A method, device and medium for detecting abnormal operation of a wind bag pump

Abstract

This application discloses a method, device, and medium for detecting abnormal operation of an airbag pump, belonging to the field of airbag pump technology. It addresses the problem that initial abnormalities in airbag pumps are often overlooked, leading to gradual deterioration and ultimately equipment shutdown and reduced pump efficiency. The method involves extracting features from airbag pump operating data using a Transformer model and constructing an airbag pump feature space. Data detection is performed on this feature space to identify actual abnormal operation data. Based on the Transformer time-series prediction model and the feature data in the airbag pump feature space, the operating status of the airbag pump within a preset time period is predicted, yielding predicted abnormal operation data. The actual abnormal operation data and the predicted abnormal operation data are combined to obtain an abnormal data set, determining the duration and severity of the abnormal operation, and triggering an alarm for the abnormal operation.

Description

Technical Field

[0001] This application relates to the field of airbag pump technology, and in particular to a method, equipment and medium for detecting abnormal operation of an airbag pump. Background Technology

[0002] In current industrial production and fluid transport fields, airbag pumps, as an important conveying device, are widely used in various fluid media transport tasks due to their unique structural design and high efficiency. However, despite the excellent performance of airbag pumps in most applications, their fault monitoring and diagnosis technology still faces many challenges, a problem that is particularly prominent in current industrial practice.

[0003] Traditional methods for monitoring airbag pump malfunctions often rely on manual inspections and periodic maintenance. This approach is not only time-consuming and labor-intensive, but also makes it difficult to detect potential equipment faults in a timely manner. Because airbag pumps typically operate in complex environments, changes in various parameters during operation are difficult to accurately capture with the naked eye or simple testing tools. Therefore, even if subtle abnormalities appear in the early stages of a malfunction, they are often overlooked due to a lack of effective monitoring methods, leading to the gradual deterioration of the fault and ultimately causing equipment shutdown or more serious consequences, thus reducing the airbag pump's operating efficiency. Summary of the Invention

[0004] This application provides a method, device, and medium for detecting abnormal operation of a windshield pump, which solves the following technical problem: In the early stage of windshield pump failure, even if the equipment has shown some minor abnormalities, they are often overlooked due to the lack of effective monitoring methods, which leads to the gradual deterioration of the failure and eventually causes the equipment to shut down or more serious consequences, thereby reducing the working efficiency of the windshield pump.

[0005] The embodiments of this application adopt the following technical solutions:

[0006] This application provides a method for detecting abnormal operation of an airbag pump, including: acquiring airbag pump operation data; extracting features from the airbag pump operation data using the self-attention mechanism of a Transformer model, and constructing an airbag pump feature space based on the extracted feature data; performing data detection on the airbag pump feature space using a feature point detection algorithm to determine the actual abnormal operation data of the airbag pump; predicting the operation status of the airbag pump within a preset time period based on the Transformer time series prediction model and the feature data in the airbag pump feature space to obtain predicted abnormal operation data of the airbag pump; combining the actual abnormal operation data of the airbag pump with the predicted abnormal operation data of the airbag pump to obtain an abnormal data set of the airbag pump, and determining the abnormal operation duration of the airbag pump based on the abnormal operation data set of the airbag pump; comparing the abnormal data set of the airbag pump with a preset abnormal database to determine the abnormality level of the airbag pump; and issuing an abnormal alarm for the airbag pump based on the abnormal operation duration and the abnormality level of the airbag pump.

[0007] This application's embodiments extract key features from the airbag pump's operational data using the self-attention mechanism of the Transformer model and construct a feature space to achieve comprehensive monitoring of the airbag pump's operational status. It also allows for correlation of airbag pump operational data from different structural locations, enabling more comprehensive and accurate identification of equipment anomalies. The feature point detection algorithm can accurately identify potential equipment faults in the feature space, preventing further deterioration leading to downtime or more serious consequences. Leveraging the Transformer model's time-series prediction capabilities, the operational status of the airbag pump within a preset time period is predicted, thereby identifying potential anomaly trends in advance and providing maintenance personnel with sufficient time for fault prevention and preparation. Combining actual operational anomaly data with predicted operational anomaly data forms a complete anomaly data set, allowing maintenance personnel to quickly obtain the equipment's fault type and severity, and then take appropriate countermeasures for timely maintenance of the airbag pump.

