables and the weights of edges represent the correlation strength between variables;

[0082] The abnormal propagation path identification module is used to identify marked nodes, track the propagation path of node abnormalities, identify the starting point of the abnormal path and record the abnormal evolution order. At the same time, it calculates the abnormal propagation centrality of each node in the dynamic correlation network and identifies the root cause nodes based on the abnormal propagation centrality;

[0083] The root cause repair measure module is used to retrieve corresponding repair measures for different root causes according to the root cause nodes in the analyzed abnormal propagation path, generate control strategies matching the repair measures for non-root cause nodes in the abnormal propagation path, and re-evaluate the dynamic correlation network after the repair measures and control strategies are executed to determine whether the abnormality is eliminated.

[0084] The beneficial effects of the present invention are:

[0085] This solution constructs a dynamic and intelligent fault diagnosis system, significantly improving the stability of the chemical process and the efficiency of fault handling. Based on the space-time distribution matrix, this method unifies the mapping of reaction path data and multiphase flow data, realizing the dynamic fusion of variables in time and space and comprehensively reflecting the operating state inside the reactor. The introduction of a sliding time window enhances the ability to capture dynamic changes in operating conditions and provides high-precision data support for subsequent detection.

[0086] In the abnormal detection stage, the system achieves accurate identification through collaborative abnormal and isolated abnormal classification analysis, and combines the spatio-temporal matching of reaction path data and multiphase flow data to lock the source and nature of the abnormality. The construction of the dynamic correlation network further quantifies the relationship strength between variables and visually presents the abnormal propagation path and key nodes in a graph structure, providing a global perspective for abnormal propagation analysis.

[0087] Root cause analysis focuses on the starting point of the propagation path and the abnormal propagation centrality, intelligently identifying key nodes and tracing the origin of the abnormality to ensure the accuracy and scientificity of root cause location. At the same time, by tracking the propagation path and the sequence of abnormal nodes, the abnormal evolution process is clearly presented, laying a foundation for formulating effective repair measures. Brief Description of the Drawings

[0088] Figure 1 It is a flowchart of the fault diagnosis method for the operation of the chemical process provided by the embodiment of the present invention;

[0089] Figure 2Flowchart for generating the space-time distribution matrix provided by an embodiment of the present invention;

[0090] Figure 3 Flowchart for collaborative anomaly detection using reaction path data and multiphase flow data provided by an embodiment of the present invention;

[0091] Figure 4 Flowchart for generating a time-varying correlation matrix at each time window provided by an embodiment of the present invention;

[0092] Figure 5 Flowchart for identifying and marking root cause nodes based on anomaly propagation centrality provided by an embodiment of the present invention;

[0093] Figure 6 Block diagram of the fault diagnosis system for the operation of a chemical process provided by an embodiment of the present invention. Detailed implementation manners

[0094] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present application.

[0095] In the description of the present application, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present application, "a plurality" means two or more unless otherwise specifically defined.

[0096] In the description of the present application, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present application is not necessarily construed as being more preferred or having more advantages than other embodiments. For the purpose of enabling any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those skilled in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed in the present application.

[0097] Figure 1This is a flowchart of the fault diagnosis method for the operation of a chemical process provided by an embodiment of the present invention. As Figure 1 shown, the method includes:

[0098] S100, monitor the reaction path data, and collect the multiphase flow data in real time. Combining the spatial distribution of the reactor, map the reaction path data and the multiphase flow data into the same spatial coordinate system to generate a space-time distribution matrix;

[0099] In this step, high-precision sensors are distributed at different positions of the reactor to continuously monitor the concentration changes of reactants and intermediate products. The reactant concentration data not only includes the instantaneous values collected at regular intervals, but also dynamically records the different spatial positions in the reactor, enabling the data to reflect the concentration distribution of the reactants along the reaction path. In addition, through the multiphase flow monitoring device, the flow pattern distribution, flow velocity data, and gas-liquid-solid three-phase distribution in the reactor are finely collected to ensure that the dynamic change characteristics in the complex multiphase flow system can be captured in real time.

[0100] To achieve the unified fusion of space and time information, a complete three-dimensional spatial coordinate system is established, and the reactor is divided into multiple grid cells. Each grid cell serves as the basic unit for data recording and processing, and its positioning accuracy directly determines the accuracy of the subsequent analysis results. Therefore, the positions of the collected sensors are strictly mapped into this spatial coordinate system to ensure that the data of each sampling point has a clear spatial definition. To further improve the timeliness and processing efficiency of the data, the entire reaction process is divided into several fixed time windows. Within each time window, the collected data is summarized and averaged to generate representative data points, thereby ensuring the smoothing and stability of the data.

