An abnormal intelligent identification system and method for high-efficiency exhaust fan units based on OCSVM

Through the highly efficient exhaust fan unit intelligent abnormal identification method based on OCSVM, combined with multivariate data and clustering algorithm, the problem of low accuracy and efficiency of single signal training model in the existing technology is solved, and more accurate and efficient abnormal identification is achieved, ensuring that The safety and economic benefits of biosafety laboratories.

CN118690307BActive Publication Date: 2024-11-26CHINA ELECTRONICS SYST ENG NO 2 CONSTR
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411162621.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-26
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

When monitoring abnormal conditions of high-efficiency exhaust fan units, the existing technology mainly relies on vibration signals and cannot effectively integrate various information, which will affect the accuracy and efficiency of fault detection. The data quality has a great impact on model performance, which may lead to the The warning delay of potential failures affects safety and economic losses.

Method used

Using an efficient exhaust fan unit intelligent abnormal identification method based on OCSVM, data preprocessing and feature extraction are performed through multi-functional operation data acquisition (including pressure, vibration, current, temperature), tags are generated using DBSCAN and K-medoids clustering algorithms, combined with OCSVM The model performs abnormal identification, and uses a double-label cross-validation strategy to ensure data labeling accuracy and improve model robustness.

Benefits of technology

It significantly improves the accuracy and efficiency of abnormal identification, enhances the robustness and robustness of the model, ensures the stable operation of the exhaust unit and personnel safety, and reduces economic losses caused by equipment failure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118690307B_ABST
    Figure CN118690307B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of abnormality identification, and specifically to an intelligent abnormality identification system and method for a high-efficiency exhaust fan unit based on OCSVM, comprising a multivariate data acquisition module, a data preprocessing module, a data set construction module, a sample label analysis module, a model training module and a model determination module; the multivariate data acquisition module is used to collect multivariate operation data of an exhaust fan unit equipped with a sensor acquisition device; the data preprocessing module is used to obtain multivariate operation data obtained by the exhaust fan unit based on the sensor acquisition device in a historical monitoring period for data preprocessing; the data set construction module is used to extract features from the preprocessed multivariate operation data to form a data set; the sample label analysis module is used to mark and verify sample data in the data set, and analyze and output sample labels; the model training module is used to construct an OCSVM model through sample label data and train the OCSVM model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality identification, and in particular to an abnormality intelligent identification system and method for a high-efficiency exhaust fan unit based on OCSVM. Background Art

[0002] In the field of biosafety, the importance of biosafety laboratories is self-evident. It is not only a key line of defense to ensure the health of researchers from harmful microorganisms, viruses and other dangerous substances, but also bears the responsibility of avoiding the leakage of biological factors and protecting the safety and stability of the environment. More importantly, biosafety laboratories have effectively prevented malicious behavior through strict biosafety management measures. Therefore, ensuring the air quality inside the laboratory and reducing the concentration of pollutants inside the laboratory are crucial to maintaining the health of laboratory personnel and ensuring the smooth progress of experimental activities. Among them, the normal operation of the high-efficiency exhaust fan unit as the core part of the exhaust system plays a decisive role in personnel safety and the conduct of experiments. However, when monitoring the abnormal situation of the high-efficiency exhaust fan unit, the model is usually trained only by vibration signals for abnormality recognition, which cannot effectively integrate various information, resulting in the accuracy and efficiency of fault detection being affected; at the same time, when using machine learning algorithms to build models to identify equipment abnormalities, the quality of data directly determines the performance of the model. However, the existing anomaly recognition methods mainly focus on model construction, and rarely involve measures to ensure data quality, which directly affects the generalization ability and robustness of the model. When the model performance is poor, the anomaly recognition rate will be affected, which may lead to delayed warning of potential faults, thus failing to effectively ensure the safety of personnel and the environment, and may even cause economic losses. Summary of the invention

[0003] The object of the present invention is to provide an OCSVM-based high-efficiency exhaust fan unit abnormality intelligent identification system and method to solve the problems raised in the above-mentioned background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent identification method for abnormalities of a high-efficiency exhaust fan unit based on OCSVM, comprising the following analysis steps:

[0005] Step S100: Arrange a sensor collection device for collecting multivariate operation data on the exhaust fan unit; obtain the multivariate operation data of the exhaust fan unit obtained by the sensor collection device during the historical monitoring period for data preprocessing; and extract features from the preprocessed multivariate operation data to form a data set;

[0006] Step S200: Based on the data set, label and verify the sample data in the data set, and analyze and output the sample labels;

[0007] Step S300: construct an OCSVM model through sample label data and train the OCSVM model, verify whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to identify abnormalities of the exhaust fan unit based on the test set for the OCSVM model composed of the adjusted model parameters;

[0008] Step S400: Based on the abnormality recognition result of the OCSVM model, select the actual abnormality result recorded in the corresponding historical monitoring period, analyze the recognition result deviation value, and return to step S100 to update the data based on the recognition result deviation value until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormality recognition model finally applied to the exhaust fan unit.

