An abnormality judgment method and device for an oil delivery pump unit and a storage medium

Abstract

The application provides an oil delivery pump unit abnormality judgment method and device and a storage medium, and belongs to the technical field of fault detection; the first quantity of monitoring data related to a to-be-monitored variable generated by a target oil delivery pump unit is input into a pre-trained oil delivery pump unit abnormality recognition model to obtain to-be-monitored variable prediction data, the difference degree between actual operation data and the to-be-monitored variable prediction data is compared, if the difference degree exceeds a threshold value, the to-be-monitored oil delivery pump unit can be considered to be in an abnormal working state, and a warning is given; the application can more timely and accurately judge whether the oil delivery pump unit is in an abnormal working state.

Description

Technical Field

[0001] This invention mainly relates to the field of fault detection technology, specifically to a method, device and storage medium for judging abnormalities in an oil pump unit. Background Technology

[0002] Oil pump units are critical equipment in pipeline oil transportation, pressurizing the crude oil transported to ensure adequate pressure for delivery. Failures in oil pump units are common during production. In current techniques, when a pump unit at an oil pumping station malfunctions, the initial analysis and attempts to resolve the problem are typically based on the experience of on-site workers. This approach relies heavily on the experience of production personnel, lacks universality, and is prone to diagnostic errors. Therefore, it is necessary to provide an accurate and convenient method for determining whether an oil pump unit is in an abnormal operating state. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method, device and storage medium for judging abnormalities of oil pump units, in order to overcome the shortcomings of the prior art.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for judging abnormalities in an oil pump unit, comprising the following steps:

[0005] A first number of monitoring data related to the variable to be monitored generated by the target oil pump unit are obtained, and the monitoring data are input into a pre-trained oil pump unit anomaly identification model, and the predicted data of the variable to be monitored are output; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions;

[0006] Import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the degree of difference between the actual operating data and the predicted data of the variable to be monitored;

[0007] Determine whether the degree of difference exceeds a set first threshold. If it exceeds the first threshold, determine that the target oil pump unit is in an abnormal working state and issue a warning message based on the abnormal working state.

[0008] Furthermore, the process of acquiring a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit includes:

[0009] Based on the mechanical structure of the target oil pump unit and the analysis results of common failure causes of the target oil pump unit, a fault tree is constructed. The fault tree is used to describe the failure types of the target oil pump unit and the corresponding failure causes of the failure types. The failure types include failure attributes.

[0010] A second number of fault categories are set for the target oil pump unit based on the fault attributes of the fault type;

[0011] Correlation analysis of the monitored variables is performed on all fault causes using the second number of fault categories, and a first number of monitoring data related to the monitored variables are selected from the correlation analysis results.

[0012] Furthermore, each of the aforementioned fault categories includes multiple fault causes of the same type;

[0013] The process of performing correlation analysis on all fault causes using the second number of fault categories to identify the monitored variables, and then selecting the first number of monitoring data points related to the monitored variables from the correlation analysis results, includes:

[0014] For any fault cause, the monitoring dataset related to any fault cause is determined through the fault tree, and a data table is obtained to describe the fault cause and the related monitoring dataset.

[0015] From the data table, determine the first monitoring dataset related to the fault cause corresponding to the variable to be monitored;

[0016] A correlation analysis is performed between any one monitoring data point in the first monitoring dataset and the variable to be monitored to obtain multiple correlation analysis results;

[0017] Each correlation analysis result is compared with a pre-set second threshold, and monitoring data in the first monitoring dataset whose correlation analysis results are less than the second threshold are removed, thereby obtaining a first number of monitoring data related to the variable to be monitored, and the value corresponding to the first number is less than the number of monitoring data included in the first monitoring dataset.

[0018] Furthermore, after obtaining the data table used to describe the causes of the failures and the associated monitoring dataset, it includes:

[0019] Based on the fault causes included in the second number of fault categories, the predicted data of the monitored variables are constructed into a time series of prediction results. pred .