[0101] Through the organization of the above steps, a space-time distribution matrix is finally formed. This matrix stores the spatial, time, and physical information of each variable in the reactor in a three-dimensional form. Specifically, the first dimension of the matrix is the three-dimensional spatial distribution of the reactor, the second dimension is various types of collected data (such as concentration, flow velocity, and phase distribution, etc.), and the third dimension corresponds to the time series. With the help of this matrix, the dynamic behavior in the chemical process can be accurately expressed, providing a scientific basis for subsequent anomaly detection and network analysis.

[0102] By establishing a space-time distribution matrix, the unified expression of time and space information is realized. This method can not only comprehensively reflect the dynamic changes inside the reactor, but also highlight the internal relationships between various variables, providing a powerful tool for multi-variable comprehensive analysis. Secondly, through the arrangement of high-precision sensors and the division of grid cells, the accuracy of data acquisition and spatial resolution are ensured, thus providing reliable basic data support for subsequent fault diagnosis. In addition, the introduction of a time window effectively solves the problem of data fluctuation during the reaction process, making the data smoother and easier to analyze.

[0103] As Figure 2 shown, monitor the reaction path data, collect multiphase flow data in real time, and combine the spatial distribution of the reactor to uniformly map the reaction path data and multiphase flow data into the same spatial coordinate system to generate a space-time distribution matrix, specifically including:

[0104] S110, Monitor the reaction path data in the reactor in real time, including: the concentrations of reactants and intermediate products;

[0105] S120, Collect multiphase flow data in real time, including: the flow pattern distribution of the multiphase flow in the reactor, the flow velocity data of the multiphase flow in the reactor, and the real-time data of the gas, liquid, and solid three-phase distribution;

[0106] S130, Establish the spatial coordinate system of the reactor, divide the reactor into multiple grid cells, each cell is used as the basic unit of the unified coordinate, and map the position of the acquisition sensor into the spatial coordinate system to define the spatial position of each sampling point;

[0107] S140, According to the dynamic characteristics of the reaction process, divide the entire operation time into several time windows, take the average value of the data values within each time window, and organize them as a time series in the spatial coordinate system;

[0108] S150, Integrate the time-serialized data into a space-time distribution matrix.

[0109] In this step, the integration of the time-serialized data into a space-time distribution matrix is specifically:

[0110] Synthesize the space-time distribution matrix X ij (t k ), where i represents the spatial coordinate of the reactor, j represents the data collected in real time, and k is the time dimension, which is directly related to the time point:

[0111]

[0112] In the space-time distribution matrix X ijk (t):

[0113] The first dimension is the three-dimensional spatial distribution of the reactor, the second dimension is the data set, and the third dimension is the time series.

[0114] S200. Use the reaction path data and multiphase flow data for collaborative anomaly detection, compare with the expected values, record the anomaly time and the position on the reaction path, simultaneously monitor the multiphase flow data anomalies, match the data at the abnormal position of the reaction path with the multiphase flow abnormal data in space and time, and mark according to the matching results;

[0115] In this step, based on historical data and theoretical models, the expected values of key variables under normal operating conditions are calculated, including the expected concentrations of reactants and intermediate products, the normal flow pattern distribution, the normal flow velocity data and their typical distribution ranges. The determination of these expected values and typical distribution ranges is based on a profound understanding of the physical and statistical laws of chemical processes. By analyzing a large number of operating condition data in the normal operating state, representative statistical characteristic values are extracted.

[0116] To cope with the dynamic characteristics in chemical processes, the expected concentration values in the normal state and multiphase flow-related information (such as flow pattern, flow velocity) are not static, but need to be dynamically updated in the form of a sliding time window. This dynamic update can effectively capture the slow change trends in chemical operations and avoid false alarms caused by minor adjustments in operating conditions. The design of the sliding time window enables the expected value interval in each time period to reflect the current normal state fluctuation range in real time, providing a dynamic reference standard for real-time anomaly detection.

[0117] During real-time monitoring, the concentrations of reactants and intermediate products and multiphase flow data collected will be automatically compared with the corresponding expected value intervals. Within the sliding time window, the moving average of the deviation of the variable is calculated. If the moving average continuously exceeds the expected value interval, it is determined as a persistent anomaly. This way of identifying persistent anomalies can not only effectively filter out the interference of accidental fluctuations, but also highlight the abnormal points that have a substantial impact on the operation stability of chemical processes. The detected abnormal points will be detailedly recorded in terms of time and space dimensions, including the specific time points when the anomalies occur and the spatial positions on the reaction path, which provides a basis for subsequent fault location and propagation path analysis.