[0009] Furthermore, step S100 includes the following specific steps:

[0010] Step S110: The multivariate operating data includes but is not limited to pressure, vibration, current, and temperature; the collected multivariate operating data is clustered and divided using the DBSCAN method, core data and outliers are identified, and the outliers are removed to update the multivariate operating data into multivariate valid operating data;

[0011] The DBSCAN method is used to preprocess multivariate operating data without the need to pre-specify the number of clusters. It can discover clusters of any shape and automatically mark noise points as outliers, thereby improving the computing power of data processing and the efficiency and ability of large-scale data processing, making it possible to effectively screen data when faced with multivariate operating data collected by exhaust fans.

[0012] Step S120: normalize the multivariate effective operation data, the formula is:

[0013] x0=[x-min(x)] / [max(x)-min(x)],

[0014] Generate normalized multivariate valid operating data x0; wherein x represents the multivariate valid operating data before normalization, min(x) represents the minimum value in the multivariate valid operating data, and max(x) represents the maximum value in the multivariate valid operating data;

[0015] Step S130: extract features from the multivariate effective operation data using a signal processing method, and construct a data set corresponding to data features and data feature values.

[0016] Furthermore, step S200 includes the following specific steps:

[0017] Step S210: Selecting the spatial geometric center formed by the data feature values ​​corresponding to the data features in the data set as the initial clustering center;

[0018] Step S220: starting from the initial cluster center, the iteration direction is the cluster end point, where the coordinate value of the cluster end point is derived from the b% quantile of each feature data; the iteration step value is 100%-b% of the direction vector modulus formed between the initial cluster center and the cluster end point; the iteration termination condition is that the normal data volume accounts for b% of the total data volume. If the condition is not met, the iteration continues; if the condition is met, the iteration ends;

[0019] Step S230: After the iteration, each sample is divided into each cluster, and then the samples in each cluster are marked; the normal data label is 1, and the abnormal data label is 0; the result obtained by the K-medoids clustering algorithm for the multivariate effective operation data is recorded as the first category sample label;

[0020] Step S240: performing data subset labeling on a single feature. Data subset labeling means executing step S220 on the single feature data respectively, and recording the obtained result as the second type sample label;

[0021] Step S250: cross-validate the first category sample label records and the second category sample label records, extract the same data features in the two categories of sample label records, perform an “AND” operation, and output the feature labels; when the data labels of all features are 0 and the data labels obtained by a single feature are 0, the corresponding correction feature label is output as 0; in other cases, the correction feature label is output as 1;

[0022] Step S260: The final sample label is determined by multiple calibration feature labels. The weight ratio of each calibration feature label is determined based on historical data, using the formula:

[0023] Y=k1×Y P +k2×Y V +k3×Y E +k4×Y T ;

[0024] Calculate the sample label Y, where Y P , Y V , Y E , Y T They are the correction feature labels of the four data types of pressure, vibration, current and temperature. k1, k2, k3 and k4 correspond to the weight proportions of the correction feature labels of the four data types of pressure, vibration, current and temperature respectively.

[0025] Furthermore, step S260 determines the weight ratio of each correction feature label based on historical data, including the following:

[0026] Step S261: Obtain the number of single features N contained in each calibration feature tag, and the number of times D that each single feature is correctly identified as an abnormal feature in the abnormal identification events recorded in the historical data of the corresponding monitoring period. Correct identification means that the only abnormal feature of the exhaust fan unit is determined in the abnormal identification event and the abnormality is finally determined; using the formula:

[0027] Z = a1 × N / N1 + a2 × D / D1;

[0028] Calculate the influence index Z of each calibration feature label, a1 and a2 represent the corresponding reference coefficients; N1 represents the total number of all single features contained in all calibration feature labels, and D1 represents the total number of abnormal identification events recorded in the historical data of the corresponding monitoring period;

[0029] Step S262: Generate label ratio values ​​f1:f2:...:f based on the influence index of each calibration feature label m , m represents the total number of calibration feature labels;

[0030] Step S263: Based on the label ratio value, use the formula: 100% = f1×b+f2×b+......+f m ×b;

[0031] Calculate the unit weight b of each correction feature label, and calculate the weight proportion b×f of each correction feature label based on the unit weight b j , f j ={f1,f2,......,f m}, f j Represents the label ratio value of the j-th correction feature label, j≤m.