[0008] In one implementation of this application, feature extraction is performed on the windshield pump operation data using the self-attention mechanism of the Transformer model. A feature space is then constructed based on the extracted feature data. Specifically, this includes: converting the windshield pump operation data into sequential data; obtaining sequential features based on the acquisition order of each data point in the windshield pump operation data; performing positional encoding on the sequential features to generate embedding vectors corresponding to the sequential features; and using the global self-attention mechanism of the Transformer model to determine the first derived vector group corresponding to each embedding vector; and multiplying and summing the first derived vector groups to obtain the globally most relevant feature corresponding to each windshield pump operation feature. Information; where the globally most relevant information is the comprehensive correlation information between each windshield pump operating feature and all other windshield pump operating features; through the local self-attention mechanism of the Transformer model, the second derived vector group corresponding to each embedding vector is determined, and the second derived vector group is multiplied and summed to obtain the adjacent most relevant information corresponding to each windshield pump operating feature; where the adjacent most relevant information is the correlation information adjacent to the current windshield pump operating feature position; based on the globally most relevant information and the adjacent most relevant information, the correlation relationship between windshield pump operating features is obtained, and the windshield pump feature space is constructed based on the correlation relationship between windshield pump operating features.

[0009] In one implementation of this application, a feature space for the windbag pump is constructed based on the correlation between the operational features of the windbag pump. Specifically, this includes: obtaining a global correlation matrix based on the most relevant global information, and obtaining a local correlation matrix based on adjacent most relevant information; determining corresponding weight combinations based on the operational feature types of the windbag pump, and weighting the global correlation matrix and the local correlation matrix based on the weight combinations to obtain an integrated correlation matrix; comparing the integrated correlation matrix with a preset matrix threshold to filter out correlations that do not meet the correlation conditions; and generating a node graph based on the remaining correlations to construct the feature space for the windbag pump.

[0010] In one implementation of this application, a node graph is generated based on the remaining associations to construct the feature space of the airbag pump. Specifically, this includes: determining the position of each node in the airbag pump based on the node graph; determining the feature type corresponding to each node; determining the association coefficient between each node based on the position and feature type; and adjusting the feature value corresponding to each node based on the association coefficient to construct the feature space of the airbag pump based on the adjusted feature value.

[0011] In one implementation of this application, a feature point detection algorithm is used to detect data in the feature space of the airbag pump to identify actual abnormal operating data of the airbag pump. Specifically, this includes: sorting different types of data in the feature space of the airbag pump to generate multiple feature trajectories; determining the low-dimensional representation of each data point in the multiple feature trajectories using a pre-set autoencoder; determining the distance between each data point and other data points based on the low-dimensional representation to construct a distance matrix; determining the K nearest neighbors of each data point based on the distance matrix; and performing data detection on each data point in the feature trajectory based on a pre-set anomaly threshold and the K nearest neighbors to identify actual abnormal operating data of the airbag pump.

[0012] In one implementation of this application, the operating status of the airbag pump within a preset future time period is predicted based on the Transformer time series prediction model and the feature data in the feature space of the airbag pump, to obtain predicted abnormal operating data of the airbag pump. Specifically, this includes: inputting the feature data in the feature space of the airbag pump into the Transformer time series prediction model to obtain predicted operating data of the airbag pump within the preset future time period; constructing an operating data line chart based on the predicted operating data of the airbag pump, determining the difference between adjacent data in the operating data line chart, and determining the airbag pump operation change segment and the airbag pump operation stability segment based on the difference; comparing the difference between adjacent data corresponding to the airbag pump operation stability segment with a first preset difference threshold to determine the first abnormal data; comparing the difference between adjacent data corresponding to the airbag pump operation change segment with a second preset difference threshold to determine the second abnormal data; and obtaining the predicted abnormal operating data of the airbag pump based on the first abnormal data and the second abnormal data.

[0013] In one implementation of this application, the actual abnormal operation data of the airbag pump and the predicted abnormal operation data of the airbag pump are combined to obtain an abnormal data set of the airbag pump. The abnormal operation duration of the airbag pump is determined based on the abnormal data set of the airbag pump. Specifically, this includes: arranging the actual abnormal operation data and the predicted abnormal operation data of the airbag pump in chronological order and marking the corresponding time points; determining the time difference between adjacent time points and determining the abnormal change rate of the airbag pump based on the ratio between the later time difference and the earlier time difference; inputting the abnormal change rate of the airbag pump into the airbag pump's operational life prediction model, so as to output the abnormal operation duration of the airbag pump through the airbag pump's operational life prediction model.

[0014] In one implementation of this application, the abnormal data set of the airbag pump is compared with a preset abnormal database to determine the degree of abnormality of the airbag pump. Specifically, this includes: comparing the abnormal data set of the airbag pump with the preset abnormal database to obtain a reference abnormality type; wherein, the preset abnormal database contains multiple abnormal data sets of airbag pumps, and also includes the abnormal types of airbag pumps corresponding to each of the multiple abnormal data sets of airbag pumps; determining the abnormal data difference between the abnormal data set of the airbag pump and the data set corresponding to the reference abnormality type, and determining a first weight based on the abnormal data difference; comparing the reference abnormality type with a preset abnormality type level table to determine a second weight based on the comparison result; and determining the degree of abnormality of the airbag pump abnormal data set based on the abnormal data set of the airbag pump, the first weight, and the second weight.

