A training method and system based on model fusion of petrochemical equipment

Abstract

The present invention belongs to the technical field of deep integration of the petrochemical industry. The present invention provides a training method and system based on model fusion of petrochemical equipment, including: identifying a diagnostic model that meets the requirements of petrochemical equipment fault diagnosis, obtaining a unit parameter set through cluster analysis and similar division and integration of different operating parameters of the petrochemical equipment when it is currently faulty, and determining the optimal diagnostic model corresponding to the unit parameter set; combining the training and verification performance of the optimal diagnostic model on different unit parameter sets, extracting and integrating the unit parameter groups in different unit parameter sets to obtain a training database of the optimal diagnostic model; training and verifying the optimal diagnostic model, determining the training and verification weights of each optimal diagnostic model, and fusing all the optimal diagnostic models in parallel to obtain a target diagnostic model; taking the operating parameters of the petrochemical equipment when it is currently faulty as input, and outputting the fault index coefficient of the petrochemical equipment, thereby improving the accuracy of petrochemical equipment fault diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of deep integration technology in the petrochemical industry, and specifically relates to a training method and system for model fusion based on petrochemical equipment. Background Art

[0002] With the rapid development of information technology, data-driven fault diagnosis methods have become a research hotspot. Among them, the use of machine learning and deep learning algorithms to construct diagnostic models for petrochemical equipment fault diagnosis has shown great potential. However, in practical applications, existing diagnostic model construction and application methods still have many limitations. First, during the model training phase, most simply utilize a subset of the equipment's operating parameters from its historical operating cycles, failing to fully explore and utilize the rich operating data of the equipment at different fault moments. This results in the model's inability to comprehensively and deeply learn the operating characteristics and fault signatures of the petrochemical equipment. This leads to insufficient generalization and difficulty ensuring diagnostic accuracy when faced with complex and changing real-world fault scenarios. Second, different fault severity levels often correspond to abnormal variations in different operating parameters of petrochemical equipment. Existing methods often fail to carefully classify and analyze these operating parameters, instead lumping all parameters together for model training. This makes it difficult for the model to accurately capture key parameter information that is highly correlated with a specific fault severity, further limiting the improvement of the model's diagnostic performance.

[0003] To this end, the present invention provides a training method and system for model fusion based on petrochemical equipment. Summary of the Invention

[0004] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a training method based on model fusion of petrochemical equipment, comprising the following steps:

[0006] Using the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, the models in the existing model library are trained and verified to obtain a diagnostic model library;

[0007] Through cluster analysis of different operating parameters of petrochemical equipment during current faults, the basic parameter group is split into unit parameter groups. The unit parameter groups are then divided and integrated according to the type of operating parameters to obtain unit parameter sets. The diagnostic models in the diagnostic model library are trained and verified to determine the optimal diagnostic model.

[0008] Combining the training and verification performance of the optimal diagnostic model on different unit parameter sets, the unit parameter groups in different unit parameter sets are extracted and integrated to obtain the training database of the optimal diagnostic model;

[0009] Through training and verification of the training database, the training and verification weights of each optimal diagnostic model are determined, and all the optimal diagnostic models are integrated in parallel to obtain the target diagnostic model;

[0010] The operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model, and the fault index coefficient of the petrochemical equipment is output.

[0011] Furthermore, the diagnostic model library is obtained in the following manner:

[0012] The model is trained and validated using the basic parameter sets of petrochemical equipment at different fault moments during the historical operation cycle;

[0013] By comparing the errors between the fault index coefficients output by the model verification and the fault index coefficients contained in the basic parameter group, the basic parameter group is marked as a verification matching group and a non-verification matching group, and the verification matching features and verification deviation features are obtained through processing and analysis. The deviation analysis of the verification matching features and verification deviation features is performed to obtain the verification reflection value;

[0014] Models with verification response values ​​less than the verification response threshold are marked as diagnostic models, and the models are summarized to obtain a diagnostic model library.

[0015] Furthermore, the verification matching feature and the verification deviation feature are obtained in the following manner:

[0016] Statistically calculate the ratio of the number of verification matching groups in the basic parameter group to obtain verification matching features;

[0017] The error between the fault index coefficient of the model verification output that is not within the preset allowable error range and the fault index coefficient contained in the basic parameter group is marked as the fault index error. The absolute deviation ratio between the fault index error and the nearest endpoint value of the preset allowable error range is calculated to obtain the out-of-bounds error ratio of the non-verification matching group. The out-of-bounds error ratios of all non-verification matching groups are averaged to obtain the verification deviation characteristics.

[0018] Furthermore, the unit parameter set is obtained in the following manner:

[0019] The different operating parameters of petrochemical equipment during the current fault are obtained, and cluster analysis is performed using a hierarchical clustering algorithm. The operating parameters that are classified into one category after cluster analysis are split from the basic parameter group and combined to form unit parameter groups. Unit parameter groups with the same operating parameter type are integrated into unit parameter sets.