[0032] The larger the impact index is, the more the corresponding correction feature label can be used as an identification feature in abnormal identification events, and the higher the corresponding data identification ability is.

[0033] Furthermore, the OCSVM model is constructed through sample label data and the OCSVM model is trained, including the following steps:

[0034] Divide the sample label data into training set, validation set and test set;

[0035] Based on the training set data, OCSVM is used to find an optimal hyperplane to achieve model optimization by minimizing an objective function. The optimization process is as follows:

[0036]

[0037] Where ω is the normal vector of the hyperplane, ξi is the slack variable, ρ is the distance from the hyperplane to the origin, θ is the regularization parameter used to control the trade-off between the error term and the margin, and n is the amount of data;

[0038] Constraint: The objective function is restricted by the following formula,

[0039]

[0040] where, Φ(x i ) is a function that maps data points to the feature space;

[0041] Using the Lagrange operator, the optimization objective function is transformed into: ; Using the decision function of OCSVM, we get: ;

[0042] where, if f(x) = 1, then x is regarded as a normal point, and if f(x) = -1, then x is regarded as an abnormal point.

[0043] Based on the above model, the model is trained using the training set.

[0044] Furthermore, to verify whether the OCSVM model parameters need to be adjusted, the following steps are included:

[0045] Verify the accuracy of the model based on the validation set, using the formula: L = TP / (TP + FN);

[0046] Calculate the accuracy L of the trained model, where TP represents the number of samples correctly identified as positive examples by the trained model, and FN represents the number of samples that are actually positive examples but are misidentified as negative examples by the model;

[0047] Set the accuracy threshold L0. When L < L0, adjust the model parameters, and use grid search to construct different parameter combinations to find the optimal parameter combination when the model output accuracy applied to the validation set is greater than or equal to L0 and is the maximum value, as the parameters applied to the model output.

[0048] Furthermore, step S400 includes the following specific steps:

[0049] Step S410: Obtain the effective recognition number V recorded based on the abnormal recognition results of the OCSVM model, and the recognition number V0 corresponding to the actual abnormal results. Effective recognition refers to the remaining recognition situations excluding incorrect abnormal recognition results and unrecognized results; Calculate the result deviation value G, G = V0 - V, and set the deviation value threshold G0;

[0050] Step S420: If G < G0, output the current OCSVM model as the abnormal recognition model finally applied to the exhaust fan group; if G ≥ G0, return to step S100 for data update. The data update includes increasing the monitoring cycle duration to increase the quantity of multivariate operation data or increasing the type of multivariate operation data.

[0051] The OCSVM-based high-efficiency exhaust fan unit abnormality intelligent identification system includes a multivariate data acquisition module, a data preprocessing module, a data set construction module, a sample label analysis module, a model training module, and a model determination module;

[0052] The multivariate data acquisition module is used to collect multivariate operation data of the exhaust fan unit equipped with sensor acquisition equipment;

[0053] The data preprocessing module is used to obtain the multivariate operating data obtained by the exhaust fan unit based on the sensor acquisition device during the historical monitoring period for data preprocessing;

[0054] The data set construction module is used to extract features from the preprocessed multivariate operating data to form a data set;

[0055] The sample label analysis module is used to label and verify the sample data in the data set, and analyze and output the sample labels;

[0056] The model training module is used to construct an OCSVM model through sample label data and train the OCSVM model, verify whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to identify abnormalities of the exhaust fan unit in the OCSVM model composed of the adjusted model parameters;

[0057] The model determination module is used to select the actual abnormal results recorded in the corresponding historical monitoring period based on the abnormal recognition results of the OCSVM model, analyze the deviation value of the recognition result, and return to step S100 to update the data based on the deviation value of the recognition result until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormal recognition model finally applied to the exhaust fan group.

[0058] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: in order to solve the accuracy and efficiency difficulties of the model trained with a single signal in abnormal identification, the present invention introduces multi-source data such as pressure, vibration, current, temperature, etc. as model training data to enhance model performance and comprehensively evaluate the equipment status; in order to effectively deal with the problems caused by uneven data quality on model performance, DBSCAN is used to identify and remove potential outliers. The present invention generates two types of data labels through the K-medoids clustering algorithm, obtains sample labels through weight calculation, realizes self-learning automatic labeling of sample labels, ensures the quality of training data, and further ensures the accuracy of the model; finally, the present invention proposes an efficient exhaust fan unit abnormality intelligent identification method based on OCSVM, which starts from the two key dimensions of data source and data quality, and integrates diversified data sources to train the model, thereby significantly reducing the cost of data collection. In addition, a double-label cross-validation strategy is adopted to ensure the accuracy of data labeling and significantly enhance the robustness and robustness of the model. In addition, the present invention increases the important influence of different characteristic data reflected by abnormal identification events based on historical data records in the process of data label processing to measure the adaptive weight conditions of different labels based on the corresponding environment, and no longer performs a single equal division to improve the accuracy of data label analysis; the present invention provides a more solid guarantee for the operation of biosafety laboratories by improving the abnormal identification rate of high-efficiency exhaust fan units. And through the implementation of the present invention, it is possible to ensure the stable operation of the exhaust fan unit, timely discover and handle potential abnormal situations, effectively ensure personnel safety, and reduce economic losses caused by equipment failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0060] Figure 1 It is an overall flow chart of the abnormal intelligent identification method of the high-efficiency exhaust fan unit based on OCSVM of the present invention;