[0015] This application provides a device for detecting abnormal operation of a windshield pump, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: acquire windshield pump operation data; extract features from the windshield pump operation data using the self-attention mechanism of a Transformer model, and construct a windshield pump feature space based on the extracted feature data; perform data detection on the windshield pump feature space using a feature point detection algorithm to determine the actual abnormal operation data of the windshield pump; predict the operation status of the windshield pump within a preset time period based on the Transformer time series prediction model and the feature data in the windshield pump feature space, to obtain predicted abnormal operation data of the windshield pump; combine the actual abnormal operation data of the windshield pump with the predicted abnormal operation data of the windshield pump to obtain a windshield pump abnormal data set, and determine the abnormal operation duration of the windshield pump based on the windshield pump abnormal data set; compare the windshield pump abnormal data set with a preset abnormal database to determine the abnormality level of the windshield pump; and issue an abnormal alarm for the windshield pump based on the abnormal operation duration and the abnormality level of the windshield pump.

[0016] This application provides a non-volatile computer storage medium storing computer-executable instructions, which are configured to: acquire windshield pump operating data; extract features from the windshield pump operating data using the self-attention mechanism of a Transformer model, and construct a windshield pump feature space based on the extracted feature data; perform data detection on the windshield pump feature space using a feature point detection algorithm to determine the actual abnormal operating data of the windshield pump; predict the operating status of the windshield pump within a preset time period based on the Transformer time series prediction model and the feature data in the windshield pump feature space, and obtain predicted abnormal operating data of the windshield pump; combine the actual abnormal operating data of the windshield pump with the predicted abnormal operating data of the windshield pump to obtain an abnormal data set of the windshield pump, and determine the abnormal operating duration of the windshield pump based on the abnormal data set of the windshield pump; compare the abnormal data set of the windshield pump with a preset abnormal database to determine the abnormality value of the windshield pump; and perform an abnormal alarm for the windshield pump based on the abnormal operating duration and the abnormality value of the windshield pump.

[0017] The above-mentioned technical solutions adopted in this application embodiment can achieve the following beneficial effects: This application embodiment extracts key features from the operation data of the airbag pump through the self-attention mechanism of the Transformer model and constructs a feature space to achieve comprehensive monitoring of the operation status of the airbag pump. It can also correlate the operation data of the airbag pump at different structural locations, and more comprehensively and accurately identify abnormal conditions of the equipment. The feature point detection algorithm can accurately identify potential faults of the equipment in the feature space, avoiding further deterioration of the fault leading to downtime or more serious consequences. Through the time series prediction capability of the Transformer model, the operation status of the airbag pump in the future within a preset time period can be predicted, thereby discovering potential abnormal trends in advance and providing maintenance personnel with sufficient time for fault prevention and preparation. Combining actual abnormal operation data with predicted abnormal operation data forms a complete abnormal data group, enabling maintenance personnel to quickly obtain the fault type and severity of the equipment, and then take appropriate countermeasures to maintain the airbag pump in a timely manner. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0019] Figure 1 A flowchart of a method for detecting abnormal operation of a windshield pump provided in this application embodiment;

[0020] Figure 2 This is a schematic diagram of a device for detecting abnormal operation of a wind pump, provided in an embodiment of this application.

[0021] Figure label:

[0022] 200: Detection device for abnormal operation of airbag pump; 201: Processor; 202: Memory. Detailed Implementation

[0023] This application provides a method, equipment, and medium for detecting abnormal operation of a wind pump.

[0024] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0025] The technical solutions proposed in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Figure 1 A flowchart of a method for detecting abnormal operation of a windshield pump provided in this application embodiment is shown below. Figure 1 As shown, the method for detecting abnormal operation of the airbag pump includes the following steps:

[0027] Step 101: Obtain the operating data of the airbag pump.

[0028] In one embodiment of this application, multiple operational data of the airbag pump are acquired, such as air supply pressure data, flow rate data, discharge pressure data, transmission efficiency, vibration data at different locations of the airbag pump, and temperature data at different locations of the airbag pump. This data can be acquired in real time using sensors installed at different locations on the airbag pump. For example, a pressure sensor installed on the air supply pipe monitors the air supply pressure of the airbag pump in real time; a pressure sensor installed on the discharge pipe acquires the pressure value of the gas discharged by the airbag pump; vibration sensors are installed at different locations on the airbag pump (such as bearings, housings, etc.) to monitor the vibration of the airbag pump in real time; and a temperature sensor is used to monitor temperature changes at different locations on the airbag pump.

[0029] Step 102: Extract features from the windshield pump operation data using the self-attention mechanism of the Transformer model, and construct the windshield pump feature space based on the extracted feature data.

