A method for monitoring the operating status of an engine thermal test device

Abstract

This application relates to the field of condition monitoring technology, and in particular to a method for monitoring the operating condition of an engine thermal testing device. The method includes: clustering the operating patterns of historical operating data to obtain multiple operating types; using historical operating data of the operating type to which the real-time operating data belongs as test samples, and using the LOF algorithm to obtain the degree of anomaly of the real-time operating data under multiple candidate K values ​​within the test samples; using the median of the candidate K values ​​as the starting point, calculating the objective function for each step size, and using the step size at which the objective function is minimized as the target step size; obtaining multiple target K values ​​based on the target step size, and comparing the average degree of anomaly and the degree threshold of the real-time operating data under each target K value to obtain the condition monitoring result. The technical solution of this application can improve the accuracy of the operating condition monitoring results of the engine thermal testing device.

Description

Technical Field

[0001] The present application relates to the technical field of condition monitoring, and in particular to a method for monitoring the operating condition of an engine thermal testing device. Background Art

[0002] Engine thermal test equipment is a key device used to evaluate the performance, heat dissipation capacity, durability and reliability of an engine in a high temperature environment. In the process of using the engine thermal test equipment to test the engine, the operating status of the equipment directly affects the accuracy of the test results. Therefore, accurately obtaining the operating status of the engine thermal test equipment is a problem that needs to be solved urgently.

[0003] At present, a patent application document with publication number CN118568620A discloses a method for identifying the operating status of a high-voltage line based on the LOF algorithm, including: collecting operating data of multiple high-voltage cables through multiple types of sensors within a specific time range; first, cleaning and detecting the collected data and processing missing values, abnormal values ​​and error values ​​in the data to ensure data quality; secondly, standardizing the data to ensure that the data features have similar numerical ranges; finally, reducing the dimensionality of high-dimensional data to reduce computational complexity; taking the status data of each cable in the reduced-dimensional data as a data point for LOF calculation; combining the LOF algorithm with the ML algorithm and the isolation forest algorithm to identify the operating status of the high-voltage line; and statistically analyzing the operating status of the high-voltage line.

[0004] The above method first reduces the dimension of all collected operating data, and performs LOF calculation on the status data of each cable in all the data after dimension reduction, and then combines the LOF algorithm with the ML algorithm and the isolation forest algorithm to obtain the operating status. However, when the high-voltage line is in different working states (such as large load fluctuations), there will be large differences between the collected operating data. Performing LOF calculation on all the data after dimension reduction ignores the impact of the working state on the operating data, resulting in inaccurate operating status monitoring results. Summary of the invention

[0005] In order to solve the technical problem of inaccurate operating status monitoring results, the present application provides an operating status monitoring method for an engine thermal test equipment to improve the accuracy of the operating status monitoring results of the engine thermal test equipment.

[0006] In a first aspect of the present application, a method for monitoring the operating status of an engine thermal test device is provided, the monitoring method comprising: clustering operating modes of a plurality of historical operating data to obtain a plurality of operating types, the operating modes comprising first-order gradients and second-order gradients of each dimension in the historical operating data, the historical operating data comprising labeled data and unlabeled data; using the historical operating data of the operating type to which the real-time operating data belongs as a test sample, and in the test sample, using the LOF algorithm to obtain the degree of abnormality of the real-time operating data under a plurality of candidate K values; using the median of the candidate K values ​​as a starting point, calculating the objective function of each step length, and taking the step length when the objective function takes the minimum value as the target step length; step length The objective function value is :

[0007] , , and are the minimum, median and maximum values ​​of the candidate K values, respectively. , and The intervals , and The fitted slope of the degree of internal anomaly, is the preset coefficient, is the number of labeled data in the test sample, To label data In the candidate K value The degree of abnormality, To label data Abnormality degree label; obtain multiple target K values ​​according to the target step length, compare the average abnormality degree and degree threshold of the real-time operation data under each target K value, and obtain the status monitoring result.