[0061] Figure 2 It is a flow chart of obtaining two types of labels of the method for intelligently identifying abnormalities of high-efficiency exhaust fan units based on OCSVM of the present invention;

[0062] Figure 3 It is an output diagram of the correction feature label of the abnormal intelligent identification method of the high-efficiency exhaust fan unit based on OCSVM of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 are within the scope of protection of the present invention.

[0064] See also Figure 1-Figure 3 The present invention provides a technical solution: an intelligent identification method for abnormalities of a high-efficiency exhaust fan unit based on OCSVM, comprising the following analysis steps:

[0065] Step S100: Arrange a sensor collection device for collecting multivariate operation data on the exhaust fan unit; obtain the multivariate operation data of the exhaust fan unit obtained by the sensor collection device during the historical monitoring period for data preprocessing; and extract features from the preprocessed multivariate operation data to form a data set;

[0066] Step S200: Based on the data set, label and verify the sample data in the data set, and analyze and output the sample labels;

[0067] Step S300: construct an OCSVM model through sample label data and train the OCSVM model, verify whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to identify abnormalities of the exhaust fan unit based on the test set for the OCSVM model composed of the adjusted model parameters;

[0068] Step S400: Based on the abnormality recognition result of the OCSVM model, select the actual abnormality result recorded in the corresponding historical monitoring period, analyze the recognition result deviation value, and return to step S100 to update the data based on the recognition result deviation value until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormality recognition model finally applied to the exhaust fan unit.

[0069] Step S100 includes the following specific steps:

[0070] Step S110: The multivariate operating data includes but is not limited to pressure, vibration, current, and temperature; the collected multivariate operating data is clustered and divided using the DBSCAN method, core data and outliers are identified, and the outliers are removed to update the multivariate operating data into multivariate valid operating data;

[0071] As shown in the embodiment, using the DBSCAN method to identify outliers in multivariate operation data includes the following:

[0072] Initialize the radius and minimum density, and randomly select an unvisited data point in the multivariate run data.

[0073] Determine core data: Find the number of data points in the circle with the data point as the center and the initialization radius that exceeds the minimum density, then the data point is the core data; mark it as visited;

[0074] Find density-reachable points: For all density-reachable points of the core data, mark them as visited and add them to the same cluster;

[0075] The above process is repeated until all core data are visited and all clusters are formed, and all unvisited data points are considered outliers.

[0076] The DBSCAN method is used to preprocess multivariate operating data without the need to pre-specify the number of clusters. It can discover clusters of any shape and automatically mark noise points as outliers, thereby improving the computing power of data processing and the efficiency and ability of large-scale data processing, making it possible to effectively screen data when faced with multivariate operating data collected by exhaust fans.

[0077] Step S120: normalize the multivariate effective operation data, the formula is:

[0078] x0=[x-min(x)] / [max(x)-min(x)],

[0079] Generate normalized multivariate valid operating data x0; wherein x represents the multivariate valid operating data before normalization, min(x) represents the minimum value in the multivariate valid operating data, and max(x) represents the maximum value in the multivariate valid operating data;

[0080] Step S130: extract features from the multivariate effective operation data using a signal processing method, and construct a data set corresponding to data features and data feature values.

[0081] For example, the data features are: static pressure, dynamic pressure, ...., frequency, temperature...;

[0082] Static pressure, dynamic pressure, and total pressure are data characteristics under pressure data; velocity effective value, acceleration effective value, and acceleration envelope peak value are data characteristics under vibration data; average value, peak value, and frequency are data characteristics under current data;

[0083] The corresponding data characteristic values ​​are 2000, 1500,......, 50, 30........