[0030] In one embodiment of this application, the operation data of the windshield pump is converted into sequence data. Sequence features are obtained based on the acquisition order of each data point in the windshield pump operation data. Position encoding is performed on the sequence features to generate embedding vectors corresponding to the sequence features. Using the global self-attention mechanism of the Transformer model, a first derived vector group corresponding to each embedding vector is determined. The first derived vector groups are multiplied and summed to obtain the globally most relevant information corresponding to each windshield pump operation feature. The globally most relevant information is the comprehensive correlation information between each windshield pump operation feature and all other windshield pump operation features. Using the local self-attention mechanism of the Transformer model, a second derived vector group corresponding to each embedding vector is determined. A global correlation matrix is ​​obtained based on the globally most relevant information, and a local correlation matrix is ​​obtained based on adjacent most relevant information. Based on the windshield pump operation feature type, a corresponding weight combination is determined. The global correlation matrix and the local correlation matrix are weighted based on the weight combination to obtain an integrated correlation matrix. The integrated correlation matrix is ​​compared with a preset matrix threshold to filter out correlations that do not meet the correlation conditions. Based on the remaining associated node graph, a feature space set of the windshield pump is constructed and multiplied and summed to obtain the nearest most relevant information corresponding to each windshield pump operating feature. The nearest most relevant information is the association information adjacent to the current windshield pump operating feature. Based on the global most relevant information and the nearest most relevant information, the association relationships between the windshield pump operating features are obtained, and the windshield pump feature space is constructed based on these association relationships.

[0031] Specifically, based on the installation order of the sensors on the airbag pump, the data is organized into sequential data. Each data point in the sequence is position-encoded to distinguish its location within the sequence, and each data point is converted into an embedding vector. Using the global self-attention mechanism of the Transformer model, attention weights are calculated between each embedding vector and all other embedding vectors. These weights reflect the degree of correlation between each embedding vector and the others. Based on the attention weights, a first derived vector group is generated for each embedding vector. This first derived vector group contains comprehensive correlation information between each embedding vector and the others. The first derived vector group is multiplied and summed to obtain the globally most relevant information for each airbag pump operating feature.

[0032] Furthermore, using the local self-attention mechanism of the Transformer model, attention weights are calculated between each embedding vector and its neighboring embedding vectors. These weights reflect the degree of correlation between each embedding vector and its neighboring embedding vectors. Based on the attention weights, a second derived vector group is generated for each embedding vector. The second derived vector group contains the correlation information between each embedding vector and its neighboring embedding vectors. The second derived vector group is multiplied and summed to obtain the most relevant neighboring information for each windshield pump operating feature.

[0033] Furthermore, based on the globally most relevant information and adjacent most relevant information, the correlations between the operating features of the airbag pump are extracted. These correlations reflect the mutual influence and dependence between different operating features. Using the extracted correlations, a feature space for the airbag pump is constructed. In this embodiment, each dimension of the feature space corresponds to one airbag pump operating feature, and the relative position and distance between features reflect the degree of correlation between them.

[0034] Suppose we have an airbag pump under test, whose operating data includes supply pressure, flow rate, discharge pressure, data from two vibration sensors, and data from two temperature sensors. This data is organized into a sequence according to the sensor installation order and converted into embedding vectors. Then, the global and local self-attention mechanisms of the Transformer model are used to calculate the correlation between each embedding vector and other embedding vectors. For example, for the supply pressure embedding vector, the global self-attention mechanism might output that it has a strong correlation with the flow rate and discharge pressure embedding vectors, but a weaker correlation with the vibration and temperature embedding vectors. The local self-attention mechanism might output that it has the strongest correlation with its immediate neighbor, the flow rate embedding vector. Based on this correlation information, a feature space for the airbag pump is constructed, where the connection lines between the supply pressure feature and the flow rate and discharge pressure features are thicker to indicate a stronger correlation, while the connection lines with the vibration and temperature features are thinner to indicate a weaker correlation. This constructed feature space allows for a better understanding of the airbag pump's operating status and the detection of potential faults or anomalies.

[0035] In one embodiment of this application, a global correlation matrix is ​​obtained based on the most relevant global information, and a local correlation matrix is ​​obtained based on adjacent most relevant information. Based on the operational characteristic type of the windshield pump, a corresponding weight combination is determined. The global correlation matrix and the local correlation matrix are then weighted based on this weight combination to obtain an integrated correlation matrix. The integrated correlation matrix is ​​compared with a preset matrix threshold to filter out correlations that do not meet the correlation criteria. A node graph is generated based on the remaining correlations to construct the windshield pump feature space.

[0036] Specifically, the most relevant global information is organized into a matrix, namely the global correlation matrix. Each element in the matrix represents the degree of correlation between two features; a larger element value indicates a stronger correlation. Adjacent most relevant information is also organized into a matrix, namely the local correlation matrix. Each element in this matrix represents the degree of correlation between two adjacent features. Based on the type and importance of the ventilator pump's operational features, different weights are assigned to the global and local correlation matrices. These weights can be determined based on expert experience. Using the determined weight combination, the global and local correlation matrices are weighted. The purpose of this weighting is to combine the global and local correlation information to obtain a more comprehensive and accurate integrated correlation matrix.