[0020] Furthermore, the optimal diagnostic model is determined as follows:

[0021] The unit parameter groups contained in the unit parameter set are used to train and verify each diagnostic model in the diagnostic model library. The fault index coefficients output by the diagnostic model are compared with the fault index coefficients contained in the unit parameter groups for error calculation and analysis. The unit parameter groups are marked as matching unit parameter groups and non-matching unit parameter groups. The verification compliance features and verification non-compliance features are obtained through processing and analysis. The verification compliance value is obtained through deviation analysis of the verification compliance features and verification non-compliance features.

[0022] Based on any unit parameter set, the diagnostic model corresponding to the maximum verification compliance value of the unit parameter set when performing training and verification tests on each diagnostic model is obtained as the optimal diagnostic model corresponding to the unit parameter set.

[0023] Furthermore, the verification of compliance features and verification of non-compliance features are obtained in the following manner:

[0024] The number ratio of the matching unit parameter group in the unit parameter set is counted to obtain verification that the characteristics are met;

[0025] The error between the fault index coefficient of the diagnostic model verification output that is not within the preset error range and the fault index coefficient contained in the unit parameter group is marked as a non-matching index error. The absolute deviation ratio between the non-matching index error and the nearest adjacent preset error range endpoint value is calculated to obtain the out-of-bounds error ratio of the non-matching unit parameter group. The out-of-bounds error ratios of all non-matching unit parameter groups are averaged to obtain the verification non-conformity feature.

[0026] Furthermore, the process of acquiring the training database is as follows:

[0027] Obtain the verification compliance features of the optimal diagnostic model after training, verification and testing on different unit parameter sets, and sum them up to obtain the total verification compliance value. Calculate the ratio between the verification compliance features and the total verification compliance value to obtain the extraction ratio of the unit parameter group corresponding to the unit parameter set. Extract and integrate the unit parameter groups in each unit parameter set according to the extraction ratio of the unit parameter group corresponding to the unit parameter set to obtain the training database of the optimal diagnostic model.

[0028] Furthermore, the method of determining the training and verification weights of the optimal diagnostic model is:

[0029] Based on any optimal diagnosis model, the fault index coefficient output by the optimal diagnosis model is obtained, and an error comparison analysis is performed with the fault index coefficient contained in the unit parameter group to identify the adaptive unit parameter group in the unit parameter group;

[0030] Count the number ratio of adaptive unit parameter groups in the training database to obtain the verified adaptive features;

[0031] The ratio of the validation adaptation feature to the total validation adaptation value is calculated to obtain the training validation weight of the optimal diagnostic model;

[0032] Among them, the total value of verified adaptation is the sum of all verified adaptation features;

[0033] Each optimal diagnostic model is combined with the corresponding training and verification weights and then added together to obtain the target diagnostic model.

[0034] Furthermore, the method of outputting the fault index coefficient of the petrochemical equipment is:

[0035] The operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model, and the output is obtained through each optimal diagnosis model in the target diagnosis model. The output results of each optimal diagnosis model are combined with the corresponding weights to output the fault index coefficient of the petrochemical equipment.

[0036] A training system based on model fusion for petrochemical equipment, comprising:

[0037] Diagnostic model library acquisition module: uses the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle to train and verify the models in the existing model library to obtain the diagnostic model library;

[0038] Optimal diagnostic model identification module: This module clusters the different operating parameters of petrochemical equipment during current faults, splits the basic parameter groups into unit parameter groups, and then uses the unit parameter sets obtained by classifying and integrating the unit parameter groups according to the operating parameter types to train and verify the various diagnostic models in the diagnostic model library to determine the optimal diagnostic model.

[0039] Training database construction module: Combined with the training and verification performance of the optimal diagnostic model on different unit parameter sets, the unit parameter groups in different unit parameter sets are extracted and integrated to obtain the training database of the optimal diagnostic model;

[0040] Target model acquisition module: Determine the training and verification weights of each optimal diagnostic model through training and verification of the training database, and integrate all the optimal diagnostic models in parallel to obtain the target diagnostic model;

[0041] Fault diagnosis output module: takes the operating parameters of the petrochemical equipment at the time of the current fault as the input of the target diagnosis model and outputs the fault index coefficient of the petrochemical equipment.

[0042] The beneficial effects of the present invention are as follows: obtaining the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, and training and verifying the models in the existing model library, identifying the diagnostic models that meet the requirements of petrochemical equipment fault diagnosis, and integrating them to obtain the diagnostic model library; through cluster analysis of different operating parameters of the petrochemical equipment at the time of current fault, splitting the basic parameter groups to obtain unit parameter groups, and classifying and integrating the unit parameter groups according to the type of petrochemical equipment operating parameters to obtain unit parameter sets, using the unit parameter sets to train and verify each diagnostic model in the diagnostic model library, and determining the optimal diagnostic model corresponding to the unit parameter set; combining the training and verification performance of the optimal diagnostic model on different unit parameter sets, determining the extraction ratio of unit parameter groups in different unit parameter sets, The unit parameter groups in different unit parameter sets are extracted and integrated according to the extraction ratio to obtain a training database for the optimal diagnostic model; the optimal diagnostic model is trained and verified using the training database to determine the training and verification weights of each optimal diagnostic model, and all the optimal diagnostic models are fused in parallel according to the training and verification weights to obtain a target diagnostic model; the operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnostic model, and the fault index coefficient of the petrochemical equipment is output. While the model of the present invention can comprehensively train and learn the operating data of the petrochemical equipment, it can also conduct more in-depth training and learn the operating data of the petrochemical equipment that is more suitable for the application, thereby improving the accuracy of the diagnostic model, and performing fault diagnosis on the petrochemical equipment by integrating multiple models in parallel, thereby improving the accuracy of the fault diagnosis of the petrochemical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be further described below with reference to the accompanying drawings.