[0084] Step S200 includes the following specific steps:

[0085] Step S210: Selecting the spatial geometric center formed by the data feature values ​​corresponding to the data features in the data set as the initial clustering center;

[0086] Step S220: starting from the initial cluster center, the iteration direction is the cluster end point, where the coordinate value of the cluster end point is derived from the b% quantile of each feature data; the iteration step value is 100%-b% of the direction vector modulus formed between the initial cluster center and the cluster end point; the iteration termination condition is that the normal data volume accounts for b% of the total data volume. If the condition is not met, the iteration continues; if the condition is met, the iteration ends; b is generally 95;

[0087] Step S230: After the iteration, each sample is divided into each cluster, and then the samples in each cluster are marked; the normal data label is 1, and the abnormal data label is 0; the result obtained by the K-medoids clustering algorithm for the multivariate effective operation data is recorded as the first category sample label;

[0088] Step S240: performing data subset labeling on the single feature. Data subset labeling means executing step S220 on the single feature data respectively and recording the obtained results as the second type of sample labels. The single feature data here refers to pressure, vibration, and current data.

[0089] Step S250: cross-validate the first category sample label records and the second category sample label records, extract the same data features in the two categories of sample label records, perform an “AND” operation, and output the feature labels; when the data labels of all features are 0 and the data labels obtained by a single feature are 0, the corresponding correction feature label is output as 0; in other cases, the correction feature label is output as 1;

[0090] Step S260: The final sample label is determined by multiple calibration feature labels. The weight ratio of each calibration feature label is determined based on historical data, using the formula:

[0091] Y=k1×Y P +k2×Y V +k3×Y E +k4×Y T ;

[0092] Calculate the sample label Y, where Y P , Y V , Y E , Y T They are the correction feature labels of the four data types of pressure, vibration, current and temperature. k1, k2, k3 and k4 correspond to the weight proportions of the correction feature labels of the four data types of pressure, vibration, current and temperature respectively.

[0093] Step S260 determines the weight ratio of each correction feature label based on historical data, including the following:

[0094] Step S261: Obtain the number of single features N contained in each calibration feature tag, and the number of times D that each single feature is correctly identified as an abnormal feature in the abnormal identification events recorded in the historical data of the corresponding monitoring period. Correct identification means that the only abnormal feature of the exhaust fan unit is determined in the abnormal identification event and the abnormality is finally determined; using the formula:

[0095] Z = a1 × N / N1 + a2 × D / D1;

[0096] Calculate the influence index Z of each calibration feature label, a1 and a2 represent the corresponding reference coefficients; N1 represents the total number of all single features contained in all calibration feature labels, and D1 represents the total number of abnormal identification events recorded in the historical data of the corresponding monitoring period;

[0097] Step S262: Generate label ratio values ​​f1:f2:...:f based on the influence index of each calibration feature label m , m represents the total number of calibration feature labels;

[0098] Step S263: Based on the label ratio value, use the formula: 100% = f1×b+f2×b+......+f m ×b;

[0099] Calculate the unit weight b of each correction feature label, and calculate the weight proportion b×f of each correction feature label based on the unit weight b j , f j ={f1,f2,......,f m}, f j Represents the label ratio value of the j-th correction feature label, j≤m.

[0100] The larger the impact index is, the more the corresponding correction feature label can be used as an identification feature in abnormal identification events, and the higher the corresponding data identification ability is.

[0101] Building an OCSVM model through sample label data and training the OCSVM model includes the following steps:

[0102] Divide the sample label data into training set, validation set and test set;

[0103] Based on the training set data, OCSVM is used to find an optimal hyperplane to achieve model optimization by minimizing an objective function. The optimization process is as follows:

[0104]

[0105] Where ω is the normal vector of the hyperplane, ξi is the slack variable, ρ is the distance from the hyperplane to the origin, θ is the regularization parameter used to control the trade-off between the error term and the margin, and n is the amount of data;

[0106] Constraint: The objective function is restricted by the following formula

[0107]

[0108] where Φ(x i ) is the function that maps data points to the feature space;

[0109] Using the Lagrange operator, the optimization objective function is transformed into: ; Using the decision function of OCSVM, we get: ;

[0110] where, if f(x) = 1, then x is regarded as a normal point, and if f(x) = -1, then x is regarded as an abnormal point.

[0111] Based on the above model, the model is trained using the training set.

[0112] Verify whether the OCSVM model parameters need to be adjusted, including the following steps:

[0113] Verify the accuracy of the model based on the validation set, using the formula: L = TP / (TP + FN);

[0114] Calculate the accuracy L of the trained model, where TP represents the number of samples correctly identified as positive examples by the trained model, and FN represents the number of samples that are actually positive examples but are misidentified as negative examples by the model;

[0115] Set the accuracy threshold L0. When L < L0, adjust the model parameters. Use grid search to construct different parameter combinations to find the optimal parameter combination when the model output accuracy on the validation set is greater than or equal to L0 and is the maximum value, as the parameters for model output application.