[0118] Furthermore, to clarify the systematic characteristics of anomalies, the abnormal spatial positions on the reaction path are matched with the abnormal data of multiphase flow. By comparing the coincidence of the two in terms of spatial position and time dimension, the types of anomalies can be clearly distinguished. If the abnormal positions of the reaction path and the abnormal positions of the multiphase flow completely coincide in space and time, it is marked as a collaborative anomaly, indicating that the anomaly may be caused by comprehensive problems arising from the operating conditions; if there is no obvious spatial or temporal coincidence between the two, it is marked as an isolated anomaly, which may indicate that the anomaly is a local or independent problem of a single variable.

[0119] As Figure 3 shown, the matching of the data of the abnormal positions on the reaction path with the abnormal data of multiphase flow in terms of space and time, and the marking based on the matching results specifically include:

[0120] S210, Calculate the expected values of the concentrations of reactants and intermediate products, the normal flow pattern distribution, the normal flow velocity data, and the typical distribution range under normal operating conditions;

[0121] S220, Dynamically update the expected values of concentrations, the normal flow pattern distribution, the normal flow velocity data, and the typical flow pattern distribution in the normal state in the form of a sliding time window, and calculate the expected value interval of the variable;

[0122] S230, Compare the concentrations of reactants and intermediate products collected in real time, as well as the multiphase flow data with the expected value interval. Calculate the moving average of the deviation within the sliding time window. If the moving average continuously exceeds the expected value interval, it is determined as a persistent anomaly, and the detected anomaly points are recorded according to time and space, including the time points when the anomalies occur and the abnormal spatial positions on the reaction path;

[0123] S240, Match the abnormal spatial positions on the reaction path with the abnormal data of multiphase flow, and compare their spatial positions and time dimensions;

[0124] S250, If the abnormal positions on the reaction path and the abnormal positions of the multiphase flow coincide in space and time, it is marked as a collaborative anomaly, otherwise it is marked as an isolated anomaly.

[0125] S300, Correlate the correlation strengths between every two data in the space-time distribution matrix, generate a correlation matrix that changes with time in each time window, form a dynamic correlation matrix sequence, and map it into a dynamic correlation network;

[0126] This step extracts a time series dataset from the space-time distribution matrix. Through collaborative calculation of these data, a correlation matrix R(t representing the correlation strength between variables is generated k)。The calculation of the association strength is based on the correlation measurement method in statistics. Specifically, covariance is used to represent the joint variation trend between variables, and at the same time, the standard deviations of the two variables are combined to normalize the covariance value. The final association strength result R jxjy (t k ) ranges from -1 to 1, reflecting the positive or negative correlation between any two variables.

[0127] During the entire analysis process, such an association matrix is not static but changes dynamically over time. To systematically capture this relationship strength evolving over time, the association matrices within the entire time range are organized in an orderly manner into a sequence of dynamic association matrices. To improve the temporal stability and noise resistance of the association strength, for each element R of the association matrix at each time point jxjy (t k ), a method of sliding time window is also introduced for data smoothing. Specifically, the arithmetic mean of the association strengths at all time points within the window is taken to obtain the smoothed association value. This process can not only eliminate the interference of random fluctuations but also more clearly reflect the true association characteristics between variables.

[0128] By mapping the sequence of dynamic association matrices into a dynamic association network, the visual expression of the multi-dimensional relationships between complex variables is simplified. The node set of the dynamic association network corresponds to the key variables in the chemical process (such as reactant concentration, flow rate, flow pattern, etc.), and the edge set represents the relationship strength between the nodes. The edge weights between the nodes are determined by the values of the association matrix.

[0129] Through the above method, a static association network can be generated at each time point, and the static networks at all time points are combined together to finally form a complete sequence of dynamic association networks. This sequence can intuitively display the dynamic association changes between variables, providing rich information for subsequent abnormal propagation analysis and root cause diagnosis.

[0130] By constructing the dynamic association matrix, the interaction strength between variables in the chemical process can be quantified from a global perspective. This not only helps to reveal the potential coupling effects between variables but also provides a data basis for identifying abnormal propagation paths. The association matrix uses the method of covariance and standard deviation normalization, which can effectively avoid calculation biases caused by differences in variable scales, making the analysis results more reliable.

