AI-based petrochemical equipment fault strength evaluation method and system

Abstract

The invention relates to the technical field of petrochemical equipment fault evaluation, and discloses an AI-based petrochemical equipment fault strength evaluation method and system, and the method comprises the following steps: collecting multi-dimensional data of a full life cycle of petrochemical equipment, extracting fault data in the multi-dimensional data, and constructing a fault database; comprehensively and dynamically evaluating the fault strength of the fault database according to the degrees of fault occurrence, fault severity and fault influence; wherein according to the fault occurrence degree, fault occurrence related data are input into the fault occurrence degree prediction model, and a fault occurrence proportion prediction value is obtained. According to the method, the fault database is constructed by integrating full-life-cycle data of petrochemical equipment, a full-dimension basis is provided for fault risks, and the possibility, the hazard degree and the linkage effect of faults are comprehensively evaluated by combining dynamic analysis of multi-dimension indexes, real-time learning of equipment state changes and the multi-dimension indexes, so that the fault strength is scientifically graded, and the fault risk is improved. And thus, the comprehensiveness and timeliness of fault assessment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical equipment fault evaluation, and in particular to an AI-based petrochemical equipment fault intensity evaluation method and system. Background Art

[0002] Petrochemical equipment is the foundation of an enterprise's production activities. Once a failure occurs, production activities will be interrupted. Severe equipment failure may even lead to production interruption or property loss. Therefore, equipment managers in the petrochemical industry attach great importance to equipment failures. On the one hand, they need to repair, handle and analyze the failures to reduce losses, resume production and prevent similar failures from happening again. On the other hand, they need to conduct strength evaluation on the mechanism where the failure occurs, so as to guide the attention to failure management and serve as a warning. However, current failure intensity evaluations of petrochemical equipment mostly focus on real-time data during equipment operation, while ignoring data from the entire life cycle. For example, the long-term impact of design and manufacturing parameters, maintenance history, and scrap data on failure risk has not been fully utilized, resulting in incomplete evaluation basis. In addition, most evaluations are based on a single intensity indicator, such as the probability of failure, which results in incomplete evaluation results. Summary of the invention

[0003] The purpose of the present invention is to provide an AI-based petrochemical equipment fault intensity evaluation method and system to solve at least one of the above-mentioned prior art problems.

[0004] In a first aspect, the present invention provides a petrochemical equipment fault intensity evaluation method based on AI, comprising the following steps: Collect multi-dimensional data of the entire life cycle of petrochemical equipment, extract fault data from the multi-dimensional data, and build a fault database; Through the degree of fault occurrence, fault severity and fault impact, the fault database is comprehensively evaluated dynamically for fault intensity; Among them, the degree of fault occurrence: input the relevant data of fault occurrence into the fault occurrence prediction model to obtain the predicted value of fault occurrence ratio; use the moving average method to predict the density of fault occurrence according to the number of fault occurrences; combine the predicted value of fault occurrence ratio and the predicted fault occurrence density to evaluate the degree of fault occurrence; Fault severity: Perform similarity assessment on fault-related data and historical fault events to obtain similar fault events used to assess fault severity; Fault impact degree: Based on the fault propagation path, a set of petrochemical equipment nodes is obtained to evaluate the fault impact degree; Fault-related data: Extract data features based on the fault database and use the random forest algorithm to screen out data features with high accuracy.

[0005] As a further solution of the present invention: the process of calculating the fault intensity evaluation index and classifying the fault intensity level is as follows: The degree of fault occurrence, the degree of fault severity and the degree of fault impact are weighted to obtain the fault intensity evaluation index.

[0006] As a further solution of the present invention: the process of obtaining the degree of the fault occurrence is: Extract data from the fault database, integrate them to obtain feature matrix and target vector, and filter out data features related to fault occurrence in the feature matrix; Set the time window, based on the fault occurrence prediction model, input the relevant data features of the fault occurrence, and calculate the predicted value of the fault occurrence ratio; Count the density of faults in each time window, use the moving average method to predict the density of faults in the next time window, and perform normalization processing; The predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are weighted to obtain the degree of fault occurrence.

[0007] As a further solution of the present invention: the process of obtaining the predicted value of the fault occurrence ratio is as follows: Inputting the real-time acquired fault occurrence related data features into the fault occurrence prediction model to obtain the fault occurrence prediction value; The statistical output value is the number of faults that occurred, which is calculated by ratio with the total number of output values ​​in the time window to obtain the predicted value of the fault occurrence ratio.