[0037] Furthermore, the integrated correlation matrix is ​​compared with a preset matrix threshold to retain strongly correlated feature pairs and remove weakly correlated or irrelevant feature pairs, thereby simplifying the structure of the feature space. Based on the remaining correlations, i.e., the elements in the filtered integrated correlation matrix, a node graph is generated. Each node in the node graph represents a windshield pump operation feature, and the lines between nodes represent the correlations between features.

[0038] In one embodiment of this application, the position of each node in the airbag pump is determined based on a node graph. The feature type corresponding to each node is also determined. Based on the position and feature type, the correlation coefficient between each node is determined. The feature values ​​corresponding to each node are adjusted based on the correlation coefficient to construct the airbag pump feature space based on the adjusted feature values.

[0039] Specifically, in the nodal diagram, each node represents an operational characteristic of the airbag pump. The location of the node can be determined based on the actual installation position of the sensor on the airbag pump. Each node corresponds to a specific airbag pump operating characteristic, such as supply pressure, flow rate, discharge pressure, vibration, and temperature. In the nodal diagram, the lines connecting the nodes represent the correlation between characteristics. The thickness or color of the lines indicates the strength of the correlation. By calculating the correlation between nodes, a correlation coefficient is obtained. The magnitude of the correlation coefficient reflects the strength and direction of the interaction between characteristics. After determining the correlation coefficient, the characteristic values ​​of each node are adjusted by weighted summation. The adjusted characteristic values ​​more accurately reflect the actual operating state of the airbag pump and take into account the mutual influence between characteristics.

[0040] Step 103: Perform data detection on the feature space of the airbag pump using a feature point detection algorithm to identify the actual abnormal data of the airbag pump operation.

[0041] In one embodiment of this application, different types of data in the feature space of the windshield pump are sorted to generate multiple feature trajectories. A pre-set autoencoder is used to determine the low-dimensional representation of each data point in the multiple feature trajectories. Based on the low-dimensional representation, the distance between each data point and other data points is determined to construct a distance matrix. Based on the distance matrix, the K nearest neighbors corresponding to each data point are determined. Based on a pre-set anomaly threshold and the K nearest neighbors, data detection is performed on each data point in the feature trajectories to identify abnormal data in the actual operation of the windshield pump.

[0042] Specifically, in the feature space of the airbag pump, different types of data, such as supply pressure, flow rate, discharge pressure, vibration, and temperature, are sorted to ensure the continuity between data points, forming multiple feature trajectories. Each feature trajectory represents a specific type of airbag pump operation data.

[0043] Furthermore, a pre-defined autoencoder transforms the high-dimensional data points in the feature trajectory into a low-dimensional representation. This low-dimensional representation allows for more efficient calculation of distances between data points. Using the low-dimensional representation, the Euclidean distance between each data point and other data points is calculated, and these distances are organized into a matrix, namely the distance matrix. Each element in the matrix represents the distance between two data points.

[0044] Furthermore, for each data point in the distance matrix, its K nearest neighbors are found, i.e., the K closest data points, and a preset outlier threshold is set. For each data point in the feature trajectory, the sum of its distances to its K nearest neighbors is calculated and compared with the outlier threshold. If the sum of distances of a data point exceeds the threshold, it is marked as an outlier. These outliers represent abnormal conditions or malfunctions in the operation of the windshield pump.

[0045] Step 104: Based on the Transformer time series prediction model and the feature data in the feature space of the airbag pump, predict the operating status of the airbag pump within a preset time period in the future to obtain the predicted abnormal operating data of the airbag pump.

[0046] In one embodiment of this application, feature data from the feature space of the airbag pump is input into a Transformer time series prediction model to obtain predicted operating data of the airbag pump within a preset future time period. An operating data line chart is constructed based on the predicted operating data of the airbag pump, and the differences between adjacent data in the line chart are determined. Based on these differences, the changing operating segments and stable operating segments of the airbag pump are identified. The differences between adjacent data corresponding to the stable operating segments are compared with a first preset difference threshold to identify first abnormal data. The differences between adjacent data corresponding to the changing operating segments are compared with a second preset difference threshold to identify second abnormal data. Based on the first and second abnormal data, the predicted abnormal operating data of the airbag pump is obtained.

[0047] Specifically, feature data from the feature space of the airbag pump, such as air supply pressure, flow rate, and temperature, are used as input to the Transformer time series prediction model. These feature data are arranged in a time series manner, meaning each feature has a corresponding timestamp. The Transformer model learns the temporal dependencies between feature data through a self-attention mechanism. Based on these relationships, the model can predict the operating data of the airbag pump within a preset time period in the future.

[0048] Furthermore, the future operational data predicted by the Transformer model is plotted as a line graph, where each point represents a predicted value at a given time point. The differences between adjacent data points in the line graph are calculated; these differences reflect the changing trends of the airbag pump's operational data. Based on the magnitude of the differences, the line graph is divided into fluctuating and stable operating segments. Fluctuating segments refer to those with larger differences, indicating significant changes in the airbag pump's operational data over a short period. Stable segments refer to those with smaller differences, indicating relatively stable operational data.