[0020] Furthermore, the prediction data for the variable to be monitored includes the time series of prediction results. pred ;

[0021] The process of importing the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and comparing the degree of difference between the actual operating data and the predicted data of the variable to be monitored, includes:

[0022] Import the actual operating data corresponding to the monitored variables generated by the target oil pump unit within a predetermined time range. The actual operating data includes the time series data s of the monitored variables. real ;

[0023] Using the time series data of the variable to be monitored s real and the time series of the prediction results s pred Construct a deviation vector, where the deviation vector Δs is represented as Δs = |s| real -s pred |1, the symbol l represents the norm;

[0024] The deviation vector Δs is tested using the ordinal probability ratio test method to obtain the degree of difference.

[0025] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: an abnormality judgment device for an oil pump unit, comprising:

[0026] The prediction result acquisition module is used to acquire a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit, and input the monitoring data into a pre-trained oil pump unit anomaly identification model, and output the predicted data of the variable to be monitored; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions;

[0027] The difference degree acquisition module is used to import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the difference degree between the actual operating data and the predicted data of the variable to be monitored;

[0028] An abnormal working state determination module is used to determine whether the degree of difference exceeds a set first threshold. If it exceeds the first threshold, the target oil pump unit is determined to be in an abnormal working state, and a warning message is issued based on the abnormal working state.

[0029] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: an abnormal judgment device for an oil pump unit, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the abnormal judgment method for the oil pump unit as described above.

[0030] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the abnormal judgment method of the oil pump unit as described above is implemented.

[0031] The beneficial effects of this invention are as follows: by inputting the first number of monitoring data related to the variable to be monitored generated by the target oil pump unit into the pre-trained oil pump unit anomaly identification model, the predicted data of the variable to be monitored is obtained. The difference between the actual operating data and the predicted data of the variable to be monitored is compared. If the difference exceeds the threshold, the oil pump unit to be monitored can be regarded as being in an abnormal working state and an early warning is issued. This invention can determine whether the oil pump unit is in an abnormal working state in a timely and accurate manner. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the abnormality judgment method for an oil pump unit provided in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram illustrating the prediction results of a vibration failure case of an oil pump unit's pump bearing in a real-world scenario, provided by an embodiment of the present invention.

[0034] Figure 3 This is a schematic diagram illustrating the prediction results of a pump bearing vibration failure case in another real-world scenario provided by an embodiment of the present invention.

[0035] Figure 4 This is a block diagram of an abnormality detection device for an oil pump unit provided in an embodiment of the present invention. Detailed Implementation

[0036] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0037] Figure 1 This is a flowchart illustrating an anomaly detection method for an oil pump unit provided in an embodiment of this specification. From a programming perspective, the execution entity of the process can be a program hosted on a server of a service platform.

[0038] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for judging abnormalities in an oil pump unit, including:

[0039] S101. Obtain a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit, and input the monitoring data into a pre-trained oil pump unit anomaly identification model, and output the predicted data of the variable to be monitored; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions.

[0040] S102. Import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the degree of difference between the actual operating data and the predicted data of the variable to be monitored.

[0041] S103. Determine whether the degree of difference exceeds the set first threshold. If it exceeds the first threshold, determine that the target oil pump unit is in an abnormal working state and issue a warning message based on the abnormal working state.

[0042] In this embodiment, by inputting the first number of monitoring data related to the variable to be monitored generated by the target oil pump unit into the pre-trained oil pump unit anomaly identification model, the predicted data of the variable to be monitored is obtained. The difference between the actual operating data and the predicted data of the variable to be monitored is compared. If the difference exceeds the threshold, the oil pump unit to be monitored can be regarded as being in an abnormal working state and an early warning is issued. The present invention can determine whether the oil pump unit is in an abnormal working state in a timely and accurate manner.