[0008] As a further solution of the present invention: the process of screening the data features related to the occurrence of faults in the feature matrix is: Train the random forest model through the feature matrix and target vector; for any feature data; Get the out-of-bag data corresponding to each decision tree, input it into the trained random forest model for prediction, and get the predicted values. By comparing these predicted values ​​with the actual values ​​of the out-of-bag data, calculate the prediction accuracy of the original out-of-bag data and mark it as the original accuracy. Use the trained random forest model to predict the out-of-bag data after the feature values ​​are randomly sorted to obtain new predicted values. Compare the new predicted values ​​with the true values, calculate the new out-of-bag data prediction accuracy, and mark it as the new accuracy. Calculate the difference between the original accuracy and the new accuracy to obtain the accuracy deviation value; The accuracy deviation values ​​corresponding to all features are ranked in descending order, and the fault features before the ranking percentage threshold are extracted and marked as important features.

[0009] As a further solution of the present invention: the process of obtaining the severity of the fault is as follows: Obtain historical fault event data and real-time fault event data, based on the combination of the real-time fault event and any historical fault event data; Calculate the similarity value of each combination, and extract historical fault events that are greater than the similarity limit as similar fault events; The similarity values ​​corresponding to similar fault events are used as weights, and the weighted average is calculated to obtain the severity of the fault.

[0010] As a further solution of the present invention: the process of obtaining the similarity value is: Based on the combination of real-time fault events and any historical fault event data, the corresponding fault occurrence-related data features are extracted respectively; The similarity value of each combination is calculated by the cosine similarity method.

[0011] As a further solution of the present invention: the process of obtaining the degree of the fault impact is as follows: Set the nodes and edges of the petrochemical equipment related graph respectively; When a petrochemical equipment fails, the propagation process of the fault in the petrochemical equipment related graph is simulated to obtain and construct the set of petrochemical equipment nodes affected during the fault propagation process; The number of nodes in the petrochemical equipment node set is counted, and the ratio is calculated with the total number of petrochemical equipment. Then it is multiplied by the normalized value of the sum of the weights of all edges on the fault propagation path to obtain the fault impact evaluation coefficient.

[0012] As a further solution of the present invention: the process of evaluating the fault intensity is: If the fault intensity evaluation index is less than the first limit value, it is classified as a low intensity fault; If the fault intensity evaluation index is greater than or equal to the first limit value and if the fault intensity evaluation index is less than the second limit value, it is classified as a medium-low intensity fault; If the fault intensity evaluation index is greater than or equal to the second limit value and if the fault intensity evaluation index is less than the third limit value, it is determined to be a medium-high intensity fault; If the fault intensity evaluation index is greater than or equal to the third limit value, it is classified as a high intensity fault.

[0013] In a second aspect, the present invention provides an AI-based petrochemical equipment fault intensity evaluation system, the system comprising: Fault data construction module: collects multi-dimensional data of the entire life cycle of petrochemical equipment, extracts fault data from the multi-dimensional data, and builds a fault database; Fault intensity evaluation module: It dynamically evaluates the fault intensity of the fault database based on the occurrence, severity and impact of the fault. Among them, the degree of fault occurrence: input the relevant data of fault occurrence into the fault occurrence prediction model to obtain the predicted value of fault occurrence ratio; use the moving average method to predict the density of fault occurrence according to the number of fault occurrences; combine the predicted value of fault occurrence ratio and the predicted fault occurrence density to evaluate the degree of fault occurrence; Fault severity: Perform similarity assessment on fault-related data and historical fault events to obtain similar fault events used to assess fault severity; Fault impact degree: Based on the fault propagation path, a set of petrochemical equipment nodes is obtained to evaluate the fault impact degree; Fault-related data: Extract data features based on the fault database and use the random forest algorithm to screen out data features with high accuracy.