[0049] Furthermore, the difference between adjacent data corresponding to the stable segment is compared with a first preset difference threshold. If a certain difference exceeds the first preset difference threshold, the data point is considered to be the first abnormal data. The first preset difference threshold is used to determine the stability of the stable segment, that is, the difference between adjacent data remains within a stable range.

[0050] Furthermore, in the variation range, although the operating data of the airbag pump will change significantly, this change is usually gradual. The difference between adjacent data corresponding to the variation range is compared with a second preset difference threshold. If a difference exceeds the second preset difference threshold, for example, if the change is too drastic or does not conform to the expected trend, then the data point is considered the second abnormal data. The first and second abnormal data are summarized to obtain the complete set of predicted abnormal operating data of the airbag pump.

[0051] Step 105: Combine the actual abnormal operation data of the airbag pump with the predicted abnormal operation data of the airbag pump to obtain the abnormal data group of the airbag pump, and determine the abnormal operation duration of the airbag pump based on the abnormal data group of the airbag pump.

[0052] In one embodiment of this application, the actual operational anomaly data and the predicted operational anomaly data of the airbag pump are arranged in chronological order, and the corresponding time points are marked. The time difference between adjacent time points is determined, and the anomaly change rate of the airbag pump is determined based on the ratio between the later time difference and the earlier time difference. The anomaly change rate of the airbag pump is input into the airbag pump's operational life prediction model, so that the abnormal operational life of the airbag pump is output through the airbag pump's operational life prediction model.

[0053] Specifically, the two types of abnormal data are arranged in chronological order, with each data point corresponding to a specific time point, representing the moment the abnormal event occurred. In the arranged data sequence, each abnormal data point is marked with its corresponding time point, which can be a specific date and time or an offset relative to a starting point. The time difference between adjacent abnormal data points is calculated, representing the time interval between two abnormal events. Based on the ratio of the subsequent time difference to the preceding time difference, the abnormal change rate of the windshield pump is calculated. If the ratio is greater than 1, it indicates that the frequency of abnormal events is increasing; if the ratio is less than 1, it indicates that the frequency of abnormal events is decreasing; if the ratio is close to 1, it indicates that the frequency of abnormal events is relatively stable.

[0054] Furthermore, the calculated abnormal change rate is used as input to the windshield pump's operational life prediction model. The training process for this model involves using the change rates of historical abnormal events as input samples and the corresponding windshield pump operational life as output to train a pre-defined neural network model. This model is used to predict the operational life of the windshield pump under a given abnormal change rate. The predicted operational life of the windshield pump under the current abnormal change rate is then obtained through this model. This duration can be used to assess the remaining service life of the windshield pump and whether early maintenance or replacement is necessary.

[0055] Step 106: Compare the abnormal data set of the airbag pump with the preset abnormal database to determine the degree of abnormality of the airbag pump.

[0056] In one embodiment of this application, an abnormal data set of the airbag pump is compared with a preset abnormality database to obtain a reference abnormality type. The preset abnormality database includes multiple abnormal data sets of airbag pumps, as well as the corresponding abnormal types of airbag pumps for each of these data sets. The abnormal data difference between the abnormal data sets of the airbag pumps and the data sets corresponding to the reference abnormality types is determined, and a first weight is determined based on this difference. The reference abnormality type is compared with a preset abnormality type level table to determine a second weight based on the comparison result. Based on the abnormal data sets of the airbag pumps, the first weight, and the second weight, the abnormality degree value of the abnormal data sets of the airbag pumps is determined.

[0057] Specifically, the identified abnormal data sets of the airbag pump are compared with a preset abnormal database. The preset abnormal database in this embodiment contains various known abnormal data sets of airbag pumps and their corresponding abnormal types. For example, suppose a set of airbag pump operating data is monitored in real time and it is found that the pressure and flow values ​​are both lower than normal, forming an abnormal data set. This abnormal data set is compared with the preset abnormal database and it is found that it is most similar to the "blockage" type of abnormal data set in the database.

[0058] Furthermore, the data difference between the real-time monitored abnormal data set and the data set corresponding to the reference abnormality type is determined. This difference reflects the degree of deviation between the actual abnormality and the known abnormality type. Based on this abnormal data difference, a first weight is determined. The weight is typically inversely proportional to the abnormal data difference; that is, the smaller the difference, the larger the weight, indicating that the abnormality is closer to a known abnormality type. The determined reference abnormality type is compared with a pre-set abnormality type ranking table, which typically categorizes abnormality types into different levels based on their severity or impact. Based on the comparison results, a second weight is determined, reflecting the severity of the reference abnormality type.