[0043] It should be understood that in S101, the monitored variable can be interpreted as the oil pump unit frequently experiencing various abnormalities during actual operation, thus affecting normal production on-site. Common fault types of oil pump units can include low pressure after unit startup, centrifugal pump cavitation, reduced flow rate during operation, and excessively high pump body temperature, among others. Each fault type can include several specific causes. For example, the causes of low pressure after startup can include incorrect motor rotation direction, excessively low speed, excessive suction height, excessively low inlet pressure, inlet leakage, or gas inside the pump, resulting in insufficient oil filling in the suction pipeline. From the perspective of fault causes, typical abnormalities of the oil pump unit can be divided into abnormal unit parameters and abnormalities caused by component damage. Furthermore, abnormal unit parameters can include abnormalities such as motor overload due to low voltage or reduced flow rate during operation due to excessively low speed. These abnormalities can be caused by abnormal unit parameters such as speed, voltage, and inlet pressure, and the causes of these abnormal unit parameters are related to the specific equipment status on-site. Component damage-related anomalies, such as reduced flow rate due to internal pump wear or motor overload caused by bearing damage, arise from component wear during long-term operation or failure due to unexpected impacts within the oil pump unit. When an oil pump unit experiences an operational anomaly, its observable operating parameters typically show abnormal values. Therefore, in practical scenarios, several monitoring indicators can be pre-set. By monitoring one or more of these indicators, the anomaly status of the unit can be assessed. Thus, in this embodiment, the monitored variable can refer to pre-set monitoring indicators used to assist in assessing whether the unit is in an abnormal operating state.

[0044] In S101, the anomaly detection model can be a neural network model, specifically a machine learning model such as a convolutional neural network (CNN). Since the input layer of a neural network model requires the values ​​of several variables as input, the first quantity of monitoring data in this embodiment can refer to the specific values ​​of several variables used as the basis for obtaining the prediction results of the variable to be monitored. For each specific variable to be monitored, this embodiment can train a corresponding anomaly detection model and set corresponding monitoring data. For example, for the specific variable to be monitored, pump bearing vibration, the corresponding first quantity of monitoring data can include monitoring data such as pump end bearing temperature, motor end bearing temperature, motor bearing temperature, motor A-phase winding temperature, motor B-phase winding temperature, pump bearing vibration value, and motor bearing vibration value.

[0045] It should be noted that in this embodiment, the training data used in training the anomaly detection model for the oil pump unit is collected when the oil pump unit is in normal operation. Normal operation refers to the state where the unit under monitoring is operating without faults. In practical applications, the data initially collected for training the anomaly detection model is often noisy, incomplete, and inconsistent. The data can be cleaned before model training to improve its quality and ensure it meets the algorithm's specifications and requirements.

[0046] It should be understood that in S102, the actual operating data corresponding to the variable to be monitored can refer to the data value of the variable to be monitored that is actually measured. In actual production, most oil pump units use SCADA systems as monitoring systems. SCADA systems are industrial process monitoring systems that can monitor the operation of oil pump units and collect operating status parameter data of oil pump units based on industrial process monitoring systems. Therefore, in this embodiment, the actual operating data corresponding to the variable to be monitored can be monitored and obtained through the SCADA system, and then the degree of difference between the two can be compared. The specific form of the degree of difference can be a percentage or the difference between the two, and this embodiment does not make a specific limitation.

[0047] It should be understood that in S103, it is determined whether the degree of difference exceeds a set first threshold. The first threshold is a pre-set indicator for assessing whether the monitored oil pump unit is in an abnormal working state. If the degree of difference exceeds the first threshold, the monitored oil pump unit is likely in an abnormal working state and can be regarded as being in an abnormal working state. At this time, relevant detection and other activities can be performed. For a specific variable to be monitored, the specific value of the pre-determined first threshold can be flexibly set according to the attributes of the specific variable to be monitored. This embodiment does not impose specific limitations. Specifically, it can be set based on the historical maintenance data records of relevant maintenance personnel of the oil pump unit.

[0048] Figure 1 The method described herein involves inputting a first set of monitoring data related to the variable to be monitored, generated by the oil pump unit to be monitored, into a trained oil pump unit anomaly identification model to obtain a prediction result for the variable to be monitored. Since the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit under normal operating conditions, if the discrepancy between the actual measured operating data corresponding to the variable to be monitored and the prediction result exceeds a certain level, the oil pump unit to be monitored can be considered to be in an abnormal operating state. This achieves the technical effect of accurate and timely fault monitoring of the oil pump unit.