[0014] Beneficial effects of the present invention: 1. The present invention integrates the data of the entire life cycle of petrochemical equipment to build a fault database, provides a full-dimensional basis for fault risks, dynamically analyzes multi-dimensional indicators, learns equipment status changes in real time, and comprehensively evaluates the possibility, degree of harm and chain effects of faults with multi-dimensional indicators, thereby scientifically grading the fault intensity and improving the comprehensiveness and timeliness of fault assessment; 2. The present invention constructs a fault occurrence prediction model based on the selected important features. Since the input features all have a significant impact on the fault occurrence, when the random forest algorithm is used for retraining, the model can better capture the relationship between the features and the target vector. The trained model has better prediction performance and can more accurately predict the fault occurrence of petrochemical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 It is a flow chart of a petrochemical equipment fault intensity evaluation method based on AI of the present invention; Figure 2 It is a structural schematic diagram of an AI-based petrochemical equipment fault intensity evaluation system of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0018] Embodiment 1 like Figure 1 As shown, an AI-based petrochemical equipment fault intensity evaluation method provided by an embodiment of the present invention specifically includes the following steps: Step 1: Collect multi-dimensional data of the entire life cycle of petrochemical equipment and build a petrochemical equipment failure database; Glossary: ​​The entire life cycle of petrochemical equipment includes: design and manufacturing period, operation period, maintenance period, and scrap period; In step 1, the multi-dimensional data includes but is not limited to the following: basic parameters of the design and manufacturing period; data collection of basic parameters: the equipment type is a centrifugal pump, used for crude oil centrifugation, and the design pressure is 10Mpa; Process parameters, vibration signals, and lubricant status during operation; process parameters include: flow, pressure, temperature, and liquid level; vibration signals include: acceleration, displacement, and velocity; lubricant status includes: viscosity, particle size, and moisture content; Fault records and maintenance results during the maintenance period; Fault records include: fault occurrence time, fault phenomenon, failed components, and failure consequences; Maintenance results include: reduction ratio of vibration value after maintenance; The reasons for the end of life and the residual strength during the scrap period. The reasons for the end of life include: failure mode, cumulative operating time, and the number of start-stop cycles; the residual strength includes: wall thickness measurement, material mechanical properties, and crack extension length; The process of building a petrochemical equipment fault database is as follows: The acquired petrochemical equipment is cleaned and preprocessed, wherein the data cleaning and preprocessing include but are not limited to: for the missing values ​​in the extracted feature data, methods such as deletion, interpolation or filling can be used to process; identify and process the outliers in the feature data; normalize or standardize different types of feature data; Integrate the full life cycle data of petrochemical equipment and build a petrochemical equipment failure database; For example, the data integration of the main air blower unit of the catalytic cracking unit is as follows: Design period: API617 standard design, impeller material Inconel718; Operation period: The vibration amplitude of 1 times frequency suddenly increased in the third year, and the oil particle size increased from ISO16 / 14 to 19 / 17; Maintenance period: Disassembly found that the impeller was coked, and online cleaning was carried out, MTTR = 24 hours; Scrap period: 10 years of operation with a total of 43,800 hours, retired due to impeller fatigue cracks; Step 2: Based on the construction of petrochemical equipment fault database, dynamically analyze the fault database to obtain fault intensity evaluation indicators; Among them, the fault intensity evaluation index is calculated by the fault occurrence evaluation coefficient, the fault severity evaluation coefficient and the fault impact evaluation coefficient; In some embodiments, based on a petrochemical equipment fault database, a fault occurrence evaluation coefficient and a fault severity evaluation coefficient are dynamically calculated; In some possible embodiments, the process of obtaining the fault occurrence evaluation coefficient is as follows: Extract data from petrochemical equipment failure database and integrate them to obtain feature matrix and target vector; The target vector is used to indicate whether the device has a fault, with a fault being 1 and a non-fault being 0; Use the random forest algorithm to rank the importance of all features in the feature matrix. The process is as follows: Setting the parameters of the random forest model, wherein the parameters include: the number of decision trees, which is set by a person skilled in the art according to the complexity of the feature data. Since petrochemical equipment contains multiple feature data, the number of decision trees is generally set in the range of 100-500 trees; The maximum depth of a single tree is generally set to 5-15 because the fault features of petrochemical equipment are hierarchically associated. For example, the feature association path is: abnormal vibration-bearing wear-shutdown; The minimum sample size for node splitting is set by technical personnel in the field based on historical fault data obtained during the maintenance period. The minimum sample size for node splitting is generally set in the range of 2-10. The number of features randomly sampled by a single tree is sampled according to the square root of the total number of features, thereby enhancing model diversity. For example, if the number of all features is 100, the number of features randomly sampled by a single tree is 10; Divide the training set and validation set in a ratio of 7:3 and train the random forest model; The indicator factor data that did not participate in the training during the modeling process of the random forest model was taken as out-of-bag data, and the original out-of-bag error of the random forest model was calculated using the out-of-bag data; Among them, out-of-bag data refers to data that is not involved in the training process of the random forest model, which can be used to evaluate the performance of the model. In the process of building a decision tree each time, each sample has a certain probability of not being selected into the training sample set of the current decision