[0059] Furthermore, based on the real-time monitored abnormal data sets of the airbag pump, the first weight, and the second weight, the degree of abnormality of the abnormal data sets is determined by weighted summation.

[0060] Step 107: Based on the abnormal running time and abnormality level of the airbag pump, issue an abnormal alarm for the airbag pump.

[0061] In one embodiment of this application, when an abnormality is detected in the airbag pump, the start time of the abnormality is recorded, and the duration of abnormal operation of the airbag pump is calculated based on the start time and the current time. Alarm triggering conditions are set by combining the abnormality severity value and the abnormal operation duration. For example, an alarm is triggered when the abnormality severity exceeds a certain level and the abnormal operation duration exceeds a set time. The alarm information should include key information such as the abnormality type, severity, and abnormal operation duration of the airbag pump, so that relevant personnel can quickly understand the fault and respond.

[0062] This application provides a device 200 for detecting abnormal operation of a windshield pump, including: at least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: acquire windshield pump operation data; extract features from the windshield pump operation data using the self-attention mechanism of the Transformer model, and construct a windshield pump feature space based on the extracted feature data; and perform data detection on the windshield pump feature space using a feature point detection algorithm, so as to... The process involves identifying actual abnormal operating data of the airbag pump; predicting the operating status of the airbag pump within a preset time period based on the Transformer time series prediction model and the feature data in the airbag pump's feature space, thus obtaining predicted abnormal operating data; combining the actual abnormal operating data and the predicted abnormal operating data to obtain an abnormal data set, and determining the abnormal operating duration based on this data set; comparing the abnormal data set with a pre-set abnormal database to determine the degree of abnormality; and issuing an alarm for abnormal operation based on the abnormal operating duration and degree of abnormality.

[0063] This application provides a non-volatile computer storage medium storing computer-executable instructions, which are configured to: acquire windshield pump operating data; extract features from the windshield pump operating data using the self-attention mechanism of a Transformer model, and construct a windshield pump feature space based on the extracted feature data; perform data detection on the windshield pump feature space using a feature point detection algorithm to determine the actual abnormal operating data of the windshield pump; predict the operating status of the windshield pump within a preset time period based on the Transformer time series prediction model and the feature data in the windshield pump feature space, and obtain predicted abnormal operating data of the windshield pump; combine the actual abnormal operating data of the windshield pump with the predicted abnormal operating data of the windshield pump to obtain an abnormal data set of the windshield pump, and determine the abnormal operating duration of the windshield pump based on the abnormal data set of the windshield pump; compare the abnormal data set of the windshield pump with a preset abnormal database to determine the abnormality value of the windshield pump; and perform an abnormal alarm for the windshield pump based on the abnormal operating duration and the abnormality value of the windshield pump.

[0064] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0065] The above descriptions are merely embodiments of this application and are not intended to limit the scope of this application. For those skilled in the art, various modifications and variations can be made to the embodiments of this application. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions in the embodiments of this application.

Claims

1. A method for detecting abnormal operation of a wind pump, characterized in that, The method includes: Obtain the operating data of the airbag pump; The windbag pump operation data is feature extracted using the self-attention mechanism of the Transformer model, and a feature space of the windbag pump is constructed based on the extracted feature data. The feature space of the airbag pump is inspected using a feature point detection algorithm to identify abnormal data in the actual operation of the airbag pump. Specifically, this includes: sorting different types of data in the feature space of the airbag pump to generate multiple feature trajectories; determining the low-dimensional representation of each data point in the multiple feature trajectories using a pre-set autoencoder; determining the distance between each data point and other data points based on the low-dimensional representation to construct a distance matrix; determining the K nearest neighbors of each data point based on the distance matrix; and performing data detection on each data point in the feature trajectory based on a pre-set anomaly threshold and the K nearest neighbors to identify abnormal data in the actual operation of the airbag pump. Based on the Transformer time series prediction model and the feature data in the feature space of the airbag pump, the operating status of the airbag pump in the future preset time period is predicted to obtain the predicted abnormal operating data of the airbag pump. The actual abnormal operation data of the airbag pump is combined with the predicted abnormal operation data of the airbag pump to obtain the abnormal operation data group of the airbag pump, and the abnormal operation duration of the airbag pump is determined based on the abnormal operation data group of the airbag pump. The abnormal data set of the airbag pump is compared with a preset abnormal database to determine the degree of abnormality of the airbag pump. An alarm is triggered for the abnormal operation duration and degree of abnormality of the airbag pump.