[0049] based on Figure 1 In addition to the method described above, this embodiment also provides some specific implementation schemes of the method, which will be explained below.

[0050] In an optional embodiment, the process of obtaining a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit includes: constructing a fault tree based on the mechanical structure of the target oil pump unit and the analysis results of common fault causes of the target oil pump unit. The fault tree is used to describe the fault types of the target oil pump unit and the fault causes corresponding to the fault types. The fault types include fault attributes.

[0051] In this embodiment, considering that the oil pump unit mainly consists of an oil pump, a motor, and a coupling, and that most current oil pump units use centrifugal pumps, this embodiment summarizes the common fault types and corresponding causes of the oil pump unit based on its mechanical structure, thereby facilitating the classification and analysis of oil pump unit faults. As shown in Table 1, Table 1 lists common fault types of the oil pump unit and their corresponding causes. To facilitate rapid analysis of the causes of oil pump unit faults, a fault tree for the oil pump unit can be established based on the fault types and corresponding causes listed in Table 1. Specifically:

[0052] In an optional embodiment, before acquiring several types of monitoring data related to the variable to be monitored generated by the oil pump unit to be monitored, it may include: constructing a fault tree to describe the fault types and corresponding fault causes of the oil pump unit to be monitored based on the mechanical structure of the oil pump unit to be monitored and the analysis results of common fault causes of the oil pump unit to be monitored.

[0053]

[0054]

[0055] From the perspective of fault characterization variables, the common faults given in Table 1 can be reclassified into temperature faults, pressure faults, vibration faults, sealing faults, and flow faults based on the type of fault characterization variables. Based on this classification method, a fault tree can be established as the basis for anomaly identification. If the oil pump unit fails, the cause of the fault can be analyzed according to the location of the monitoring point and the type of the variable.

[0056] Based on the fault tree established for the oil pump unit as described above, in an optional embodiment, after constructing the fault tree to describe the fault types and corresponding causes of the oil pump unit to be monitored, a second number of fault categories for the target oil pump unit are set according to the fault attributes of the fault types; specifically:

[0057] Based on the fault attributes of the fault types included in the fault tree, a second number of fault categories are set for the oil pump unit to be monitored, and any one of the second number of fault categories includes several fault causes of the same type.

[0058] In this embodiment, fault attributes can be used to describe the specific type of fault involved in the oil pump unit, such as faults related to current, pressure, temperature, or vibration. Then, a second number of fault categories can be set based on the fault attributes, and each fault category can have several similar fault causes. For example, in a practical scenario, as shown in Table 2, which lists the internal detection variables of the oil pump unit dataset, faults can be categorized into current-type faults, pressure-type faults, temperature-type faults, and vibration-type faults. Among these, vibration-type faults include four specific fault causes: pump end bearing vibration, pump waist bearing vibration, motor end bearing vibration, and motor waist bearing vibration.

[0059] Table 2:

[0060]

[0061] In some embodiments, after setting a first number of fault categories for the oil pump unit to be monitored, a correlation analysis of the variable to be monitored is performed on all fault causes using the second number of fault categories, and a first number of monitoring data related to the variable to be monitored are selected from the correlation analysis results. Specifically:

[0062] Since each of the aforementioned fault categories includes multiple fault causes of the same type, for any given fault cause, the fault tree is used to determine the monitoring dataset associated with that fault cause, resulting in a data table describing the fault cause and its associated monitoring dataset.

[0063] In this embodiment, a monitoring dataset related to each fault cause within each fault category is established. Taking pump bearing vibration, a specific variable to be monitored, as described above, as an example, abnormal pump bearing vibration can be considered a specific fault cause of the oil pump unit. The corresponding monitoring data can include pump end bearing temperature, motor end bearing temperature, motor bearing temperature, motor A-phase winding temperature, motor B-phase winding temperature, pump bearing vibration value, and motor bearing vibration value. In this embodiment, the correspondence between different fault causes and their specific related monitoring datasets can be presented in the form of a data table.