[0131] Secondly, the introduction of the sliding time window is an important innovation point in this step. By smoothing the association data, the anti-interference ability of the system to random noise can be significantly improved, and at the same time, the capture effect of the evolving trend of the association strength over time can be enhanced. This dynamic smoothing method can more realistically reflect the association changes in the chemical process, laying a stable foundation for subsequent anomaly detection and propagation analysis.

[0132] The mapping of the dynamic association network further improves the visualization level of data. Expressing the relationships between variables in the form of a graph structure makes the complex associations between variables intuitive and clear. Through the threshold filtering strategy, the network removes weakly related or unimportant edges, thus simplifying the network structure and enabling engineers to focus more quickly on key variables and important relationships. In addition, the dynamic association network can also visually display the temporal evolution process of variable relationships, helping to more accurately depict the propagation path and dynamic evolution of anomalies.

[0133] As Figure 4 shown, generating an association matrix that changes over time in each time window to form a sequence of dynamic association matrices and mapping them into a dynamic association network specifically includes:

[0134] S310, extracting the real-time collected data j, j = {j ij (t k ) from the spatio-temporal distribution matrix X 1 , j 2 , j 3 ,..., j m};

[0135] S320, calculating the association strength between the data j k at the spatial position i at different time points t x , j y and generating an association matrix R(t k );

[0136] S330, generating a sequence of dynamic association matrices that change over time within the entire time range;

[0137] S340, mapping the sequence of dynamic association matrices into a dynamic association network.

[0138] In this step, calculating the association strength between the data j k at the spatial position i at different time points t x , j y and generating an association matrix specifically is:

[0139]

[0140] Among them, is the association strength value, is the covariance of the data j x and j y , and are the standard deviations of the data j x and j y respectively;

[0141] Construction time point t k 's association matrix R(t k ), used to represent the association strength between different data:

[0142]

[0143] The generation of a sequence of dynamic association matrices that change over time within the entire time range is specifically as follows:

[0144] Combine the association matrices of all time points into a sequence of dynamic association matrices:

[0145] {R(t 1 ), R(t 2 ),..., R(t K )}, where K is the total number of time points;

[0146] Use a sliding time window to smooth the matrix elements :

[0147]

[0148] where n is the width of the sliding window.

[0149] The mapping of the sequence of dynamic association matrices to a dynamic association network is specifically as follows:

[0150] The node set of the dynamic association network:

[0151] V = {v 1 , v 2 , v 3 ,..., v m}, where v j Each node corresponds to a data j;

[0152] Define the edge weights through the in the association matrix:

[0153]

[0154] And set a threshold τ, and only when keep the corresponding edge;

[0155] For each time point t k , generate a static network G(t k ) = (V, E(t k ))), where E(t k ) is the edge set at time point t k ;

[0156] Combine the static networks generated at all time points into a sequence of dynamic networks:

[0157] {G(t 1 ), G(t 2 ), G(t 3 ),..., G(t K )}。

[0158] S400 identifies the nodes with markers, traces the propagation path of node anomalies, identifies the starting point of the abnormal path and records the abnormal evolution order, and at the same time calculates the abnormal propagation centrality of each node in the dynamic association network, and identifies the root cause nodes with markers based on the abnormal propagation centrality;

[0159] In this step, the data marked as collaborative anomalies or isolated anomalies and their corresponding time points are obtained from the previous step. In the dynamic association network, the network nodes corresponding to these data are marked as abnormal nodes. Each abnormal node represents a key variable identified as abnormal at the current time point and becomes the starting point of the propagation path analysis.

[0160] Starting from these abnormal nodes, the system further analyzes their association strength with other nodes in the network. Specifically, for each abnormal node, it searches sequentially along the edge with the highest association strength (i.e., the maximum value of the relationship weight between nodes in the network) and records the path order of these connected nodes, thereby constructing a chain for the anomaly to spread from one node to other nodes. By traversing the entire network, a complete node sequence of each abnormal node on its propagation path can be generated, forming an abnormal propagation sequence. These sequences can not only reflect the propagation trend of anomalies in time and space but also provide data support for the precise positioning of root cause nodes.

[0161] Next, each propagation path is analyzed in depth, especially in combination with timestamp and association strength information, to identify the node where the anomaly first appears on the propagation chain (i.e., the starting point in the time dimension). These earliest abnormal nodes are recorded as the path starting points and are aggregated into a set of starting abnormal nodes. Determining the path starting points is a key step in root cause node identification because these nodes are often where the anomaly initially occurred.