[0008] The operation modes of multiple historical operation data are clustered to obtain multiple operation types, and the historical operation data under each operation type are obtained; the historical operation data of the operation type to which the real-time operation data belongs is used as a test sample, and in the test sample, the LOF algorithm is used to obtain the abnormality of the real-time operation data under multiple candidate K values, so as to eliminate the influence of different operation modes on the operation data and improve the accuracy of anomaly detection; further, the curve composed of the abnormality under multiple candidate K values ​​can be divided into an early growth segment, a stable segment and a late growth segment, and the candidate K value corresponding to the stable segment can better distinguish normal points from abnormal points, therefore, the candidate K value corresponding to the stable segment can be used to distinguish normal points from abnormal points. The median of the K value is selected as the starting point, and the objective function of each step is calculated. The objective function evaluates the division effect of each step from two aspects: whether the early growth segment, the stable segment and the late growth segment can be accurately divided, and whether the candidate K value of the stable segment can accurately predict the abnormality of the labeled data. The smaller the value of the objective function, the better the division effect of the corresponding step; the step length when the objective function takes the minimum value is taken as the target step length; multiple target K values ​​are obtained based on the target step length, and the average abnormality and degree threshold of the real-time operating data under each target K value are compared to obtain the status monitoring results, thereby improving the accuracy of the operating status monitoring results of the engine thermal test equipment.

[0009] Preferably, the dimension The first-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The value of; dimension The second-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The numerical value of .

[0010] The first-order gradient is used to characterize the change amount of the corresponding dimension when collecting historical operating data, and the second-order gradient is used to characterize the change speed of the corresponding dimension when collecting historical operating data; the change amount and change speed of all dimensions can characterize the working state of the engine thermal test equipment when collecting historical operating data, and realize the accurate quantification of the working state of the engine thermal test equipment.

[0011] Preferably, clustering the operation modes of multiple historical operation data to obtain multiple operation types includes: clustering the operation modes of the historical operation data using a Kmeans algorithm, and determining the number of clusters using an elbow method to obtain multiple clusters, wherein the multiple clusters correspond to multiple operation types.

[0012] The clustering algorithm is used to divide the operation mode into multiple operation types, and the historical operation data of each operation type is obtained to facilitate the subsequent determination of test samples for real-time operation data.

[0013] Preferably, one operation type corresponds to one standard pattern vector, and the standard pattern vector is the average operation pattern of historical operation data belonging to the operation type. Taking the historical operation data of the operation type to which the real-time operation data belongs as a test sample includes: calculating the similarity between the operation pattern of the real-time operation data and the standard pattern vector of each operation type, and taking the operation type corresponding to the maximum similarity value as the operation type to which the real-time operation data belongs.

[0014] Among all the historical operation data of the operation type to which the real-time operation data belongs, the LOF algorithm is calculated on the real-time operation data, which can eliminate the influence of different operation modes on the operation data and improve the accuracy of anomaly detection.

[0015] Preferably, in the LOF algorithm, the method for calculating the data distance between the running data includes: in response to the number of labeled data in the test sample being less than a preset number, taking the sum of the absolute values ​​of the differences in each dimension as the data distance between the running data; in response to the number of labeled data in the test sample being not less than a preset number, calculating the total variance of the abnormality degree labels of all labeled data in the test sample; calculating the conditional variance of any dimension, taking the difference between the total variance and the conditional variance as the abnormal sensitivity of each dimension, and performing weighted summation on the absolute values ​​of the differences in each dimension according to the abnormal sensitivity to obtain the data distance between the running data; wherein, the dimension conditional variance of for: , For Dimension All values ​​of is the dimension of the labeled data of the test sample The value is The proportion of is the dimension in the test sample The value is The variance of the anomaly labels of the labeled data.

[0016] When the number of annotated data is not less than a preset number, the abnormal sensitivity of each dimension under the operation type to which the real-time operation data belongs can be accurately obtained based on the annotated data, so that the data distance can reflect the operation status of the engine thermal test equipment.

[0017] Preferably, in response to the number of annotated data in the test sample being not less than a preset number, the running data and operation data The data distance between for: , For Dimension abnormal sensitivity, is the sum of abnormal sensitivity of all dimensions, and The operation data and operation data Medium Dimension The value of is the number of dimensions in the run data.

[0018] Preferably, the preset coefficient is the proportion of the labeled data in the test sample.

[0019] The more annotated data there is, the more abnormality labels the annotated data can provide, and the more effective supervisory information can be provided. Therefore, the value of the preset coefficient is increased to make full use of the supervisory information provided by the annotated data.

[0020] Preferably, after obtaining the status monitoring result, the monitoring method further comprises: using the average abnormality degree as an abnormality degree label and storing the real-time operation data as annotated data.

[0021] After obtaining the status monitoring results, the real-time operation data is stored as annotated data, which realizes the continuous updating of historical operation data and continuously increases the amount of annotated data in the historical operation data.