[0064] For machine learning models such as neural networks, the processed data is generally not directly fed into the model for learning. Instead, feature selection is performed first. This is because the monitoring variables in the dataset obtained in actual production usually have a certain correlation. This means that the dataset contains redundant information for machine learning models. This redundant information will cause the prediction results after model training to depend on this redundant information, resulting in a decrease in prediction accuracy. Therefore, feature extraction and selection of the dataset are required before training.

[0065] Therefore, in an optional embodiment of the present invention, the method for determining the data category of the first number of monitoring data related to the variable to be monitored may include: determining a first monitoring dataset related to the fault cause corresponding to the variable to be monitored from the data table; performing a correlation analysis on any one monitoring data in the first monitoring dataset with the variable to be monitored to obtain several correlation analysis results; comparing each correlation analysis result in the several correlation analysis results with a predetermined second threshold, and removing monitoring data in the first monitoring dataset whose correlation analysis results are less than the second threshold to obtain the first number of monitoring data, wherein the value corresponding to the first number is less than the number of monitoring data included in the first monitoring dataset.

[0066] In this embodiment, for the variable to be monitored, a set of more statistically significant features are selected from the first monitoring dataset using a feature selection method, thereby achieving the purpose of data dimensionality reduction and simplifying the training process of the anomaly recognition model.

[0067] In some embodiments, the anomaly identification model described above can be validated before being put into use to test its effectiveness in identifying abnormal states of the oil pump unit. Specifically, a specific variable to be monitored can be selected, and a series of actual operating data values ​​of the specific variable to be monitored under normal conditions can be obtained within a certain time period. Simultaneously, the anomaly identification model described above can be used to obtain a series of prediction results for the specific variable to be monitored within that specific time period. The consistency between the two sets of data in a Cartesian coordinate system is then assessed to test whether the anomaly identification model can accurately predict the variable to be monitored under normal conditions. Correspondingly, a series of actual operating data values ​​of the specific variable to be monitored under abnormal conditions can be obtained within another time period. The anomaly identification model described above can also be used to obtain a series of prediction results for the specific variable to be monitored within that specific time period. If the two sets of data do not match well in a Cartesian coordinate system, it indicates that the anomaly identification model described above can accurately identify anomalies in the variable to be monitored.

[0068] The following uses data from real-world scenarios to illustrate this, such as... Figure 2 and Figure 3 As shown, Figure 2 This is a schematic diagram illustrating the prediction results of a vibration failure case of an oil pump unit's pump bearing in a real-world scenario. Figure 3 This is a schematic diagram illustrating the predicted results of a pump bearing vibration failure case in another real-world scenario. Figure 2 and Figure 3 In the diagram, A represents actual data, and B represents predicted data. Figure 2 and Figure 3 It can be seen that under normal conditions, the prediction results of the anomaly identification model match the actual values ​​very well, while under fault conditions, the predicted values ​​of the model deviate significantly from the anomalies.

[0069] In an optional embodiment, the process of importing the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and comparing the degree of difference between the actual operating data and the predicted data of the variable to be monitored, includes:

[0070] Import the actual operating data corresponding to the monitored variables generated by the target oil pump unit within a predetermined time range. The actual operating data includes the time series data s of the monitored variables. real ;

[0071] Using the time series data of the variable to be monitored s real and the time series of the prediction results s pred Construct a deviation vector, where the deviation vector Δs is represented as Δs = |s| real -spred |1, the symbol l represents the norm;

[0072] The deviation vector Δs is tested using the ordinal probability ratio test method to obtain the degree of difference.

[0073] In this embodiment, the Sequential Probability Ratio Test (SPRT) method is used to test the deviation vector. The test process is described in detail below.

[0074] In the SPRT method, for monitoring data from SCADA systems (which may be in normal or abnormal states), two hypotheses are presented: Hypothesis H0: Under normal conditions; Hypothesis H1: Under abnormal conditions. Testing these two hypotheses requires pre-defining the distribution of the data under different hypotheses, generally assuming a normal distribution with two different parameters. For a given residual e1, e2, ..., e n Suppose the formulas for calculating the joint probability distributions of H0 and H1 are as follows:

[0075]

[0076]

[0077] The likelihood ratio is defined as:

[0078]

[0079] Define the false alarm rate α and the false alarm rate β respectively, and give the decision thresholds A and B required for fault warning:

[0080]

[0081]

[0082] For ease of use, all three formulas above are in logarithmic form, that is:

[0083]

[0084]

[0085]

[0086] Basis for judgment:

[0087] lnR≤lnA,

[0088] lnR≥lnB,

[0089] lnA < lnR < lnB.