[0162] In addition, to evaluate the importance of nodes from a global perspective, the system calculates the abnormal propagation centrality of each node in the dynamic association network. The centrality reflects the degree to which a node serves as an information dissemination hub in the network. Among them, degree centrality measures the number of direct connections of a node to other nodes in the network. The more central the network structure position occupied by the node, the higher the degree centrality; betweenness centrality measures the frequency of a node serving as a hub for the shortest paths between other nodes. The higher the betweenness centrality, the more important the "bridge" role of the node in information dissemination; closeness centrality evaluates the average shortest path length between a node and all other nodes in the network. The higher the value, the faster the node can reach other nodes. By comprehensively weighting the three metrics, the comprehensive abnormal propagation centrality of the node can be obtained.

[0163] Combining the starting point of the propagation path and the propagation centrality, a centrality threshold is set to determine the root cause node. If the comprehensive centrality value of the starting node of a certain propagation path is higher than the threshold, it is directly marked as the root cause node; otherwise, the node with the highest comprehensive centrality is selected from the propagation path as the root cause node. This determination method ensures that the selection of the root cause node takes into account both the chronological order of the occurrence of the anomaly and fully weighs the importance of the node in the global network.

[0164] As Figure 5 shown, identifying the starting point of the abnormal path and recording the abnormal evolution order, while calculating the abnormal propagation centrality of each node in the dynamic association network, and identifying and marking the root cause node based on the abnormal propagation centrality specifically includes:

[0165] S410, Obtain the abnormal data with co-abnormal or isolated abnormal marks and their corresponding time points, and mark the data node where the abnormal data is located as an abnormal node in the dynamic association network;

[0166] S420, Starting from each abnormal node, read the association strength between the abnormal node and all surrounding nodes, and search along the edge with the highest association strength, recording the path order and propagation chain of the nodes;

[0167] S430, For each abnormal node, determine all the nodes on its propagation path, generating an abnormal propagation sequence of the nodes:

[0168]

[0169] S440, Analyze each propagation path, use the timestamp and association strength information to identify the earliest abnormal node on the propagation chain, and record it as the path starting point, generating a set of starting abnormal nodes;

[0170] S450, Calculate the abnormal propagation centrality of each node in the dynamic association network, sort the nodes according to the abnormal propagation centrality, and identify the node with the highest centrality;

[0171] S460, integrate the starting point and propagation centrality of the comprehensive propagation path, set the centrality threshold, and determine the root cause node. If the centrality of the starting point node is higher than the centrality threshold, it is directly marked as the root cause node; otherwise, select the node with the highest centrality in the propagation path as the root cause node.

[0172] In this step, calculating the abnormal propagation centrality of each node in the dynamic association network specifically includes:

[0173] For each time point t k , calculate the centrality of all nodes in the current network G(t k ), including:

[0174] Degree centrality C D (v jx ):

[0175]

[0176] where deg(v) is the number of edges connected to the node, and N is the total number of nodes in the network;

[0177] Betweenness centrality C B (v jx ):

[0178]

[0179] is the number of paths passing through node v jy in the shortest path from node v jz to node v jx , is the total number of shortest paths from node v jy to node v jz ;

[0180] Closeness centrality C C (v jx ):

[0181]

[0182] d(v jx ,v jz ) is the length of the shortest path from node v jx to node v jz ;

[0183] Calculate the comprehensive abnormal propagation centrality C S (v jx ):

[0184] C S (v jx) = β 1 C D (v jx ) + β 2 C B (v jx ) + β 3 C C (v jx );

[0185] where β 1 , β 2 , β 3 are weight coefficients;

[0186] Rank all nodes according to the comprehensive abnormal propagation centrality to obtain the node importance ranking.

[0187] S500. According to the root cause nodes in the analyzed abnormal propagation path, retrieve corresponding repair measures for different root causes, generate control strategies matching the repair measures for non-root cause nodes in the abnormal propagation path, and after the repair measures and control strategies are executed, re-evaluate the dynamic association network to determine whether the anomaly is eliminated.

[0188] In this step, for the root cause nodes analyzed from the dynamic association network, combined with their physical locations, variable types, and abnormal characteristics, matching repair solutions will be retrieved from the preset repair measure library. The repair measure library is established based on the specific process requirements, equipment characteristics, and past failure experiences of the chemical process, covering standardized operation suggestions for different types of variables (such as flow rate, concentration, temperature, etc.) and different abnormal phenomena (such as deviation, fluctuation, stagnation, etc.). For example, for an anomaly caused by too high a reactant concentration, the repair measures may include adjusting the feed rate or increasing the addition of diluent; for a multiphase flow pattern anomaly, it may be necessary to adjust the stirring rate or optimize the gas-liquid distribution.