[0044] Figure 1 This is a flowchart of the steps of a training method for model fusion based on petrochemical equipment according to an embodiment of the present invention;

[0045] Figure 2 This is a module diagram of a training system for model fusion based on petrochemical equipment described in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0047] Example 1: Please refer to Figure 1 As shown, a training method and system for model fusion based on petrochemical equipment according to an embodiment of the present invention includes the following steps:

[0048] Step 1: Obtain the basic parameter set of petrochemical equipment at different fault moments during the historical operation cycle, train and verify the models in the existing model library, identify diagnostic models that meet the requirements of petrochemical equipment fault diagnosis, and integrate them to obtain a diagnostic model library;

[0049] In step 1, the basic parameter group includes operating data and fault index coefficients of the petrochemical equipment at different fault moments during a historical operating cycle, wherein the operating data includes but is not limited to the temperature, pressure, flow rate, speed, etc. of the petrochemical equipment, wherein the fault index coefficient may be the efficiency drop coefficient, product variation coefficient, fatigue damage coefficient, etc. of the petrochemical equipment;

[0050] It should be noted that each petrochemical equipment has a basic parameter group at each fault moment during its historical operation cycle. Each basic parameter group consists of the operating parameters and fault index coefficients of the petrochemical equipment at the same fault moment. It should also be noted that the basic parameter group is obtained from the operation record report of the petrochemical equipment.

[0051] In step 1, the identification method of the diagnostic model that meets the requirements of petrochemical equipment fault diagnosis is as follows:

[0052] Based on any model in the existing model library;

[0053] The model is trained and verified using the basic parameter set of petrochemical equipment at different fault moments during its historical operation cycle. The fault index coefficients output by the model verification are compared with the fault index coefficients contained in the basic parameter set.

[0054] If the error between the fault index coefficient output by the model verification and the fault index coefficient contained in the basic parameter group is within the preset allowable error range, the basic parameter group is marked as a verification matching group;

[0055] If the error between the fault index coefficient output by the model verification and the fault index coefficient contained in the basic parameter group is not within the preset allowable error range, the basic parameter group is marked as a non-verification matching group;

[0056] After the model is trained and verified using the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, the proportion of the verification matching group in the basic parameter group is counted to obtain the verification matching features;

[0057] If the error between the fault indicator coefficient output by the model verification and the fault indicator coefficient contained in the basic parameter group is not within the preset allowable error range, the error between the fault indicator coefficient output by the model verification and the fault indicator coefficient contained in the basic parameter group is marked as the fault indicator error, and the deviation ratio between the fault indicator error and the nearest endpoint value of the preset allowable error range is calculated, and the absolute value is taken to obtain the out-of-bounds error ratio of the non-verification matching group, and the out-of-bounds error ratios of all non-verification matching groups are averaged to obtain the verification deviation feature;

[0058] For example, assuming that the fault indicator error is xi, the preset allowable error range is [xz-xp], and assuming that the fault indicator error xi is not within the preset allowable error range [xz-xp], and xi>xp, then the out-of-bounds error ratio WB is:

[0059]

[0060] Perform difference processing on the verification matching feature and the verification deviation feature to obtain the verification reflection value;

[0061] In some preferred embodiments, the verification response value is compared to a verification response threshold;

[0062] If the verification response value is greater than or equal to the verification response threshold, no processing is performed;

[0063] If the validation response value is less than the validation response threshold, the model is marked as a diagnostic model;

[0064] Integrate all diagnostic models to obtain a diagnostic model library;

[0065] It should be noted that the physical meaning of the verification reflection value is that the verification reflection value is obtained by differentiating the verification matching feature and the verification deviation feature. The verification matching feature represents the proportion of basic parameter groups that meet the model verification requirements. The larger the proportion, the higher the verification pass rate of the model. The verification deviation feature reflects the verification pass deviation of the model for basic parameter groups that fail the overall verification. The larger the verification pass deviation, the worse the model's adaptability to the verification of the basic parameter groups. The verification reflection value comprehensively reflects the verification performance of the model.

[0066] Step 2: Through cluster analysis of different operating parameters of petrochemical equipment during current faults, the basic parameter group is split into unit parameter groups. The unit parameter groups are then classified and integrated according to the type of petrochemical equipment operating parameters to obtain unit parameter sets. The unit parameter sets are used to train, verify and test each diagnostic model in the diagnostic model library to determine the optimal diagnostic model corresponding to the unit parameter set.