[0090] Where, lnR ≤ lnA indicates acceptance of hypothesis H0, representing normal operating conditions; lnR ≥ lnB indicates acceptance of hypothesis H1, representing abnormal operating conditions; lnA ＜ lnR < lnB indicates that the test should continue.

[0091] Based on the above SPRT early warning method criteria, abnormal early warnings for oil pump units can be achieved, with the following assumptions made regarding the parameters (assuming early warning is given for the monitored variable X):

[0092] Algorithm: SPRT early warning for monitored variable X,

[0093] Input: Distribution of monitoring variable X under normal operating conditions (assuming the variable follows a normal distribution under normal operating conditions, i.e., input its mean and variance).

[0094] (1) For variable X under normal working conditions, it can be assumed that it follows a normal distribution. Then the inputs σ and μ are the variance and mean of the data of the monitored variable X under normal working conditions.

[0095] (2) For the distribution of abnormal operating conditions, the variance ratio is defined as ,

[0096]

[0097] In this embodiment, V = 1.5;

[0098] (3) For the distribution of abnormal working conditions, the average value is taken as the normal working condition average value with fluctuation above and below, and the fluctuation level is taken as three times the normal working condition average value. When the monitored variable increases under abnormal working conditions, it is taken as upward fluctuation, and when the monitored variable decreases under abnormal working conditions, it is taken as downward fluctuation.

[0099] (4) The false alarm rate α and the false alarm rate β in the SPRT test are both set to 0.1, that is, the significance level of the hypothesis test is 0.1.

[0100] In this embodiment, in order to specifically use the SPRT method for fault early warning of oil pump units, after obtaining the data table describing the fault causes and related monitoring datasets, the following may also be included:

[0101] Based on the fault causes included in the second number of fault categories, several variables to be monitored are set, and based on the several variables to be monitored, a sequence of variables to be monitored (i.e., the time series of prediction results s) is constructed. pred ).

[0102] In other words, this embodiment sequentially numbers the multiple variables to be monitored. Then, in order to apply the SPRT method, the predicted time series s is... pred Formally, it can be expressed as follows:

[0103]

[0104] In this formula, t represents the current time, i represents the index of the variable to be monitored, and so on. Let represent the predicted value of the monitored variable numbered i at the current time t, and let n represent the total amount of multiple monitored variables. This represents the predicted value of the i-th monitored variable at the j-th time point before the current time t. Similarly, the symbol... The predicted value of the i-th monitored variable at the (j-1)-th time before the current time t. Because in the anomaly detection model described above, predicting the value of a specific monitored variable at the current time requires certain specific monitoring data at the current time as input data for the anomaly detection model's input layer. Therefore, the symbol... The meaning can be understood as predicting the value of the monitored variable with prediction number i at the current time t. The required dataset of monitoring data is used as input data for the input layer of the anomaly detection model corresponding to number i. In this sense, the values ​​of the symbols k and i are not equal, i.e., k ≠ i. The symbol "|" indicates that the specific predicted value of the data listed before the symbol depends on the dataset consisting of the corresponding monitoring data following the symbol. Accordingly, the degree of anomaly of the deviation vector Δs described above can be formally expressed as follows, where anomaly means anomaly, i.e.

[0105]

[0106] like Figure 4 As shown, this embodiment of the invention provides an anomaly detection device for an oil pump unit, comprising:

[0107] The prediction result acquisition module is used to acquire a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit, and input the monitoring data into a pre-trained oil pump unit anomaly identification model, and output the predicted data of the variable to be monitored; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions;

[0108] The difference degree acquisition module is used to import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the difference degree between the actual operating data and the predicted data of the variable to be monitored;

[0109] An abnormal working state determination module is used to determine whether the degree of difference exceeds a set first threshold. If it exceeds the first threshold, the target oil pump unit is determined to be in an abnormal working state.