tree. This part of the unselected data is the out-of-bag data; For any feature data; Get the out-of-bag data corresponding to each decision tree, input it into the trained random forest model for prediction, and get the predicted values. By comparing these predicted values ​​with the actual values ​​of the out-of-bag data, calculate the prediction accuracy of the original out-of-bag data and mark it as the original accuracy. Randomly sort the values ​​of the feature in the out-of-bag data, while the values ​​of other features remain unchanged; In a data set containing multiple features, each sample is described by multiple feature values. When evaluating the importance of a feature, the value of the feature in the out-of-bag data is randomly shuffled, while the values ​​of other features remain the same. In this way, the original data order of the feature is changed, and the possible intrinsic connection between it and the target variable is cut off. Then, the model is used to make predictions with the adjusted data, and the changes in the prediction results are compared to determine the importance of the feature. Use the trained random forest model to predict the out-of-bag data after the feature values ​​are randomly sorted to obtain new predicted values. Compare the new predicted values ​​with the true values, calculate the new out-of-bag data prediction accuracy, and mark it as the new accuracy. Calculate the difference between the original accuracy and the new accuracy to obtain the accuracy deviation value; Among them, the larger the accuracy deviation value, the more important the feature is and the higher its prediction accuracy; Sort the accuracy deviation values ​​corresponding to all features in descending order to obtain a fault feature correlation ranking table; Set the ranking percentage threshold, extract the fault features before the ranking percentage threshold, and mark them as important features, that is, the data related to the fault occurrence: For example, if the ranking percentage threshold is 30%, and there are P features in the feature matrix, then the features with 0.3P ranking numbers are extracted as important features. If 0.3P is not an integer, it is rounded forward, that is, the integer digit of the value 0.3P is extracted; The feature importance ranking results of the random forest model are highly interpretable. In petrochemical equipment failure analysis, technicians can intuitively understand the impact of each feature on the occurrence of failures based on the feature importance ranking table, so as to carry out equipment maintenance and troubleshooting in a targeted manner. For example, if the importance of the lubricant status feature is found to be high, the monitoring and replacement of the lubricant can be strengthened to prevent equipment failures. Based on the important features, a fault occurrence prediction model is constructed, and the random forest algorithm is used again to train the important features and target vectors to obtain the fault occurrence prediction model; It should be noted that the fault occurrence prediction model is constructed based on the selected important features. Since the input features have a significant impact on the fault occurrence, when the random forest algorithm is used for retraining, the model can better capture the relationship between the features and the target vector. The trained model has better prediction performance and can more accurately predict the fault occurrence of petrochemical equipment. A time window is set, and the real-time data of the petrochemical equipment operation period is input into the fault occurrence prediction model to obtain the fault occurrence prediction value, i.e., 1 or 0; Count the number of output values ​​1, calculate the ratio between it and the total number of output values ​​in the time window, and get the predicted value of the fault occurrence ratio; Count the number of faults that occur in each time window (i.e., fault density), and use the moving average method to predict the fault density based on the number of faults that occur in all time windows; Set the moving average window size, where the moving average window size is based on the equipment operation cycle and historical fault frequency, and can be set to 3 or 5; Calculate the moving average of N windows in sequence, that is, the average level of the number of faults in N consecutive time windows including the current window; Get the latest moving average value as the predicted value of the fault density in the next time window; Normalize the predicted value of the fault occurrence density and map its value range to [0,1]; The weights of the predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are calculated based on the entropy weight method, and the predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are combined with the calculated weights for weighted calculation to obtain the degree of fault occurrence; The entropy value of the indicator in the entropy weight method is used to measure the uncertainty of the indicator data or the degree of information disorder. The indicator with a low entropy value has a large data difference and a strong ability to distinguish samples. For example, in the evaluation of petrochemical equipment, if the entropy value of the "fault occurrence density" indicator is low, it means that its numerical difference in different time windows is significant, which is more critical for judging the equipment fault status; It should be explained that the predicted value of the failure rate is the quantification of the probability dimension, and the predicted value of the failure density is the quantification of the time dimension, thereby comprehensively and dynamically evaluating the intensity of the failure occurrence of petrochemical equipment; In some possible embodiments, the process of obtaining the fault severity evaluation coefficient is as follows: Based on the petrochemical equipment failure database, historical fault event data of petrochemical equipment failures is obtained, and corresponding data features related to the occurrence of the failures are extracted; The technical personnel in this field classify the severity level according to the severity of the fault event, wherein the severity level includes four levels: slight, general, severe, and very severe, and are assigned values ​​of 1, 2, 3, and 4 respectively; When a new petrochemical equipment failure occurs (i.e., a real-time failure), the corresponding failure event data is obtained and the relevant data features of the failure are extracted; Normalize the historical fault occurrence-related data features and fault occurrence-related data features respectively, and map the value of each feature to the [0,1] interval; Based on the combination of the real-time fault occurrence related data features and any historical fault occurrence related data features, the similarity value of each combination is calculated by the cosine