[0067] In step 2, the unit parameter group is obtained as follows:

[0068] Obtain different operating parameters of petrochemical equipment during the current fault, perform cluster analysis using a hierarchical clustering algorithm, split the operating parameters that are clustered together after cluster analysis from the basic parameter group, and combine them to form a unit parameter group;

[0069] For example, the basic parameter group includes operating parameters such as temperature, pressure, flow, and speed. If the clustering result shows that temperature and pressure are the same type of operating parameters, then temperature and pressure are separated from the basic parameter group to form a unit parameter group. The unit parameter group also includes a fault index coefficient, and its fault index coefficient is the fault index coefficient of the separated basic parameter group.

[0070] Among them, the process of cluster analysis using the hierarchical clustering algorithm is:

[0071] Initially, the operating parameters were normalized, and after the processing, each operating parameter was considered as a separate cluster;

[0072] Calculate the similarity between all clusters and merge the two clusters with the highest similarity into a new cluster. The similarity calculation method can be the single linkage algorithm (nearest neighbor algorithm), the full linkage algorithm (furthest neighbor algorithm), or the average linkage algorithm.

[0073] Repeat the above steps until all the operating parameters are merged into one cluster, and the clustering is completed;

[0074] In step 2, the unit parameter set is obtained as follows:

[0075] Integrate unit parameter groups with the same operating parameter type into unit parameter sets;

[0076] For example, all unit parameter groups formed by temperature and pressure combinations are summarized and integrated into unit parameter sets;

[0077] In step 2, the process of determining the optimal diagnostic model corresponding to the unit parameter set is as follows:

[0078] The unit parameter groups contained in the unit parameter set are used to train, verify and test each diagnostic model in the diagnostic model library. Based on any diagnostic model, the fault index coefficient output by the diagnostic model is obtained, and the error is calculated with the fault index coefficient contained in the unit parameter group. If the error is within the preset error range, the unit parameter group is marked as a matching unit parameter group. Conversely, if the error is not within the preset error range, the unit parameter group is marked as a non-matching unit parameter group.

[0079] The number ratio of the matching unit parameter group in the unit parameter set is counted to obtain verification that the characteristics are met;

[0080] The error between the fault index coefficient of the diagnostic model verification output that is not within the preset error range and the fault index coefficient contained in the unit parameter group is marked as a non-matching index error. The deviation ratio between the non-matching index error and the nearest endpoint value of the preset error range is calculated, and the absolute value is taken to obtain the out-of-bounds error ratio of the non-matching unit parameter group. The out-of-bounds error ratios of all non-matching unit parameter groups are averaged to obtain the verification non-conformity feature.

[0081] Calculate the deviation between the verification compliance feature and the verification non-compliance feature to obtain the verification compliance value;

[0082] Based on any unit parameter set, the diagnostic model corresponding to the maximum verification compliance value of the unit parameter set during the training and verification tests of each diagnostic model is obtained as the optimal diagnostic model corresponding to the unit parameter set;

[0083] For example, it is assumed that the existing diagnostic models include: diagnostic model 1, diagnostic model 2, diagnostic model 3...diagnostic model n, where n is the number of diagnostic models, and the diagnostic models are trained, validated and tested using the unit parameter set, wherein there is a validation compliance value between the unit parameter set and each diagnostic model, and the diagnostic model corresponding to the maximum validation compliance value is selected as the optimal diagnostic model of the unit parameter set;

[0084] It should be noted that the calculation and analysis principles for verifying the compliance value and verifying the response value are the same, and both are verification training performances of the reaction model;

[0085] Step 3: Combine the training and verification performance of the optimal diagnostic model on different unit parameter sets to determine the extraction ratio of unit parameter groups in different unit parameter sets. Extract and integrate the unit parameter groups in different unit parameter sets according to the extraction ratio to obtain the training database of the optimal diagnostic model.

[0086] In step 3, the process of obtaining the training database is as follows:

[0087] Based on any optimal diagnostic model;

[0088] Obtain the verification compliance features of the optimal diagnostic model after training, verification and testing of different unit parameter sets, and sum them to obtain the total verification compliance value. Calculate the ratio between the verification compliance features and the total verification compliance value to obtain the extraction ratio of the unit parameter group corresponding to the unit parameter set. Extract and integrate the unit parameter groups in each unit parameter set according to the extraction ratio of the unit parameter group corresponding to the unit parameter set to obtain the training database of the optimal diagnostic model.

[0089] For example, assuming that the verification compliance features of the optimal diagnostic model after training, verification and testing of different unit parameter sets are YZ1, YZ2, YZ3...YZb, where YZb represents the verification compliance feature of the optimal diagnostic model after training, verification and testing of the bth unit parameter set, and b represents the number of unit parameter sets, then the extraction ratio cqi of the unit parameter group corresponding to the i-th unit parameter set is:

[0090]

[0091] Among them, YZi represents the verification compliance characteristics after training, validation and testing of the parameter set of the i-th unit;

[0092] Assuming that there are three unit parameter sets, and the corresponding extraction ratios of unit parameter groups are 70%, 20%, and 10%, then the unit parameter groups are extracted from the three unit parameter sets and integrated. The training database of the optimal diagnostic model has the extracted unit parameter groups accounting for 70%, 20%, and 10% of the training database.