[0022] The technical solution of this application has the following beneficial technical effects:

[0023] The operation modes of multiple historical operation data are clustered to obtain multiple operation types, and the historical operation data under each operation type are obtained; the historical operation data of the operation type to which the real-time operation data belongs is used as a test sample, and in the test sample, the LOF algorithm is used to obtain the abnormality of the real-time operation data under multiple candidate K values, so as to eliminate the influence of different operation modes on the operation data and improve the accuracy of anomaly detection; further, the curve composed of the abnormality under multiple candidate K values ​​can be divided into an early growth segment, a stable segment and a late growth segment, and the candidate K value corresponding to the stable segment can better distinguish normal points from abnormal points, therefore, the candidate K value corresponding to the stable segment can be used to distinguish normal points from abnormal points. The median of the K value is selected as the starting point, and the objective function of each step is calculated. The objective function evaluates the division effect of each step from two aspects: whether the early growth segment, the stable segment and the late growth segment can be accurately divided, and whether the candidate K value of the stable segment can accurately predict the abnormality of the labeled data. The smaller the value of the objective function, the better the division effect of the corresponding step; the step length when the objective function takes the minimum value is taken as the target step length; multiple target K values ​​are obtained based on the target step length, and the average abnormality and degree threshold of the real-time operating data under each target K value are compared to obtain the status monitoring results, thereby improving the accuracy of the operating status monitoring results of the engine thermal test equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0025] Figure 1 It is a flow chart of a method for monitoring the operating status of an engine thermal test device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0027] It should be understood that when the terms "first", "second", etc. are used in the claims, specification and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the specification and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0028] According to a first aspect of the present application, the present application provides a method for monitoring the operating status of an engine thermal test device. Figure 1 FIG. 1 is a flow chart of a method for monitoring the operating status of an engine thermal test device according to an embodiment of the present application. Figure 1 As shown, the operating status monitoring method of the engine thermal testing equipment includes steps S101 to S104, which are described in detail below.

[0029] S101, clustering operation modes of a plurality of historical operation data to obtain a plurality of operation types, wherein the operation mode includes a first-order gradient and a second-order gradient of each dimension in the historical operation data, and the historical operation data includes labeled data and unlabeled data.

[0030] In one embodiment, the historical operation data is the operation data of the engine thermal test equipment collected at historical moments, and one historical operation data includes multiple dimensions such as temperature, pressure, power, and amplitude. The first-order gradient and the second-order gradient of each dimension in the historical operation data are calculated, and the vector composed of the first-order gradient and the second-order gradient of each dimension is used as the operation mode of the historical operation data.

[0031] The first-order gradient is used to characterize the change amount of the corresponding dimension when collecting historical operation data, and the second-order gradient is used to characterize the change speed of the corresponding dimension when collecting historical operation data; the change amount and change speed of all dimensions are used as the operation mode of the historical operation data, and one operation model corresponds to a working state of the engine thermal test equipment. The first-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The value of; dimension The second-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The numerical value of .

[0032] In one embodiment, clustering the operation modes of multiple historical operation data to obtain multiple operation types includes: clustering the operation modes of the historical operation data using a Kmeans algorithm, and determining the number of clusters using an elbow method to obtain multiple clusters, wherein the multiple clusters correspond to multiple operation types.

[0033] Among them, one operation type includes multiple historical operation data, and multiple historical operation data can be divided into unlabeled data and labeled data. Labeled data refers to historical operation data with abnormality degree labels, and unlabeled data refers to historical operation data without abnormality degree labels.

[0034] It should be noted that all historical operation data are stored in the database. When the method is put into use for the first time, all historical operation data are unlabeled data. As the method is put into use for a longer time, more historical operation data with abnormality degree labels can be collected. At this time, the database includes both labeled data and unlabeled data.

[0035] In this way, the historical operation data is divided into a plurality of operation types, and one operation model corresponds to a working state of the engine thermal test equipment.

[0036] S102, taking historical operation data of the operation type to which the real-time operation data belongs as a test sample, and using the LOF algorithm in the test sample to obtain the abnormality degree of the real-time operation data under multiple candidate K values.

[0037] In one embodiment, during the process of using the engine thermal test device to test the engine, the real-time operation data of the engine thermal test device at the current moment and the operation mode of the real-time operation data are collected. The current moment is recorded as , then the dimension in the operation mode of real-time operation data The first-order gradient for: , Dimensions in real-time operation data The numerical value of For the moment Collection time dimension The value, dimension The second-order gradient of is not described here.