[0116] Grid search is a method of finding the best parameter configuration by exhaustively trying different parameter combinations.

[0117] Determine the value range of each parameter and create a grid of parameter combinations.

[0118] For each set of parameters in the grid, train the model using the training set and evaluate the performance of the model using the validation set.

[0119] Select the parameter combination with the best performance on the validation set as the final parameters.

[0120] Step S400 includes the following specific steps:

[0121] Step S410: Obtain the effective recognition count V recorded based on the anomaly recognition results of the OCSVM model and the recognition count V0 corresponding to the actual anomaly results. Effective recognition refers to the remaining recognition situations excluding incorrect anomaly recognition results and unrecognized results; calculate the result deviation value G, where G = V0 - V, and set the deviation value threshold G0.

[0122] Step S420: If G < G0, output the current OCSVM model as the final anomaly recognition model applied to the exhaust fan group; if G ≥ G0, return to Step S100 for data update. The data update includes increasing the monitoring cycle duration to increase the quantity of multivariate operation data or increasing the types of multivariate operation data.

[0123] As shown in the embodiment: After increasing the types of multivariate operation data: For example, from the initial analysis of pressure, vibration, current, and temperature, it is increased to pressure, vibration, current, temperature, and noise.

[0124] When corresponding to the calculation and analysis step S200, the calculation of the sample labels is performed based on the five features increased currently.

[0125] Such as Figure 1 This is the detailed process of the multi-source data anomaly recognition method for an efficient exhaust fan group based on OCSVM. First, key data during the operation of the exhaust fan group, such as pressure, vibration, current, and temperature, are collected through various sensors. Subsequently, feature extraction is performed on these multi-source data. Secondly, label setting is performed on the extracted data features. This step classifies the data into different categories to indicate the normal or abnormal state of the equipment. The data is further divided into a training set, a validation set, and a test set. The training set is used to train the OCSVM model so that it can identify abnormal data significantly deviating from the normal mode; the validation set is used to adjust the model parameters to improve its recognition accuracy; while the test set is used to evaluate the model performance, thereby obtaining a judgment result on whether the equipment has anomalies.

[0126] Such as Figure 2 This is the data label annotation process. When constructing the sample data of the exhaust fan group, multiple key features are considered, including static pressure, dynamic pressure, total pressure, root mean square velocity, root mean square acceleration, peak acceleration envelope, average current, peak current, current frequency, and temperature. The K-medoids clustering algorithm is used to annotate these data. Specifically, two clustering operations are performed in total: The first is full-feature clustering, that is, the complete data set including all parameter features such as pressure, vibration, current, and temperature is input into the K-medoids clustering algorithm to obtain the first type of labeled data; the second is single-feature clustering, that is, only the feature subset of a single parameter such as pressure is selected and input into the same K-medoids clustering algorithm to obtain the second type of labeled data.

[0127] like Figure 3 As shown in the figure, it is the sample data obtained by the sample label. The data shown in the figure are the first-class data labels and the second-class data labels obtained by executing the clustering algorithm twice. Specifically, the label of each feature in the first-class data label is consistent with the sample label, and the label of each feature in the second-class label is consistent with the parameter label. Subsequently, the first-class label and the second-class label of the same sample and the same feature are compared. When the values ​​in the first-class label and the second-class label are both 0, the label of the corresponding feature of the sample is also marked as 0. In other cases, that is, when the label values ​​are inconsistent or neither is 0, the feature label corresponding to the sample is marked as 1. Taking into account the importance of different parameters, different weights are given to them when calculating the final sample label. Taking the figure as an example, the weights of pressure, vibration and current account for 30% each, and the weight of temperature accounts for 10%. By comprehensively considering these weighted feature labels, the final label of the sample can be obtained to identify the abnormal state of the equipment.

[0128] The OCSVM-based high-efficiency exhaust fan unit abnormality intelligent identification system includes a multivariate data acquisition module, a data preprocessing module, a data set construction module, a sample label analysis module, a model training module, and a model determination module;

[0129] The multivariate data acquisition module is used to collect multivariate operation data of the exhaust fan unit equipped with sensor acquisition equipment;

[0130] The data preprocessing module is used to obtain the multivariate operating data obtained by the exhaust fan unit based on the sensor acquisition device during the historical monitoring period for data preprocessing;

[0131] The data set construction module is used to extract features from the preprocessed multivariate operating data to form a data set;

[0132] The sample label analysis module is used to label and verify the sample data in the data set, and analyze and output the sample labels;

[0133] The model training module is used to construct an OCSVM model through sample label data and train the OCSVM model, verify whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to identify abnormalities of the exhaust fan unit in the OCSVM model composed of the adjusted model parameters;

[0134] The model determination module is used to select the actual abnormal results recorded in the corresponding historical monitoring period based on the abnormal recognition results of the OCSVM model, analyze the deviation value of the recognition result, and return to step S100 to update the data based on the deviation value of the recognition result until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormal recognition model finally applied to the exhaust fan group.