[0189] After executing the repair measures for the root cause nodes, the system also needs to generate corresponding control strategies for other non-root cause nodes in the abnormal propagation path. Although non-root cause nodes are not the direct source of the anomaly, they are affected by the anomalies of the root cause nodes and may exhibit chain reaction-like deviations or anomalies. To prevent these secondary anomalies from having further negative impacts on the system stability, the formulation of control strategies needs to precisely consider the association strength between these nodes and the root cause nodes in the dynamic association network and their influence on the overall network. Specifically, through the "abnormal propagation centrality" calculated in the dynamic association network, key nodes with higher propagation centrality are preferentially repaired to ensure that the parts most influential to the system recovery are controlled first. For example, for some variables in the propagation chain, the anomaly diffusion can be slowed down by adjusting operation parameters (such as temperature, pressure) or optimizing the controller set values.

[0190] The introduction of the critical node priority repair strategy makes the entire repair process more targeted and efficient. Nodes with high propagation centrality in the dynamic correlation network usually mean that these nodes are at the core of the network and have an important impact on the propagation of anomalies and the stability of the system. Prioritizing the repair of these nodes helps to quickly contain the chain effect of anomalies and create favorable conditions for the restoration of other nodes to normal.

[0191] After the implementation of the repair measures and control strategies, the system will re-evaluate the dynamic correlation network. This includes recalculating the correlation matrix, updating the structure of the dynamic correlation network, and monitoring whether the status of each node in the network returns to the expected value range. If all anomalies are eliminated and the dynamic correlation network returns to the normal state again, it means that the diagnosis and repair process is successfully completed; if there are still anomalies not eliminated, it is necessary to iteratively execute steps S100 to S500, further dig out the potential root causes of the unsolved problems, and re-design the repair measures and control strategies until the system fully resumes stability.

[0192] Figure 6 For the structural block diagram of the fault diagnosis system for the operation of the chemical process provided by the embodiment of the present invention, as Figure 6 shown, the system includes:

[0193] A space-time distribution matrix generation module 100, configured to monitor reaction path data, collect multiphase flow data in real time, combine the reactor space distribution, map the reaction path data and the multiphase flow data to the same space coordinate system, and generate a space-time distribution matrix;

[0194] A collaborative anomaly detection module 200, configured to perform collaborative anomaly detection using the reaction path data and the multiphase flow data, compare with the expected value, record the anomaly time and the position in the reaction path, monitor the multiphase flow data anomaly at the same time, match the data at the anomaly position of the reaction path with the multiphase flow anomaly data in space and time, and mark according to the matching result;

[0195] A dynamic correlation matrix and network mapping module 300, configured to correlate the correlation strength between every two data in the space-time distribution matrix, generate a correlation matrix that changes with time in each time window, form a dynamic correlation matrix sequence, and map it into a dynamic correlation network, where the nodes represent process variables and the weight of the edge represents the correlation strength between variables;

[0196] An anomaly propagation path identification module 400, configured to identify the nodes with marks, track the propagation path of the anomaly of the node, identify the starting point of the anomaly path and record the anomaly evolution order, calculate the anomaly propagation centrality of each node in the dynamic correlation network at the same time, and identify and mark the root cause nodes according to the anomaly propagation centrality;

[0197] The root cause repair measure module 500 is used to retrieve corresponding repair measures for different root causes according to the root cause nodes in the analyzed abnormal propagation path, generate control strategies matching the repair measures for the non-root cause nodes in the abnormal propagation path, and re-evaluate the dynamic association network after the repair measures and control strategies are executed to determine whether the abnormality is eliminated.

[0198] It should be noted that in the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0199] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0200] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded computers, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0201] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing the functions in Figure 1 one process or multiple processes and / or blocksFigure 1 Steps of the functions specified in one or more boxes.