[0038] Specifically, one operation type corresponds to one standard pattern vector, and the standard pattern vector is the average operation pattern of historical operation data belonging to the operation type. Using the historical operation data of the operation type to which the real-time operation data belongs as a test sample includes: calculating the similarity between the operation pattern of the real-time operation data and the standard pattern vector of each operation type, and taking the operation type corresponding to the maximum similarity value as the operation type to which the real-time operation data belongs.

[0039] The similarity is calculated using a similarity calculation method based on Euclidean distance.

[0040] After determining the operation type to which the real-time operation data belongs, all historical operation data of the operation type to which the real-time operation data belongs are used as test samples. The LOF algorithm is calculated on the real-time operation data in the test samples, which can eliminate the impact of different operation modes on the operation data and improve the accuracy of anomaly detection.

[0041] All test samples and real-time operation data are treated as a data distribution, and the LOF algorithm is used to obtain the abnormality of the real-time operation data under multiple candidate K values. The candidate K value range is .

[0042] Among them, the LOF algorithm (Local Outlier Factor) is an anomaly detection algorithm based on data distribution, which can output the degree of anomaly of any data in a data distribution. The LOF algorithm believes that the local density of non-outlier points is similar to the local density of their neighborhood points, while the local density of outliers is significantly different from the local density of their neighborhood points; in the LOF algorithm, the selection of K value will affect the number of neighborhood points and the calculation results of local density, which will lead to different degrees of anomaly output for different K values ​​for the same data.

[0043] In the LOF algorithm, local density is related to the data distance between the operating data. Since the sensitivity of each dimension in the operating data to the operating state of the engine thermal test equipment is different, in order to accurately obtain the abnormality of each operating data, when calculating the data distance, a larger weight should be assigned to the dimension with a larger sensitivity. Specifically, in the LOF algorithm, the calculation method of the data distance between the operating data includes: in response to the number of annotated data in the test sample being less than a preset number, the sum of the absolute values ​​of the differences in each dimension is used as the data distance between the operating data; in response to the number of annotated data in the test sample being not less than a preset number, the total variance of the abnormality degree labels of all annotated data in the test sample is calculated; the conditional variance of any dimension is calculated, and the difference between the total variance and the conditional variance is used as the abnormal sensitivity of each dimension. The absolute value of the difference in each dimension is weighted and summed according to the abnormal sensitivity to obtain the data distance between the operating data.

[0044] Among them, the dimension conditional variance of for: , For Dimension All values ​​of is the dimension of the labeled data of the test sample The value is The proportion of is the dimension in the test sample The value is The variance of the anomaly labels of the labeled data.

[0045] The preset number is one half of the labeled data in the test sample.

[0046] Among them, the dimension The conditional variance is regardless of the dimension The variance of the abnormality label, the total variance and the dimension The larger the difference in conditional variance, the less consideration of the dimension. After that, the variance of the abnormality degree label becomes smaller, and the dimension The greater the impact on the abnormality label, the greater the dimension The greater the abnormal sensitivity of Therefore, in response to the number of labeled data in the test sample being not less than the preset number, the running data and operation data The data distance between for: , For Dimension abnormal sensitivity, is the sum of abnormal sensitivity of all dimensions, and The operation data and operation data Medium Dimension The value of is the number of dimensions in the run data.

[0047] It can be understood that the calculation formula of the data distance is determined based on the annotated data in the test sample. When the number of annotated data is not less than the preset number, the abnormal sensitivity of each dimension under the operation type to which the real-time operation data belongs can be accurately obtained based on the annotated data, so that the data distance can reflect the operation state of the engine thermal test equipment. Among them, the abnormal sensitivity of each dimension is different under different operation types.

[0048] In this way, in the test samples of the operation type to which the real-time operation data belongs, the LOF algorithm is used to calculate the degree of abnormality under multiple candidate K values, eliminating the influence of different operation modes on the operation data and improving the accuracy of anomaly detection.

[0049] S103, taking the median of the candidate K values ​​as the starting point, calculating the objective function of each step length, and taking the step length when the objective function takes the minimum value as the target step length.