[0093] It can be understood that by combining the verification and compliance characteristics of the optimal diagnostic model on different unit parameter sets, determining the extraction ratio of unit parameter groups in different unit parameter sets, and extracting and integrating the unit parameter groups in different unit parameter sets according to the extraction ratio, a training database of the optimal diagnostic model is obtained. The purpose is to enable the diagnostic model to comprehensively train and learn the operating data of petrochemical equipment while more deeply training and learning the operating data of petrochemical equipment that is more suitable, thereby improving the accuracy of the diagnostic model;

[0094] Step 4: Use the training database to train and verify the optimal diagnostic model, determine the training and verification weights of each optimal diagnostic model, and fuse all the optimal diagnostic models in parallel according to the training and verification weights to obtain the target diagnostic model;

[0095] In step 4, the process of obtaining the target diagnostic model is as follows:

[0096] The optimal diagnostic model is trained and verified using the unit parameter groups contained in the training database;

[0097] Based on any optimal diagnostic model, the fault index coefficient output by the optimal diagnostic model is obtained, and the error is calculated with the fault index coefficient contained in the unit parameter group. If the error is within the preset error range, the unit parameter group is marked as an adaptive unit parameter group. Conversely, if the error is not within the preset error range, the unit parameter group is marked as a non-adaptive unit parameter group.

[0098] Count the number ratio of adaptive unit parameter groups in the training database to obtain the verified adaptive features;

[0099] The training and validation weights of the optimal diagnostic model are obtained by calculating the ratio of the validation adaptation feature to the total validation adaptation value;

[0100] Among them, the total value of verified adaptation is the sum of all verified adaptation features;

[0101] Each optimal diagnostic model is combined with the corresponding training and verification weights and then added together to obtain the target diagnostic model;

[0102] For example, assuming that the optimal diagnostic models include: optimal diagnostic model 1, optimal diagnostic model 2, optimal diagnostic model 3...optimal diagnostic model z, where z represents the number of optimal diagnostic models, and the training and verification weights corresponding to the optimal diagnostic models are q1, q2, q3...qz, where qz represents the training and verification weight corresponding to the zth optimal diagnostic model, then the target diagnostic model is:

[0103]

[0104] Among them, q1+q2+......+qz=1;

[0105] It can be understood that by integrating all the optimal diagnostic models in parallel according to the training and verification weights to obtain the target diagnostic model, different optimal diagnostic models have outstanding performance in the unit parameter groups corresponding to different operating data and have strong adaptability. Through parallel fusion, the performance of each model in its respective advantage field can be integrated to comprehensively improve the accuracy of the target diagnostic model in the overall data distribution;

[0106] Step 5: Use the operating parameters of the petrochemical equipment at the time of the current fault as the input of the target diagnosis model and output the fault index coefficient of the petrochemical equipment;

[0107] In step 5, the operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model. The output of each optimal diagnosis model in the target diagnosis model is combined with the corresponding weight to output the fault index coefficient of the petrochemical equipment.

[0108] For example, assuming that the fault index coefficients output by optimal diagnostic model 1, optimal diagnostic model 2, optimal diagnostic model 3, ... optimal diagnostic model z are gc1, gc2, gc3, ... gcz, respectively, where gcz is the fault index coefficient output by the zth optimal diagnostic model; then the fault index coefficient ZGX of the petrochemical equipment output by the final target diagnostic model is:

[0109]

[0110] The technical solution of the embodiment of the present invention is as follows: obtaining the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, and training and verifying the models in the existing model library, identifying the diagnostic models that meet the requirements of petrochemical equipment fault diagnosis, and integrating them to obtain the diagnostic model library; through cluster analysis of different operating parameters of the petrochemical equipment at the time of the current fault, splitting the basic parameter groups to obtain unit parameter groups, and classifying and integrating the unit parameter groups according to the type of petrochemical equipment operating parameters to obtain unit parameter sets, using the unit parameter sets to train and verify the various diagnostic models in the diagnostic model library, and determine the optimal diagnostic model corresponding to the unit parameter set; combining the training and verification performance of the optimal diagnostic model on different unit parameter sets, determining the extraction ratio of unit parameter groups in different unit parameter sets , extract and integrate the unit parameter groups in different unit parameter sets according to the extraction ratio to obtain the training database of the optimal diagnostic model; use the training database to train and verify the optimal diagnostic model, determine the training and verification weights of each optimal diagnostic model, and fuse all the optimal diagnostic models in parallel according to the training and verification weights to obtain the target diagnostic model; use the operating parameters of the petrochemical equipment at the time of the current fault as the input of the target diagnostic model, and output the fault index coefficient of the petrochemical equipment. While the model of the present invention can comprehensively train and learn the operating data of the petrochemical equipment, it can also conduct more in-depth training and learn the operating data of the petrochemical equipment that is more suitable for the application, thereby improving the accuracy of the diagnostic model, and performing fault diagnosis on the petrochemical equipment by integrating multiple models in parallel, thereby improving the accuracy of the fault diagnosis of the petrochemical equipment.