similarity method; Set a similarity limit, extract historical fault events that are greater than the similarity limit, and mark them as similar fault events; It should be noted that if there are no similar fault events, the fault severity level will be set by professional technicians in this field based on historical experience and fault characteristics; The similarity values ​​corresponding to similar fault events are used as weights to calculate the weighted average , the calculation formula is: ; Where k represents the number of similar fault events, represents the similarity value of the i-th similar fault event, represents the fault severity level of the i-th similar fault event; The weighted average value is normalized using the minimum-maximum normalization method to obtain the severity of the fault; In some possible embodiments, the process of obtaining the fault impact evaluation coefficient is as follows: Set the nodes and edges of the petrochemical equipment related graph respectively; The nodes are petrochemical equipment, and the edges are the relationships between petrochemical equipment. The edges can be directed or undirected, and each edge can be given a weight to indicate the closeness of the association between the equipment. For example, in a process flow, the weight of the associated edge between upstream and downstream petrochemical equipment can be determined based on material flow and energy transfer factors; The relationships between petrochemical equipment include but are not limited to: physical connection relationships, process flow associations, and functional dependencies; When a petrochemical equipment fails, simulate the propagation process of the fault in the petrochemical equipment related graph; In the simulation process, a rule-based propagation method can be used. For example, if there is an edge connecting device A and device B, and the weight of the edge exceeds the weight threshold, then the failure of device A may propagate to device B. Obtain and construct a set of petrochemical equipment nodes affected during the fault propagation process; The number of nodes NJ in the petrochemical equipment node set is counted, and the ratio is calculated with the total number of petrochemical equipment NS, and then multiplied by the normalized value of the sum of the weights W of all edges on the fault propagation path to obtain the degree of fault impact C. The calculation formula is: ; in, is the maximum value of the sum of edge weights on the fault propagation path; The degree of fault occurrence, the degree of fault severity and the degree of fault impact are weighted to obtain the fault intensity evaluation index; It should be explained that the fault occurrence reflects the possibility of fault occurrence; the fault severity focuses on the hazard level of the fault itself; and the fault impact focuses on the scope of fault propagation. Therefore, the three values ​​are combined and calculated to comprehensively evaluate the fault risk from multiple dimensions, providing sufficient basis for decision-making on equipment maintenance, production scheduling, safety management, etc. It should be noted that the AI ​​technologies used mainly include random forest algorithm and entropy weight method; Step 3: Classify the fault intensity level based on the fault intensity evaluation index; In some embodiments, a fault intensity evaluation index is obtained, and a fault intensity division limit is set, the division limit includes a first limit, a second limit, and a third limit, and the first limit is smaller than the second limit and smaller than the third limit; If the fault intensity evaluation index is less than the first limit value, it is classified as a low intensity fault; Based on low-intensity faults, the frequency of attention to such faults can be appropriately reduced, but changes in equipment-related parameters still need to be monitored regularly; If the fault intensity evaluation index is greater than or equal to the first limit value and if the fault intensity evaluation index is less than the second limit value, it is classified as a medium-low intensity fault; Based on low and medium intensity failures, it is necessary to strengthen the monitoring of petrochemical equipment and formulate preventive maintenance plans; If the fault intensity evaluation index is greater than or equal to the second limit value and if the fault intensity evaluation index is less than the third limit value, it is determined to be a medium-high intensity fault; For medium and high intensity faults, professional technicians need to be immediately organized to conduct assessments, including but not limited to: developing detailed repair plans, prioritizing repair resources, and repairing equipment as soon as possible to reduce losses; If the fault intensity evaluation index is greater than or equal to the third limit value, it is classified as a high intensity fault; Due to high-intensity failures, emergency measures must be taken immediately, including but not limited to: stopping related production processes, evacuating personnel, and concentrating superior resources to analyze the cause of the failure and formulate long-term improvement measures to reduce the recurrence of high-intensity failures; The technical solution of this embodiment is: first, collect multi-dimensional data of the entire life cycle of petrochemical equipment to build a fault database, then dynamically analyze the database based on AI technology, predict the fault occurrence through the random forest model, match the historical fault severity through cosine similarity, and simulate the fault propagation impact through the equipment association diagram, and calculate the fault intensity evaluation index by combining the three indicators. Finally, according to the preset threshold, the fault is divided into four levels of low, medium-low, medium-high, and high intensity, corresponding to different maintenance strategies; Therefore, by integrating the data of the entire life cycle of petrochemical equipment to build a fault database, and combining AI to dynamically analyze multi-dimensional indicators, the limitations of traditional methods that rely on single operating data are optimized. Specifically, the data of the entire life cycle provides a full-dimensional basis for fault risks, such as design defects and maintenance records that can trace the root cause of faults; AI models learn equipment status changes in real time, predict fault probabilities, and reduce static misjudgments; multi-dimensional indicators comprehensively evaluate the possibility, degree of harm, and chain effects of faults, and achieve scientific grading of fault intensity (low, medium-low, medium-high, and high intensity), thereby improving the comprehensiveness and timeliness of fault assessment, and providing data support for enterprises to optimize maintenance strategies, prevent accident escalations, and allocate resources.