[0050] In one embodiment, in the process of using the LOF algorithm to obtain the degree of abnormality of real-time operation data under multiple candidate K values, as the candidate K value increases, the number of neighborhood points of the real-time operation data gradually increases, and the degree of abnormality of the real-time operation data gradually increases; when the candidate K value increases to a certain value, the degree of abnormality of the real-time operation data tends to be stable, and the degree of abnormality at this time can better distinguish normal points from abnormal points; when the candidate K value exceeds a certain value and continues to increase, the degree of abnormality of the real-time operation data will further increase until it is expanded to all test samples, and the degree of abnormality will gradually approach 1, and the degree of abnormality at this time cannot distinguish abnormal points from normal points; that is, the curve composed of the degree of abnormality under multiple candidate K values ​​can be divided into an early growth segment, a stable segment, and a late growth segment, and the candidate K value corresponding to the stable segment can better distinguish normal points from abnormal points. Therefore, in order to ensure the accuracy of the state monitoring results, it is necessary to select the candidate K value in the stable segment as the target K value.

[0051] Specifically, the median of the candidate K values ​​is used as the starting point, and the step size of the starting point is set , according to the starting point and step size Multiple candidate K values ​​can be divided into , and There are three intervals in total, among which, , and are the minimum, median and maximum values ​​of the candidate K values, respectively, and the step size is any positive integer.

[0052] Using the least squares method to find the interval The abnormal degree within the interval is fitted by straight line, and the interval The fitted slope of , The absolute value of The changing trend of the abnormal degree within The larger the absolute value of The faster the change of the abnormal degree is; similarly, we can get and The changing trends of the abnormality degree are recorded as and .when approaches 0, and and The larger the minimum absolute value is, the larger the step length is. Can accurately divide the early growth stage, stable stage and late growth stage, step length The better the division effect.

[0053] Furthermore, according to the stable phase The candidate K value in the calculation calculates the average abnormality of the labeled data in the test sample, and the absolute value of the difference between the average abnormality and the abnormality label is used as the evaluation step length Part of the division effect, the smaller the absolute value of the difference, the more stable the segment The candidate K value in can more accurately predict the abnormality of real-time operation data.

[0054] In summary, the objective function of each step size is calculated. The smaller the value of the objective function, the better the division effect of the corresponding step size. The objective function value is :

[0055] , , and are the minimum, median and maximum values ​​of the candidate K values, respectively. , and The intervals , and The fitted slope of the degree of internal anomaly, is the preset coefficient, is the number of labeled data in the test sample, To label data In the candidate K value The degree of abnormality, To label data The abnormality level label.

[0056] The preset coefficient value is 0.5.

[0057] In another embodiment, the preset coefficient is the proportion of the number of annotated data in the test sample. The more the number of annotated data, the more abnormality labels the annotated data can provide, and the more effective supervision information can be provided. Assign greater weight.

[0058] In this way, the target step size with the best division effect is obtained. The target step size can accurately divide the degree of abnormality under multiple candidate K values ​​into the early growth segment, the stable segment and the late growth segment, and the degree of abnormality of the real-time operation data can be accurately predicted based on the candidate K values ​​in the stable segment.

[0059] S104, obtaining multiple target K values ​​according to the target step length, comparing the average abnormality degree and degree threshold of the real-time operation data under each target K value, and obtaining the status monitoring result.

[0060] In one embodiment, the multiple target K values ​​are in the range For all candidate K values ​​in the test sample, calculate the average abnormality of the real-time operation data under each target K value; compare the average abnormality with the degree threshold, and if the average abnormality is greater than the degree threshold, the state monitoring result is abnormal, otherwise, the state monitoring result is normal. The degree threshold is 0.5.

[0061] In one embodiment, after obtaining the status monitoring result, the monitoring method further includes: using the average abnormality level as an abnormality level label and storing the real-time operation data as annotated data.

[0062] In this way, after obtaining the status monitoring results, the real-time operation data is stored as annotation data, realizing the continuous updating of historical operation data.