[0111] Example 2: A training system for model fusion based on petrochemical equipment, comprising the following modules:

[0112] Diagnostic model library acquisition module: This module obtains the basic parameter groups of petrochemical equipment at different fault moments during the historical operation cycle, trains and verifies the models in the existing model library, identifies diagnostic models that meet the requirements for petrochemical equipment fault diagnosis, and integrates them to form a diagnostic model library.

[0113] Based on any model in the existing model library;

[0114] The model is trained and verified using the basic parameter set of petrochemical equipment at different fault moments during its historical operation cycle. The fault index coefficients output by the model verification are compared with the fault index coefficients contained in the basic parameter set.

[0115] If the error between the fault index coefficient output by the model verification and the fault index coefficient contained in the basic parameter group is within the preset allowable error range, the basic parameter group is marked as a verification matching group;

[0116] If the error between the fault index coefficient output by the model verification and the fault index coefficient contained in the basic parameter group is not within the preset allowable error range, the basic parameter group is marked as a non-verification matching group;

[0117] After the model is trained and verified using the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, the proportion of the verification matching group in the basic parameter group is counted to obtain the verification matching features;

[0118] If the error between the fault indicator coefficient output by the model verification and the fault indicator coefficient contained in the basic parameter group is not within the preset allowable error range, the error between the fault indicator coefficient output by the model verification and the fault indicator coefficient contained in the basic parameter group is marked as the fault indicator error, and the deviation ratio between the fault indicator error and the nearest endpoint value of the preset allowable error range is calculated, and the absolute value is taken to obtain the out-of-bounds error ratio of the non-verification matching group, and the out-of-bounds error ratios of all non-verification matching groups are averaged to obtain the verification deviation feature;

[0119] Perform difference processing on the verification matching feature and the verification deviation feature to obtain the verification reflection value;

[0120] If the validation response value is less than the validation response threshold, the model is marked as a diagnostic model;

[0121] Optimal diagnostic model identification module: This module clusters the different operating parameters of petrochemical equipment during current faults, splits the basic parameter groups into unit parameter groups, and then classifies and integrates the unit parameter groups according to the type of petrochemical equipment operating parameters to obtain unit parameter sets. The unit parameter sets are then used to train, validate, and test each diagnostic model in the diagnostic model library to determine the optimal diagnostic model corresponding to the unit parameter set.

[0122] Obtain different operating parameters of petrochemical equipment during the current fault, perform cluster analysis using a hierarchical clustering algorithm, split the operating parameters that are clustered together after cluster analysis from the basic parameter group, and combine them to form a unit parameter group;

[0123] Among them, the process of cluster analysis using the hierarchical clustering algorithm is:

[0124] Initially, the operating parameters were normalized, and after the processing, each operating parameter was considered as a separate cluster;

[0125] Calculate the similarity between all clusters and merge the two clusters with the highest similarity into a new cluster. The similarity calculation method can be the single linkage algorithm (nearest neighbor algorithm), the full linkage algorithm (furthest neighbor algorithm), or the average linkage algorithm.

[0126] Repeat the above steps until all the operating parameters are merged into one cluster, and the clustering is completed;

[0127] Integrate unit parameter groups with the same operating parameter type into unit parameter sets;

[0128] The unit parameter groups contained in the unit parameter set are used to train, verify and test each diagnostic model in the diagnostic model library. Based on any diagnostic model, the fault index coefficient output by the diagnostic model is obtained, and the error is calculated with the fault index coefficient contained in the unit parameter group. If the error is within the preset error range, the unit parameter group is marked as a matching unit parameter group. Conversely, if the error is not within the preset error range, the unit parameter group is marked as a non-matching unit parameter group.

[0129] The number ratio of the matching unit parameter group in the unit parameter set is counted to obtain verification that the characteristics are met;

[0130] The error between the fault index coefficient output by the diagnostic model verification and the fault index coefficient contained in the unit parameter group is marked as a non-matching index error. The deviation ratio between the non-matching index error and the nearest adjacent preset error range endpoint value is calculated and the absolute value is taken to obtain the out-of-bounds error ratio of the non-matching unit parameter group. The out-of-bounds error ratios of all non-matching unit parameter groups are averaged to obtain the verification non-conformity feature.

[0131] Calculate the deviation between the verification compliance feature and the verification non-compliance feature to obtain the verification compliance value;

[0132] Based on any unit parameter set, the diagnostic model corresponding to the maximum verification compliance value of the unit parameter set during the training and verification tests of each diagnostic model is obtained as the optimal diagnostic model corresponding to the unit parameter set;

[0133] Training database construction module: Combined with the training and verification performance of the optimal diagnostic model on different unit parameter sets, the extraction ratio of unit parameter groups in different unit parameter sets is determined. The unit parameter groups in different unit parameter sets are extracted and integrated according to the extraction ratio to obtain the training database of the optimal diagnostic model;

[0134] Based on any optimal diagnostic model;

[0135] Obtain the verification compliance features of the optimal diagnostic model after training, verification and testing of different unit parameter sets, and sum them to obtain the total verification compliance value. Calculate the ratio between the verification compliance features and the total verification compliance value to obtain the extraction ratio of the unit parameter group corresponding to the unit parameter set. Extract and integrate the unit parameter groups in each unit parameter set according to the extraction ratio of the unit parameter group corresponding to the unit parameter set to obtain the training database of the optimal diagnostic model.