[0019] Embodiment 2 Based on the above embodiments, Figure 2 As shown, an AI-based petrochemical equipment fault intensity evaluation system provided by an embodiment of the present invention specifically includes: Fault data construction module: collects multi-dimensional data of the entire life cycle of petrochemical equipment and builds a petrochemical equipment fault database; In this embodiment, the multi-dimensional data includes but is not limited to the following: basic parameters of the design and manufacturing period; data collection of basic parameters: the equipment type is a centrifugal pump, used for crude oil centrifugation, and the design pressure is 10Mpa; Process parameters, vibration signals, and lubricant status during operation; process parameters include: flow, pressure, temperature, and liquid level; vibration signals include: acceleration, displacement, and velocity; lubricant status includes: viscosity, particle size, and moisture content; Fault records and maintenance results during the maintenance period; Fault records include: fault occurrence time, fault phenomenon, failed components, and failure consequences; Maintenance results include: reduction ratio of vibration value after maintenance; The reasons for the end of life and the residual strength during the scrap period. The reasons for the end of life include: failure mode, cumulative operating time, and the number of start-stop cycles; the residual strength includes: wall thickness measurement, material mechanical properties, and crack extension length; The process of building a petrochemical equipment fault database is as follows: The acquired petrochemical equipment is cleaned and preprocessed, wherein the data cleaning and preprocessing include but are not limited to: for the missing values ​​in the extracted feature data, methods such as deletion, interpolation or filling can be used to process; identify and process the outliers in the feature data; normalize or standardize different types of feature data; Integrate the full life cycle data of petrochemical equipment and build a petrochemical equipment failure database; Fault data analysis module: Based on the construction of petrochemical equipment fault database, dynamically analyze the fault database and obtain fault intensity evaluation indicators; The fault intensity evaluation index is calculated by the fault occurrence evaluation coefficient, fault severity evaluation coefficient and fault impact evaluation coefficient; Based on the petrochemical equipment fault database, dynamically calculate the fault occurrence evaluation coefficient and fault severity evaluation coefficient; The process of obtaining the fault occurrence evaluation coefficient is as follows: Extract data from petrochemical equipment failure database and integrate them to obtain feature matrix and target vector; The target vector is used to indicate whether the device has a fault, with a fault being 1 and a non-fault being 0; Use the random forest algorithm to sort the importance of all features in the feature matrix and obtain a ranking table of fault feature relevance; Set a ranking percentage threshold, extract fault features before the ranking percentage threshold, and mark them as important features; Based on the important features, a fault occurrence prediction model is constructed, and the random forest algorithm is used again to train the important features and target vectors to obtain the fault occurrence prediction model; A time window is set, and the real-time data of the petrochemical equipment operation period is input into the fault occurrence prediction model to obtain the fault occurrence prediction value, i.e., 1 or 0; Count the number of output values ​​1, calculate the ratio between it and the total number of output values ​​in the time window, and get the predicted value of the fault occurrence ratio; Count the number of faults that occur in each time window (i.e., fault density), and use the moving average method to predict the fault density based on the number of faults that occur in all time windows; Set the moving average window size, where the moving average window size is based on the equipment operation cycle and historical fault frequency, and can be set to 3 or 5; Calculate the moving average of N windows in sequence, that is, the average level of the number of faults in N consecutive time windows including the current window; Get the latest moving average value as the predicted value of the fault density in the next time window; Normalize the predicted value of the fault occurrence density and map its value range to [0,1]; The weights of the predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are calculated based on the entropy weight method, and the predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are combined with the calculated weights for weighted calculation to obtain the degree of fault occurrence; The process of obtaining the fault severity evaluation coefficient is as follows: Based on the petrochemical equipment failure database, historical fault event data of petrochemical equipment failures is obtained, and corresponding data features related to the occurrence of the failures are extracted; When a new petrochemical equipment failure occurs (i.e., a real-time failure), the corresponding failure event data is obtained and the relevant data features of the failure are extracted; Normalize the historical fault occurrence-related data features and fault occurrence-related data features respectively, and map the value of each feature to the [0,1] interval; Based on the combination of the real-time fault occurrence related data features and any historical fault occurrence related data features, the similarity value of each combination is calculated by the cosine similarity method; Set a similarity limit, extract historical fault events that are greater than the similarity limit, and mark them as similar fault events; It should be noted that if there are no similar fault events, the fault severity level will be set by professional technicians in this field based on historical experience and fault characteristics; The similarity values ​​corresponding to similar fault events are used as weights to calculate the weighted average; The weighted average value is normalized using the minimum-maximum normalization method to obtain the severity of the fault; The process of obtaining the fault impact evaluation coefficient is as follows: Set the nodes and edges of the petrochemical equipment related graph respectively; The nodes are petrochemical equipment, and the edges are the relationships between petrochemical equipment. The edges can be directed or undirected, and each edge can be given a weight to indicate the closeness of the association between the equipment. The relationships between petrochemical equipment include but are not limited to: physical connection relationships, process flow associations, and functional dependencies; When a petrochemical equipment fails, simulate the propagation process of the fault in the petrochemical equipment related graph; Obtain and construct a set of petrochemical equipment nodes affected during the fault propagation process; Count the number of nodes in the petrochemical equipment node set, calculate the ratio with the total number of petrochemical equipment, and then multiply it by the normalized value of the sum of the weights of all edges on the fault propagation path to obtain the degree of fault impact. The degree of fault occurrence, the degree of fault severity and the degree of fault impact are weighted to obtain the fault intensity evaluation index; Fault intensity classification module: classifies faults into intensity levels based on fault intensity evaluation indicators; Obtaining a fault intensity evaluation index, setting a fault intensity division limit, the division limit including a first limit, a second limit and a third limit, and the first limit is smaller than the second limit and smaller than the third limit; If the fault intensity evaluation index is less than the first limit value, it is classified as a low intensity fault; If the fault intensity evaluation index is greater than or equal to the first limit value and if the fault intensity evaluation index is less than the second limit value, it is classified as a medium-low intensity fault; If the fault intensity evaluation index is greater than or equal to the second limit value and if the fault intensity evaluation index is less than the third limit value, it is determined to be a medium-high intensity fault; If the fault intensity evaluation index is greater than or equal to the third limit value, it is classified as a high intensity fault.