[0135] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0136] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent identification method for abnormalities of high-efficiency exhaust fan units based on OCSVM, characterized in that: The analysis steps include: Step S100: Arrange a sensor collection device for collecting multi-element operation data on the exhaust fan unit; obtain the multi-element operation data of the exhaust fan unit obtained by the sensor collection device during the historical monitoring period and perform data preprocessing; And perform feature extraction on the preprocessed multivariate operation data to form a data set; Step S200: Based on the data set, label and verify the sample data in the data set, and analyze and output the sample labels; The step S200 includes the following specific steps: Step S210: Selecting the spatial geometric center formed by the data feature values ​​corresponding to the data features in the data set as the initial clustering center; Step S220: starting from the initial cluster center, the iteration direction is the cluster end point, where the coordinate value of the cluster end point is derived from the b% quantile of each feature data; the iteration step value is 100%-b% of the direction vector modulus formed between the initial cluster center and the cluster end point; the iteration termination condition is that the normal data volume accounts for b% of the total data volume. If the condition is not met, the iteration continues; If the condition is met, the iteration ends; Step S230: After the iteration, each sample is divided into each cluster, and then the samples in each cluster are marked; the normal data label is 1, and the abnormal data label is 0; the result obtained by the K-medoids clustering algorithm for the multivariate effective operation data is recorded as the first category sample label; Step S240: performing data subset labeling on a single feature, wherein the data subset labeling refers to performing step S220 on the single feature data respectively, and recording the obtained result as a second type of sample label; Step S250: cross-validate the first category sample label records and the second category sample label records, extract the same data features in the two categories of sample label records, perform an "AND" operation, and output the feature label; when the data label of all features is 0 and the data label obtained by a single feature is 0, the corresponding correction feature label is output as 0; in other cases, the correction feature label is output as 1; Step S260: The final sample label is determined by multiple calibration feature labels. The weight ratio of each calibration feature label is determined based on historical data, using the formula: Y=k1×Y P +k2×Y V +k3×Y E +k4×Y T ; Calculate the sample label Y, where Y P , Y V , Y E , Y T They are the calibration feature labels of the four data types of pressure, vibration, current, and temperature, and k1, k2, k3, and k4 correspond to the weight proportions of the calibration feature labels of the four data types of pressure, vibration, current, and temperature, respectively; The step S260 determines the weight proportion of each correction feature label based on historical data, including the following: Step S261: Obtain the number N of single features contained in each calibration feature tag, and the number D of times each single feature is correctly identified as an abnormal feature in the abnormal identification events recorded in the historical data of the corresponding monitoring period, wherein the correct identification refers to determining the only abnormal feature of the exhaust fan unit in the abnormal identification event and finally determining the abnormality; Using the formula: Z = a1 × N / N1 + a2 × D / D1; Calculate the influence index Z of each calibration feature label, a1 and a2 represent the corresponding reference coefficients; N1 represents the total number of all single features contained in all calibration feature labels, and D1 represents the total number of abnormal identification events recorded in the historical data of the corresponding monitoring period; Step S262: Generate label ratio values ​​f1:f2:...:f based on the influence index of each calibration feature label m , m represents the total number of calibration feature labels; Step S263: Based on the label ratio value, use the formula: 100% = f1×b+f2×b+......+f m ×b; Calculate the unit weight b of each calibration feature label, and calculate the weight proportion b×f of each calibration feature label based on the unit weight b j , f j ={f1,f2,......,f m }, f j Represents the label ratio value of the jth correction feature label, j≤m; Step S300: construct an OCSVM model through sample label data and train the OCSVM model, verify whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to identify abnormalities of the exhaust fan unit based on the test set for the OCSVM model composed of the adjusted model parameters; Step S400: Based on the abnormality recognition result of the OCSVM model, select the actual abnormality result recorded in the corresponding historical monitoring period, analyze the recognition result deviation value, and return to step S100 to update the data based on the recognition result deviation value until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormality recognition model finally applied to the exhaust fan unit.