[0136] Target model acquisition module: Use the training database to train and verify the optimal diagnostic model, determine the training and verification weights of each optimal diagnostic model, and fuse all the optimal diagnostic models in parallel according to the training and verification weights to obtain the target diagnostic model;

[0137] The optimal diagnostic model is trained and verified using the unit parameter groups contained in the training database;

[0138] Based on any optimal diagnostic model, the fault index coefficient output by the optimal diagnostic model is obtained, and the error is calculated with the fault index coefficient contained in the unit parameter group. If the error is within the preset error range, the unit parameter group is marked as an adaptive unit parameter group. Conversely, if the error is not within the preset error range, the unit parameter group is marked as a non-adaptive unit parameter group.

[0139] Count the number ratio of adaptive unit parameter groups in the training database to obtain the verified adaptive features;

[0140] The training and validation weights of the optimal diagnostic model are obtained by calculating the ratio of the validation adaptation feature to the total validation adaptation value;

[0141] Among them, the total value of verified adaptation is the sum of all verified adaptation features;

[0142] Each optimal diagnostic model is combined with the corresponding training and verification weights and then added together to obtain the target diagnostic model;

[0143] Fault diagnosis output module: takes the operating parameters of the petrochemical equipment at the time of the current fault as the input of the target diagnosis model and outputs the fault index coefficient of the petrochemical equipment;

[0144] The operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model, and the output is obtained through each optimal diagnosis model in the target diagnosis model. The output results of each optimal diagnosis model are combined with the corresponding weight to output the fault index coefficient of the petrochemical equipment.

[0145] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A training method for model fusion based on petrochemical equipment, characterized by: The following steps are involved: Using the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle, the models in the existing model library are trained and verified to obtain a diagnostic model library; Through cluster analysis of different operating parameters of petrochemical equipment during current faults, the basic parameter group is split into unit parameter groups. The unit parameter groups are then divided and integrated according to the type of operating parameters to obtain unit parameter sets. The diagnostic models in the diagnostic model library are trained and verified to determine the optimal diagnostic model. Combining the training and verification performance of the optimal diagnostic model on different unit parameter sets, the unit parameter groups in different unit parameter sets are extracted and integrated to obtain the training database of the optimal diagnostic model; Through training and verification of the training database, the training and verification weights of each optimal diagnostic model are determined, and all the optimal diagnostic models are integrated in parallel to obtain the target diagnostic model; The operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model, and the fault index coefficient of the petrochemical equipment is output; The acquisition process of the training database is as follows: Obtain the verification compliance features of the optimal diagnostic model after training, verification and testing of different unit parameter sets, and sum them to obtain the total verification compliance value. Calculate the ratio between the verification compliance features and the total verification compliance value to obtain the extraction ratio of the unit parameter group corresponding to the unit parameter set. Extract and integrate the unit parameter groups in each unit parameter set according to the extraction ratio of the unit parameter group corresponding to the unit parameter set to obtain the training database of the optimal diagnostic model. The method for determining the training and verification weights of the optimal diagnostic model is: Based on any optimal diagnosis model, the fault index coefficient output by the optimal diagnosis model is obtained, and an error comparison analysis is performed with the fault index coefficient contained in the unit parameter group to identify the adaptive unit parameter group in the unit parameter group; Count the number ratio of adaptive unit parameter groups in the training database to obtain the verified adaptive features; The ratio of the validation adaptation feature to the total validation adaptation value is calculated to obtain the training validation weight of the optimal diagnostic model; Among them, the total value of verified adaptation is the sum of all verified adaptation features; Each optimal diagnostic model is combined with the corresponding training and verification weights and then added together to obtain the target diagnostic model.

2. The training method for model fusion based on petrochemical equipment according to claim 1, characterized in that: The diagnostic model library is obtained as follows: The model is trained and validated using the basic parameter sets of petrochemical equipment at different fault moments during the historical operation cycle; By comparing the errors between the fault index coefficients output by the model verification and the fault index coefficients contained in the basic parameter group, the basic parameter group is marked as a verification matching group and a non-verification matching group, and the verification matching features and verification deviation features are obtained through processing and analysis. The deviation analysis of the verification matching features and verification deviation features is performed to obtain the verification reflection value; Models with verification reflection values ​​less than the verification reflection threshold are marked as diagnostic models, and the models are summarized to obtain a diagnostic model library.