[0020] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0021] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A petrochemical equipment fault intensity evaluation method based on AI, characterized in that: The following steps are involved: Collect multi-dimensional data of the entire life cycle of petrochemical equipment, extract fault data from the multi-dimensional data, and build a fault database; Through the degree of fault occurrence, fault severity and fault impact, the fault database is dynamically evaluated for fault intensity; Among them, the degree of fault occurrence: input the relevant data of fault occurrence into the fault occurrence prediction model to obtain the predicted value of fault occurrence ratio; use the moving average method to predict the density of fault occurrence according to the number of fault occurrences; combine the predicted value of fault occurrence ratio and the predicted fault occurrence density to evaluate the degree of fault occurrence; Fault severity: Perform similarity assessment on fault-related data and historical fault events to obtain similar fault events used to assess fault severity; Fault impact degree: Based on the fault propagation path, a set of petrochemical equipment nodes is obtained to evaluate the fault impact degree; Fault-related data: Extract data features based on the fault database and use the random forest algorithm to screen out data features with high accuracy.

2. The AI-based petrochemical equipment fault intensity evaluation method according to claim 1 is characterized in that: The process of calculating the fault intensity evaluation index and classifying the fault intensity level is as follows: The degree of fault occurrence, the degree of fault severity and the degree of fault impact are weighted to obtain the fault intensity evaluation index.