[0203] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0204] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for fault diagnosis of chemical process operation, characterized in that: The method comprises: Monitor the reaction path data and collect multiphase flow data in real time. Combined with the reactor spatial distribution, the reaction path data and multiphase flow data are uniformly mapped to the same spatial coordinate system to generate a space-time distribution matrix. Use reaction path data and multiphase flow data to perform collaborative anomaly detection, compare with expected values, and record the anomaly time and position in the reaction path. At the same time, monitor the multiphase flow data anomaly, match the data of the reaction path anomaly position with the multiphase flow anomaly data in space and time, and mark according to the matching results; The correlation strength between each two data in the correlation space-time distribution matrix is ​​calculated, and a time-varying correlation matrix is ​​generated in each time window to form a dynamic correlation matrix sequence, which is mapped into a dynamic correlation network; Identify the nodes with marks, track the propagation path of the node anomaly, identify the starting point of the anomaly path and record the anomaly evolution order, calculate the anomaly propagation centrality of each node in the dynamic association network, and identify the marked root cause node based on the anomaly propagation centrality; According to the root cause nodes in the abnormal propagation path obtained by analysis, the corresponding repair measures are retrieved for different root causes, and control strategies matching the repair measures are generated for the non-root cause nodes in the abnormal propagation path. After the repair measures and control strategies are executed, the dynamic association network is re-evaluated to determine whether the abnormality has been eliminated.

2. The method according to claim 1, characterized in that The monitoring of reaction path data and real-time acquisition of multiphase flow data, combined with the reactor spatial distribution, uniformly mapping the reaction path data and multiphase flow data to the same spatial coordinate system, and generating a space-time distribution matrix, specifically includes: Real-time monitoring of reaction path data within the reactor, including: reactant and intermediate product concentrations; Real-time collection of multiphase flow data, including: flow pattern distribution of multiphase flow in the reactor, flow velocity data of multiphase flow in the reactor, and real-time data of gas, liquid and solid three-phase distribution; Establish a spatial coordinate system for the reactor, divide the reactor into multiple grid units, each unit is used as the basic unit of unified coordinates, and map the acquisition sensor position to the spatial coordinate system to define the spatial position of each sampling point; According to the dynamic characteristics of the reaction process, the entire running time is divided into several time windows, and the data values ​​in each time window are averaged and organized into a time series in the spatial coordinate system; Integrate the time-series data into a space-time distribution matrix.

3. The method according to claim 2, characterized in that The time-series data is integrated into a space-time distribution matrix, specifically: Synthesized space-time distribution matrix X ij (t k ), where i represents the spatial coordinates of the reactor, j represents the data collected in real time, and k is the time dimension, which is directly related to the time point: In the space-time distribution matrix X ijk (t) In: The first dimension is the three-dimensional spatial distribution of the reactor, the second dimension is the data set, and the third dimension is the time series.

4. The method according to claim 2, characterized in that: The data of the abnormal position of the reaction path is matched with the abnormal multiphase flow data in space and time, and marked according to the matching result, specifically including: Calculate the expected concentration values ​​of reactants and intermediates, normal flow pattern distribution, normal flow rate data and typical distribution range under normal operating conditions; Dynamically update the expected value of concentration in normal state, normal flow pattern distribution, normal flow velocity data and typical flow pattern distribution in the form of a sliding time window, and calculate the expected value interval of the variable; Compare the concentrations of reactants and intermediate products, and multiphase flow data collected in real time with the expected value range, and calculate the moving average of the deviation within the sliding time window. If the moving average continuously exceeds the expected value range, it is determined to be a persistent anomaly, and the detected anomaly points are recorded in time and space, including the time point when the anomaly occurred and the abnormal spatial position on the reaction path; Match the abnormal spatial position of the reaction path with the multiphase flow abnormal data and compare the spatial position and time dimension of the two; If the reaction path anomaly position and the multiphase flow anomaly position coincide in space and time, they are marked as collaborative anomalies, otherwise they are marked as isolated anomalies.

5. The method according to claim 4, characterized in that The method generates a time-varying association matrix in each time window to form a dynamic association matrix sequence and maps it into a dynamic association network, specifically including: From the space-time distribution matrix X ij (t k ) to extract the real-time collected data j, j = {j1, j2, j3, ..., j m }; Calculate different time points t k In the example, data j at spatial position i x ,j y The correlation strength between them is calculated and the correlation matrix R(t k ); Generate a time-varying sequence of dynamic correlation matrices over the entire time range; Mapping dynamic association matrix sequences into dynamic association networks.

6. The method according to claim 5, characterized in that The calculation of different time points t k In the example, data j at spatial position i x ,j y The correlation strength between them is calculated and the correlation matrix is ​​generated, which is as follows: in, is the association strength value, For data j x and j y The covariance of and are data j x and j y The standard deviation of Construction time point t k The incidence matrix R(t k ), which is used to indicate the strength of association between different data: The generation of a dynamic correlation matrix sequence that changes with time within the entire time range is specifically as follows: Combine the incidence matrices for all time points into a sequence of dynamic incidence matrices: {R(t1),R(t2),...,R(t K )}, where K is the total number of time points; Use sliding time windows to filter matrix elements To smooth: Where n is the width of the sliding window.