2. The method for detecting abnormal operation of a wind pump according to claim 1, characterized in that, The step of extracting features from the windshield pump operating data using the self-attention mechanism of the Transformer model, and constructing a windshield pump feature space based on the extracted feature data, specifically includes: The windbag pump operation data is converted into sequence data. Sequence features are obtained based on the acquisition position order of each data in the windbag pump operation data. The sequence features are then position-encoded to generate the embedding vector corresponding to the sequence features. By using the global self-attention mechanism of the Transformer model, the first derived vector group corresponding to each of the embedded vectors is determined. The first derived vector group is multiplied and summed to obtain the most relevant global information corresponding to each windshield pump operation feature. The most relevant global information is the comprehensive correlation information between each windshield pump operation feature and all other windshield pump operation features. By using the local self-attention mechanism of the Transformer model, the second derived vector group corresponding to each of the embedded vectors is determined. The second derived vector group is multiplied and summed to obtain the most relevant neighboring information corresponding to each of the windshield pump operation features. The most relevant neighboring information is the association information that is adjacent to the position of the current windshield pump operation feature. Based on the global most relevant information and the adjacent most relevant information, the correlation between the operating features of the windbag pump is obtained, and the feature space of the windbag pump is constructed based on the correlation between the operating features of the windbag pump.

3. The method for detecting abnormal operation of a wind pump according to claim 2, characterized in that, The construction of the windbag pump feature space based on the correlation between the operating characteristics of the windbag pump specifically includes: A global correlation matrix is ​​obtained based on the most relevant global information, and a local correlation matrix is ​​obtained based on the most relevant neighboring information. Based on the operating characteristics of the wind bag pump, a corresponding weight combination is determined. Based on the weight combination, the global correlation matrix and the local correlation matrix are weighted to obtain an integrated correlation matrix. The integrated association matrix is ​​compared with a preset matrix threshold to filter out associations that do not meet the association conditions. The remaining associations are used to generate a node graph to construct the feature space of the wind bag pump.

4. The method for detecting abnormal operation of a wind pump according to claim 3, characterized in that, The process of generating a node graph based on the remaining associations to construct the feature space of the windshield pump specifically includes: Based on the node diagram, the position of each node in the airbag pump is determined; And, determine the feature type corresponding to each of the nodes; Based on the location and the feature type, the correlation coefficient between each node is determined; The feature values ​​corresponding to each node are adjusted based on the correlation coefficient, and the feature space of the wind bag pump is constructed based on the adjusted feature values.

5. The method for detecting abnormal operation of a wind pump according to claim 1, characterized in that, The Transformer-based time series prediction model, together with the feature data in the feature space of the airbag pump, predicts the operating status of the airbag pump within a preset time period, obtaining predicted abnormal operating data of the airbag pump, specifically including: The feature data in the feature space of the windbag pump are input into the time series prediction model of the Transformer to obtain the predicted operation data of the windbag pump within a preset time period in the future. Based on the predicted operating data of the airbag pump, an operating data line chart is constructed, and the difference between adjacent data in the operating data line chart is determined. Based on the difference, the operating variation section and the operating stability section of the airbag pump are determined. The difference between adjacent data corresponding to the stable operating section of the windbag pump is compared with a first preset difference threshold to determine the first abnormal data. The difference between adjacent data corresponding to the change in the operation of the windbag pump is compared with a second preset difference threshold to identify the second abnormal data. Based on the first abnormal data and the second abnormal data, the predicted abnormal operation data of the wind bag pump is obtained.

6. The method for detecting abnormal operation of a wind pump according to claim 1, characterized in that, The step of combining the actual abnormal operation data of the airbag pump with the predicted abnormal operation data of the airbag pump to obtain an abnormal data set of the airbag pump, and determining the abnormal operation duration of the airbag pump based on the abnormal data set of the airbag pump, specifically includes: Arrange the actual abnormal operation data of the airbag pump and the predicted abnormal operation data of the airbag pump in chronological order, and mark the corresponding time points. The time difference between adjacent time points is determined, and the abnormal change rate of the airbag pump is determined based on the ratio between the later time difference and the earlier time difference; The abnormal change rate of the airbag pump is input into the airbag pump operating time prediction model, so as to output the abnormal operating time of the airbag pump through the airbag pump operating time prediction model.

7. The method for detecting abnormal operation of a wind pump according to claim 6, characterized in that, The step of comparing the abnormal data set of the airbag pump with a preset abnormal database to determine the degree of abnormality of the airbag pump specifically includes: The abnormal data set of the airbag pump is compared with a preset abnormal database to obtain a reference abnormal type; wherein, the preset abnormal database contains a variety of abnormal data sets of airbag pumps, and also includes a variety of abnormal types of airbag pumps corresponding to the abnormal data sets of airbag pumps respectively. The abnormal data difference between the abnormal data group of the wind bag pump and the data group corresponding to the reference abnormal type is determined, and a first weight is determined based on the abnormal data difference; The reference anomaly type is compared with a preset anomaly type level table to determine the second weight based on the comparison result. Based on the abnormal data group of the airbag pump, the first weight, and the second weight, the abnormality level value of the abnormal data group of the airbag pump is determined.

8. A device for detecting abnormal operation of a wind pump, characterized in that, The device includes a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to perform the method described in any one of claims 1-7.

9. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of performing the method described in any one of claims 1-7.

Images