3. The AI-based petrochemical equipment fault intensity evaluation method according to claim 1 is characterized in that: The process of obtaining the degree of the fault occurrence is as follows: Extract data from the fault database, integrate them to obtain feature matrix and target vector, and filter out data features related to fault occurrence in the feature matrix; Set the time window, based on the fault occurrence prediction model, input the relevant data features of the fault occurrence, and calculate the predicted value of the fault occurrence ratio; Count the density of faults in each time window, use the moving average method to predict the density of faults in the next time window, and perform normalization processing; The predicted value of the fault occurrence ratio and the predicted value of the fault occurrence density are weighted to obtain the degree of fault occurrence.

4. The AI-based petrochemical equipment fault intensity evaluation method according to claim 3 is characterized in that: The process of obtaining the predicted value of the fault occurrence ratio is as follows: Inputting the real-time acquired fault occurrence related data features into the fault occurrence prediction model to obtain the fault occurrence prediction value; The statistical output value is the number of faults that occurred, which is calculated by ratio with the total number of output values ​​in the time window to obtain the predicted value of the fault occurrence ratio.

5. The AI-based petrochemical equipment fault intensity evaluation method according to claim 3 is characterized in that: The process of screening the data features related to the occurrence of faults in the feature matrix is ​​as follows: Train the random forest model through the feature matrix and target vector; for any feature data; Get the out-of-bag data corresponding to each decision tree, input it into the trained random forest model for prediction, and get the predicted values. By comparing these predicted values ​​with the actual values ​​of the out-of-bag data, calculate the prediction accuracy of the original out-of-bag data and mark it as the original accuracy. Use the trained random forest model to predict the out-of-bag data after the feature values ​​are randomly sorted to obtain new predicted values. Compare the new predicted values ​​with the true values, calculate the new out-of-bag data prediction accuracy, and mark it as the new accuracy. Calculate the difference between the original accuracy and the new accuracy to obtain the accuracy deviation value; The accuracy deviation values ​​corresponding to all features are ranked in descending order, and the fault features before the ranking percentage threshold are extracted and marked as important features.

6. The AI-based petrochemical equipment fault intensity evaluation method according to claim 2 is characterized in that: The process of obtaining the severity of the fault is as follows: Obtain historical fault event data and real-time fault event data, based on the combination of the real-time fault event and any historical fault event data; Calculate the similarity value of each combination, and extract historical fault events that are greater than the similarity limit as similar fault events; The similarity values ​​corresponding to similar fault events are used as weights, and the weighted average is calculated to obtain the severity of the fault.

7. The AI-based petrochemical equipment fault intensity evaluation method according to claim 6 is characterized in that: The process of obtaining the similarity value is as follows: Based on the combination of real-time fault events and any historical fault event data, the corresponding fault occurrence-related data features are extracted respectively; The similarity value of each combination is calculated by the cosine similarity method.

8. The AI-based petrochemical equipment fault intensity evaluation method according to claim 1 is characterized in that: The process of obtaining the degree of the fault impact is as follows: Set the nodes and edges of the petrochemical equipment related graph respectively; When a petrochemical equipment fails, the propagation process of the fault in the petrochemical equipment related graph is simulated to obtain and construct the set of petrochemical equipment nodes affected during the fault propagation process; The number of nodes in the petrochemical equipment node set is counted, and the ratio is calculated with the total number of petrochemical equipment. Then it is multiplied by the normalized value of the sum of the weights of all edges on the fault propagation path to obtain the fault impact evaluation coefficient.

9. The AI-based petrochemical equipment fault intensity evaluation method according to claim 1 is characterized in that: The process of evaluating the fault intensity is as follows: If the fault intensity evaluation index is less than the first limit value, it is classified as a low intensity fault; If the fault intensity evaluation index is greater than or equal to the first limit value and if the fault intensity evaluation index is less than the second limit value, it is classified as a medium-low intensity fault; If the fault intensity evaluation index is greater than or equal to the second limit value and if the fault intensity evaluation index is less than the third limit value, it is determined to be a medium-high intensity fault; If the fault intensity evaluation index is greater than or equal to the third limit value, it is classified as a high intensity fault.

10. An AI-based petrochemical equipment fault intensity evaluation system, characterized in that: The system is used to execute the method described in any one of claims 1 to 9, and the system comprises: Fault data construction module: collects multi-dimensional data of the entire life cycle of petrochemical equipment, extracts fault data from the multi-dimensional data, and builds a fault database; Fault intensity evaluation module: It dynamically evaluates the fault intensity of the fault database based on the occurrence, severity and impact of the fault. Among them, the degree of fault occurrence: input the relevant data of fault occurrence into the fault occurrence prediction model to obtain the predicted value of fault occurrence ratio; use the moving average method to predict the density of fault occurrence according to the number of fault occurrences; combine the predicted value of fault occurrence ratio and the predicted fault occurrence density to evaluate the degree of fault occurrence; Fault severity: Perform similarity assessment on fault-related data and historical fault events to obtain similar fault events used to assess fault severity; Fault impact degree: Based on the fault propagation path, a set of petrochemical equipment nodes is obtained to evaluate the fault impact degree; Fault-related data: Extract data features based on the fault database and use the random forest algorithm to screen out data features with high accuracy.

Images