7. The method according to claim 6, characterized in that The mapping of the dynamic association matrix sequence into a dynamic association network is specifically as follows: A collection of nodes in a dynamically associated network: V={v1,v2,v3,...,v m }, where v j Each node corresponds to a data j; Through the correlation matrix Define edge weights: And set the threshold τ, only when Keep the corresponding edges; For each time point t k , generate a static network G(t k )=(V,E(t k )), where E(t k ) is the time point t k The edge set of ; Combine the static networks generated at all time points into a dynamic network sequence: {G(t1),G(t2),G(t3),...,G(t K )}。 8. The method according to claim 5, characterized in that The identification of the starting point of the abnormal path and recording the abnormal evolution order, while calculating the abnormal propagation centrality of each node in the dynamic association network, and identifying and marking the root cause node according to the abnormal propagation centrality, specifically includes: Obtain abnormal data with collaborative anomaly or isolated anomaly marks and their corresponding time points, and mark the data nodes where the abnormal data are located as abnormal nodes in the dynamic association network; Starting from each abnormal node, read the correlation strength between the abnormal node and all surrounding nodes, and search along the edge with the highest correlation strength, recording the path sequence and propagation chain of the nodes; For each abnormal node, determine all nodes on its propagation path and generate the abnormal propagation sequence of the node: Analyze each propagation path, use timestamp and correlation strength information to identify the earliest abnormal node on the propagation chain, and record it as the starting point of the path to generate a set of starting abnormal nodes; Calculate the anomaly propagation centrality of each node in the dynamic association network, sort the nodes according to the anomaly propagation centrality, and identify the nodes with the highest centrality; The starting point and propagation centrality of the propagation path are comprehensively considered, and the centrality threshold is set to determine the root cause node. If the centrality of the starting point node is higher than the centrality threshold, it is directly marked as the root cause node. Otherwise, the node with the highest centrality in the propagation path is selected as the root cause node.

9. The method according to claim 8, characterized in that The calculation of the abnormal propagation centrality of each node in the dynamic association network is specifically: For each time point t k , calculate the current network G(t k ), including: Degree centrality C D (v jx ): Where deg(v) is the number of edges connected to the node, and N is the total number of nodes in the network; Betweenness Centrality C B (v jx ): For slave node v jy To node v jz The shortest path passes through node v jx The number of paths, For slave node v jy To node v jz The total number of shortest paths; Closeness Centrality C C (v jx ): d(v jx ,v jz ) is the node v jx To node v jz The shortest path length; Calculate the comprehensive anomaly propagation centrality C S (v jx ): C S (v jx )=β1C D (v jx )+β2C B (v jx )+β3C C (v jx ); Among them, β1, β2, and β3 are weight coefficients; All nodes are sorted according to the comprehensive anomaly propagation centrality to obtain the node importance ranking.

10. A fault diagnosis system for chemical process operation, characterized in that: The system comprises: The space-time distribution matrix generation module is used to monitor the reaction path data and collect multiphase flow data in real time. Combined with the reactor spatial distribution, the reaction path data and multiphase flow data are uniformly mapped to the same spatial coordinate system to generate the space-time distribution matrix. The collaborative anomaly detection module is used to perform collaborative anomaly detection using reaction path data and multiphase flow data, compare them with expected values, and record the anomaly time and position in the reaction path. At the same time, it monitors the anomaly of the multiphase flow data, matches the data of the reaction path anomaly position with the multiphase flow anomaly data in space and time, and marks them according to the matching results. The dynamic association matrix and network mapping module is used to associate the association strength between each two data in the space-time distribution matrix, and generate a time-varying association matrix in each time window to form a dynamic association matrix sequence, which is mapped into a dynamic association network, where the nodes represent process variables and the edge weights represent the association strength between variables; The abnormal propagation path identification module is used to identify the nodes with marks, track the propagation path of the abnormality of the node, identify the starting point of the abnormal path and record the abnormal evolution order, and calculate the abnormal propagation centrality of each node in the dynamic association network, and identify the marked root cause node based on the abnormal propagation centrality; The root cause repair measure module is used to retrieve corresponding repair measures for different root causes based on the root cause nodes in the abnormal propagation path obtained by analysis, and to generate control strategies that match the repair measures for non-root cause nodes in the abnormal propagation path. After the repair measures and control strategies are executed, the dynamic association network is re-evaluated to determine whether the abnormality has been eliminated.

Images