2. The method for intelligently identifying abnormalities of a high-efficiency exhaust fan unit based on OCSVM according to claim 1 is characterized in that: The step S100 includes the following specific steps: Step S110: The multivariate operating data includes but is not limited to pressure, vibration, current, and temperature; the collected multivariate operating data is clustered and divided using the DBSCAN method, core data and outliers are identified, and the outliers are removed to update the multivariate operating data into multivariate valid operating data; Step S120: normalize the multivariate effective operation data using the formula: x0=[x-min(x)] / [max(x)-min(x)], Generate normalized multivariate valid operating data x0; wherein x represents the multivariate valid operating data before normalization, min(x) represents the minimum value in the multivariate valid operating data, and max(x) represents the maximum value in the multivariate valid operating data; Step S130: extract features from the multivariate effective operation data using a signal processing method, and construct a data set corresponding to data features and data feature values.

3. The method for intelligently identifying abnormalities of a high-efficiency exhaust fan unit based on OCSVM according to claim 2 is characterized in that: The method of constructing an OCSVM model through sample label data and training the OCSVM model includes the following steps: Divide the sample label data into training set, validation set and test set; Based on the training set data, OCSVM is used to find an optimal hyperplane to achieve model optimization by minimizing an objective function. The optimization process is as follows: ; Where ω is the normal vector of the hyperplane, ξi is the slack variable, ρ is the distance from the hyperplane to the origin, θ is the regularization parameter used to control the trade-off between the error term and the margin, and n is the amount of data; Constraints: The objective function is subject to the following constraints, ; In the formula, Φ(x i ) is a function that maps data points to feature space; The Lagrangian operator is used to transform the optimization objective function into: ; Using the decision function of OCSVM, we get: ; In the formula, if f(x)=1, then x is considered a normal point, and if f(x)=-1, then x is considered an abnormal point; Based on the above model, the model is trained using the training set.

4. The method for intelligently identifying abnormalities of a high-efficiency exhaust fan unit based on OCSVM according to claim 3 is characterized in that: The verification of whether the OCSVM model parameters need to be adjusted includes the following steps: Verify the accuracy of the model based on the validation set using the formula: L=TP / (TP+FN); Calculate the accuracy L of the training model, where TP represents the number of samples correctly identified as positive examples by the training model, and FN represents the number of samples that are actually positive examples but are incorrectly identified as negative examples by the model; Set the accuracy threshold L0. When L < L0, adjust the model parameters. Use grid search to construct different parameter combinations to find the optimal parameter combination when the model output accuracy on the validation set is greater than or equal to L0 and is the maximum value, and use it as the parameter for the model output application.

5. The method for intelligently identifying abnormalities of a high-efficiency exhaust fan unit based on OCSVM according to claim 4 is characterized in that: The step S400 includes the following specific steps: Step S410: Obtain the effective recognition number V recorded based on the abnormal recognition results of the OCSVM model and the recognition number V0 corresponding to the actual abnormal results. The effective recognition refers to the remaining recognition situations except for the incorrect abnormal recognition results and the unrecognized results. Calculate the result deviation value G, G = V0 - V, and set the deviation value threshold G0. Step S420: If G < G0, output the current OCSVM model as the abnormal recognition model finally applied to the exhaust fan group; if G ≥ G0, return to step S100 for data update. The data update includes increasing the monitoring period duration to increase the quantity of multivariate operation data or increasing the type of multivariate operation data.

6. An OCSVM-based high-efficiency exhaust fan unit abnormality intelligent identification system, using an OCSVM-based high-efficiency exhaust fan unit abnormality intelligent identification method as described in any one of claims 1 to 5, characterized in that: It includes a multivariate data acquisition module, a data preprocessing module, a data set construction module, a sample label analysis module, a model training module, and a model determination module; The multivariate data acquisition module is used to collect multivariate operation data of the exhaust fan group equipped with sensor acquisition devices. The data preprocessing module is used to perform data preprocessing on the multivariate operation data obtained by the exhaust fan group based on the sensor acquisition devices during the historical monitoring period. The data set construction module is used to extract features from the preprocessed multivariate operation data to form a data set. The sample label analysis module is used to label and verify the sample data in the data set, and analyze and output sample labels. The model training module is used to construct an OCSVM model through sample label data and train the OCSVM model, check whether the OCSVM model parameters need to be adjusted, and use the OCSVM algorithm to perform abnormal recognition of the exhaust fan group on the OCSVM model composed of the adjusted model parameters. The model determination module is used to select the actual abnormal results recorded during the corresponding historical monitoring period based on the abnormal recognition results of the OCSVM model, analyze the recognition result deviation value, and return to step S100 for data update based on the recognition result deviation value until the result deviation value corresponding to the trained OCSVM model is less than or equal to the deviation value threshold, and output the corresponding OCSVM model as the abnormal recognition model finally applied to the exhaust fan group.