[0110] Optionally, the prediction result acquisition module acquires a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit, specifically:

[0111] Based on the mechanical structure of the target oil pump unit and the analysis results of common failure causes of the target oil pump unit, a fault tree is constructed. The fault tree is used to describe the failure types of the target oil pump unit and the corresponding failure causes of the failure types. The failure types include failure attributes.

[0112] A second number of fault categories are set for the target oil pump unit based on the fault attributes of the fault type;

[0113] Correlation analysis of the monitored variables is performed on all fault causes using the second number of fault categories, and a first number of monitoring data related to the monitored variables are selected from the correlation analysis results.

[0114] In an optional embodiment, the difference degree acquisition module imports the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compares the difference degree between the actual operating data and the predicted data of the variable to be monitored, specifically:

[0115] The prediction data for the variable to be monitored includes the time series of prediction results. real ;

[0116] Import the actual operating data corresponding to the monitored variables generated by the target oil pump unit within a predetermined time range. The actual operating data includes the time series data s of the monitored variables. real ;

[0117] Using the time series data of the variable to be monitored s real and the time series of the prediction results s pred Construct a deviation vector, where the deviation vector Δs is represented as Δs = |s| real -s pred |1, the symbol l represents the norm;

[0118] The deviation vector Δs is tested using the ordinal probability ratio test method to obtain the degree of difference.

[0119] This invention acquires data on monitored variables related to the normal operating status of the oil pump unit to be monitored, inputs this data into a trained oil pump unit anomaly identification model, and obtains a predicted result for the monitored variable. This predicted result is compared with the actual operating data generated by the oil pump unit corresponding to the monitored variable to determine the degree of difference between the operating data and the predicted result. The difference is then judged using the Sequential Probability Ratio Test (SPRT) method, enabling the monitoring and early warning of abnormal states of the oil pump unit based on process data. When an anomaly is detected in the unit, the fault tree is used to identify potentially problematic components of the oil pump unit, providing support for maintenance and repair. This invention overcomes the drawback of current alarm settings based on the unit's tolerance limits, which result in late detection of unit anomalies. It also saves the additional investment costs required to install high-frequency vibration sensors for early detection and assessment of unit anomalies, significantly improving personnel efficiency, reducing investment costs, and achieving early warning of unit anomalies. This is of great significance for improving the safety and reliability of oil transportation stations.

[0120] Another embodiment of the present invention provides an anomaly detection device for an oil pump unit, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the anomaly detection method for the oil pump unit as described above.

[0121] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the abnormal judgment method for the oil pump unit as described above.

[0122] The above-mentioned abnormality judgment device and storage medium for an oil pump unit can be found in the implementation details and beneficial effects of the above-described abnormality judgment method for an oil pump unit, which will not be repeated here.

[0123] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing anomalies in an oil pump unit, characterized in that, Includes the following steps: A first number of monitoring data related to the variable to be monitored generated by the target oil pump unit are obtained, and the monitoring data are input into a pre-trained oil pump unit anomaly identification model, and the predicted data of the variable to be monitored are output; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions; Import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the degree of difference between the actual operating data and the predicted data of the variable to be monitored; Determine whether the degree of difference exceeds a set first threshold. If it exceeds the first threshold, determine that the target oil pump unit is in an abnormal working state and issue a warning message based on the abnormal working state. The process of acquiring a first quantity of monitoring data related to the variable to be monitored generated by the target oil pump unit includes: Based on the mechanical structure of the target oil pump unit and the analysis results of common failure causes of the target oil pump unit, a fault tree is constructed. The fault tree is used to describe the failure types of the target oil pump unit and the corresponding failure causes of the failure types. The failure types include failure attributes. A second number of fault categories are set for the target oil pump unit based on the fault attributes of the fault type; Correlation analysis of the monitored variables is performed on all fault causes using the second number of fault categories, and a first number of monitoring data related to the monitored variables are selected from the correlation analysis results.