3. The training method for model fusion based on petrochemical equipment according to claim 2, characterized in that: The verification matching feature and the verification deviation feature are obtained as follows: Statistically calculate the ratio of the number of verification matching groups in the basic parameter group to obtain verification matching features; The error between the fault index coefficient of the model verification output that is not within the preset allowable error range and the fault index coefficient contained in the basic parameter group is marked as the fault index error. The absolute deviation ratio between the fault index error and the nearest endpoint value of the preset allowable error range is calculated to obtain the out-of-bounds error ratio of the non-verification matching group. The out-of-bounds error ratios of all non-verification matching groups are averaged to obtain the verification deviation characteristics.

4. The training method for model fusion based on petrochemical equipment according to claim 1, characterized in that: The unit parameter set is obtained as follows: The different operating parameters of petrochemical equipment during the current fault are obtained, and cluster analysis is performed using a hierarchical clustering algorithm. The operating parameters that are classified into one category after cluster analysis are split from the basic parameter group and combined to form unit parameter groups. Unit parameter groups with the same operating parameter type are integrated into unit parameter sets.

5. The training method for model fusion based on petrochemical equipment according to claim 1, characterized in that: The optimal diagnostic model is determined as follows: The unit parameter groups contained in the unit parameter set are used to train and verify each diagnostic model in the diagnostic model library. The fault index coefficients output by the diagnostic model are compared with the fault index coefficients contained in the unit parameter groups for error calculation and analysis. The unit parameter groups are marked as matching unit parameter groups and non-matching unit parameter groups. The verification compliance features and verification non-compliance features are obtained through processing and analysis. The verification compliance value is obtained through deviation analysis of the verification compliance features and verification non-compliance features. Based on any unit parameter set, the diagnostic model corresponding to the maximum verification compliance value of the unit parameter set when performing training and verification tests on each diagnostic model is obtained as the optimal diagnostic model corresponding to the unit parameter set.

6. The training method for model fusion based on petrochemical equipment according to claim 5, characterized in that: The method for obtaining the verification compliance feature and the verification non-compliance feature is as follows: The number ratio of the matching unit parameter group in the unit parameter set is counted to obtain verification that the characteristics are met; The error between the fault index coefficient of the diagnostic model verification output that is not within the preset error range and the fault index coefficient contained in the unit parameter group is marked as a non-matching index error. The absolute deviation ratio between the non-matching index error and the nearest adjacent preset error range endpoint value is calculated to obtain the out-of-bounds error ratio of the non-matching unit parameter group. The out-of-bounds error ratios of all non-matching unit parameter groups are averaged to obtain the verification non-conformity feature.

7. The training method for model fusion based on petrochemical equipment according to claim 1, characterized in that: The method of outputting the fault index coefficient of the petrochemical equipment is: The operating parameters of the petrochemical equipment at the time of the current fault are used as the input of the target diagnosis model, and the output is obtained through each optimal diagnosis model in the target diagnosis model. The output results of each optimal diagnosis model are combined with the corresponding weights to output the fault index coefficient of the petrochemical equipment.

8. A training system based on model fusion of petrochemical equipment, characterized by: include: Diagnostic model library acquisition module: uses the basic parameter groups of petrochemical equipment at different fault moments in the historical operation cycle to train and verify the models in the existing model library to obtain the diagnostic model library; Optimal diagnostic model identification module: This module clusters the different operating parameters of petrochemical equipment during current faults, splits the basic parameter groups into unit parameter groups, and then uses the unit parameter sets obtained by classifying and integrating the unit parameter groups according to the operating parameter types to train and verify the various diagnostic models in the diagnostic model library to determine the optimal diagnostic model. Training database construction module: Combined with the training and verification performance of the optimal diagnostic model on different unit parameter sets, the unit parameter groups in different unit parameter sets are extracted and integrated to obtain the training database of the optimal diagnostic model; Target model acquisition module: Determine the training and verification weights of each optimal diagnostic model through training and verification of the training database, and integrate all the optimal diagnostic models in parallel to obtain the target diagnostic model; Fault diagnosis output module: takes the operating parameters of the petrochemical equipment at the time of the current fault as the input of the target diagnosis model and outputs the fault index coefficient of the petrochemical equipment; The acquisition process of the training database is as follows: Obtain the verification compliance features of the optimal diagnostic model after training, verification and testing of different unit parameter sets, and sum them to obtain the total verification compliance value. Calculate the ratio between the verification compliance features and the total verification compliance value to obtain the extraction ratio of the unit parameter group corresponding to the unit parameter set. Extract and integrate the unit parameter groups in each unit parameter set according to the extraction ratio of the unit parameter group corresponding to the unit parameter set to obtain the training database of the optimal diagnostic model. The method for determining the training and verification weights of the optimal diagnostic model is: Based on any optimal diagnosis model, the fault index coefficient output by the optimal diagnosis model is obtained, and an error comparison analysis is performed with the fault index coefficient contained in the unit parameter group to identify the adaptive unit parameter group in the unit parameter group; Count the number ratio of adaptive unit parameter groups in the training database to obtain the verified adaptive features; The ratio of the validation adaptation feature to the total validation adaptation value is calculated to obtain the training validation weight of the optimal diagnostic model; Among them, the total value of verified adaptation is the sum of all verified adaptation features; Each optimal diagnostic model is combined with the corresponding training and verification weights and then added together to obtain the target diagnostic model.

Images