2. The method for judging abnormalities in an oil pump unit according to claim 1, characterized in that, Each of the aforementioned fault categories includes multiple fault causes of the same type; The process of performing correlation analysis on all fault causes using the second number of fault categories to identify the monitored variables, and then selecting the first number of monitoring data points related to the monitored variables from the correlation analysis results, includes: For any fault cause, the monitoring dataset related to any fault cause is determined through the fault tree, and a data table is obtained to describe the fault cause and the related monitoring dataset. From the data table, determine the first monitoring dataset related to the fault cause corresponding to the variable to be monitored; A correlation analysis is performed between any one monitoring data point in the first monitoring dataset and the variable to be monitored to obtain multiple correlation analysis results; Each correlation analysis result is compared with a pre-set second threshold, and monitoring data in the first monitoring dataset whose correlation analysis results are less than the second threshold are removed, thereby obtaining a first number of monitoring data related to the variable to be monitored, and the value corresponding to the first number is less than the number of monitoring data included in the first monitoring dataset.

3. The method for judging abnormalities in an oil pump unit according to claim 2, characterized in that, After obtaining the data table used to describe the causes of the failure and the related monitoring dataset, it includes: Based on the fault causes included in the second number of fault categories, the predicted data of the monitored variable are constructed into a time series of prediction results. .

4. The method for judging abnormalities in an oil pump unit according to claim 3, characterized in that, The process of importing the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and comparing the degree of difference between the actual operating data and the predicted data of the variable to be monitored, includes: Import the actual operating data corresponding to the monitored variables generated by the target oil pump unit within a predetermined time range. The actual operating data includes time series data of the monitored variables. ; Using the time series data of the variable to be monitored and the time series of the prediction results Construct a deviation vector, wherein the deviation vector Represented as ,symbol Represents the norm; The deviation vector is analyzed according to the ordinal probability ratio test method. The degree of difference was determined through testing.

5. An anomaly detection device for an oil pump unit, characterized in that, include: The prediction result acquisition module is used to acquire a first number of monitoring data related to the variable to be monitored generated by the target oil pump unit, and input the monitoring data into a pre-trained oil pump unit anomaly identification model, and output the predicted data of the variable to be monitored; wherein, the oil pump unit anomaly identification model is trained based on the monitoring dataset generated by the oil pump unit to be monitored under normal operating conditions; The difference degree acquisition module is used to import the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored, and compare the difference degree between the actual operating data and the predicted data of the variable to be monitored; An abnormal working state determination module is used to determine whether the degree of difference exceeds a set first threshold. If it exceeds the first threshold, the target oil pump unit is determined to be in an abnormal working state. In the prediction result acquisition module, a first quantity of monitoring data related to the variable to be monitored, generated by the target oil pump unit, is acquired, specifically: Based on the mechanical structure of the target oil pump unit and the analysis results of common failure causes of the target oil pump unit, a fault tree is constructed. The fault tree is used to describe the failure types of the target oil pump unit and the corresponding failure causes of the failure types. The failure types include failure attributes. A second number of fault categories are set for the target oil pump unit based on the fault attributes of the fault type; Correlation analysis of the monitored variables is performed on all fault causes using the second number of fault categories, and a first number of monitoring data related to the monitored variables are selected from the correlation analysis results.

6. The anomaly detection device for an oil pump unit according to claim 5, characterized in that, In the difference degree acquisition module, the actual operating data generated by the target oil pump unit corresponding to the variable to be monitored is imported, and the difference degree between the actual operating data and the predicted data of the variable to be monitored is compared, specifically: The prediction data for the variable to be monitored includes the time series of prediction results. ; Import the actual operating data corresponding to the monitored variables generated by the target oil pump unit within a predetermined time range. The actual operating data includes time series data of the monitored variables. ; Using the time series data of the variable to be monitored and the time series of the prediction results Construct a deviation vector, wherein the deviation vector Represented as ,symbol Represents the norm; The deviation vector is analyzed according to the ordinal probability ratio test method. The degree of difference was determined through testing.

7. An anomaly detection device for an oil pump unit, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the abnormal judgment method for the oil pump unit as described in any one of claims 1 to 4.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the abnormal judgment method for the oil pump unit as described in any one of claims 1 to 4.

Images