[0063] The above introduces the technical principles and implementation details of an operating status monitoring method for an engine thermal test equipment of the present application through specific embodiments. Cluster the operating modes of multiple historical operating data to obtain multiple operating types, and obtain historical operating data under each operating type; use the historical operating data of the operating type to which the real-time operating data belongs as test samples, and use the LOF algorithm in the test samples to obtain the degree of abnormality of the real-time operating data under multiple candidate K values, eliminate the influence of different operating modes on the operating data, and improve the accuracy of abnormality detection; further, the curve formed by the degree of abnormality under multiple candidate K values ​​can be divided into an early growth segment, a stable segment and a late growth segment, and the candidate K value corresponding to the stable segment can better distinguish between normal points and abnormal points, therefore, the candidate K value corresponding to the stable segment can better distinguish between normal points and abnormal points. The median of the K value is selected as the starting point, and the objective function of each step is calculated. The objective function evaluates the division effect of each step from two aspects: whether the early growth segment, the stable segment and the late growth segment can be accurately divided, and whether the candidate K value of the stable segment can accurately predict the abnormality of the labeled data. The smaller the value of the objective function, the better the division effect of the corresponding step; the step length when the objective function takes the minimum value is taken as the target step length; multiple target K values ​​are obtained based on the target step length, and the average abnormality and degree threshold of the real-time operating data under each target K value are compared to obtain the status monitoring results, thereby improving the accuracy of the operating status monitoring results of the engine thermal test equipment.

[0064] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A method for monitoring the operating status of an engine thermal test device, characterized in that: The monitoring method comprises: clustering operation modes of a plurality of historical operation data to obtain a plurality of operation types, wherein the historical operation data is operation data of an engine thermal test device in a plurality of dimensions collected at a historical moment, including labeled data and unlabeled data, wherein the plurality of dimensions include but are not limited to temperature, pressure, power and amplitude, and the operation mode comprises a first-order gradient and a second-order gradient of each dimension in the historical operation data; Dimensions The first-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The value of Dimensions The second-order gradient for: , For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The numerical value of For historical moments Dimensions in the collected historical operation data The value of The historical operation data of the operation type to which the real-time operation data belongs is used as a test sample, and the LOF algorithm is used in the test sample to obtain the abnormality degree of the real-time operation data under multiple candidate K values; Take the median of the candidate K value as the starting point, calculate the objective function of each step length, and take the step length when the objective function takes the minimum value as the target step length; step length The objective function value is : , , and are the minimum, median and maximum values ​​of the candidate K values, respectively. , and The intervals , and The fitted slope of the degree of internal anomaly, is the preset coefficient, is the number of labeled data in the test sample, To label data In the candidate K value The degree of abnormality, To label data The abnormality degree label; The target step size can accurately divide the abnormality degree under multiple candidate K values ​​into the early growth stage, the stable stage and the late growth stage, select the candidate K value in the stable stage as the target K value, compare the average abnormality degree and degree threshold of the real-time operation data under each target K value, and obtain the status monitoring result.

2. The operating status monitoring method of an engine thermal testing device according to claim 1, characterized in that: Clustering the operation modes of multiple historical operation data to obtain multiple operation types includes: clustering the operation modes of the historical operation data using a Kmeans algorithm, and determining the number of clusters using an elbow method to obtain multiple clusters, wherein the multiple clusters correspond to multiple operation types.

3. The operating status monitoring method of an engine thermal testing device according to claim 1, characterized in that: One operation type corresponds to one standard mode vector, and the standard mode vector is an average operation mode of historical operation data belonging to the operation type. The historical operation data of the operation type to which the real-time operation data belongs is used as a test sample, including: The similarity between the operation mode of the real-time operation data and the standard mode vector of each operation type is calculated, and the operation type corresponding to the maximum similarity value is taken as the operation type to which the real-time operation data belongs.

4. The operating status monitoring method of an engine thermal testing device according to claim 1, characterized in that: In the LOF algorithm, the method for calculating the data distance between running data includes: In response to the number of annotated data in the test sample being less than a preset number, the sum of the absolute values ​​of the differences in each dimension is used as the data distance between the running data; in response to the number of annotated data in the test sample being not less than a preset number, the total variance of the abnormality degree labels of all annotated data in the test sample is calculated; the conditional variance of any dimension is calculated, and the difference between the total variance and the conditional variance is used as the abnormal sensitivity of each dimension. The absolute values ​​of the differences in each dimension are weighted and summed according to the abnormal sensitivity to obtain the data distance between the running data; wherein, the dimension conditional variance of for: , For Dimension All values ​​of is the dimension of the labeled data of the test sample The value is The proportion of is the dimension in the test sample The value is The variance of the anomaly labels of the labeled data.

5. The operating status monitoring method of an engine thermal testing device according to claim 4, characterized in that: In response to the number of labeled data in the test sample being not less than a preset number, the data is run and operation data The data distance between for: , For Dimension abnormal sensitivity, is the sum of abnormal sensitivity of all dimensions, and The operation data and operation data Medium Dimension The value of is the number of dimensions in the run data.

6. The operating status monitoring method of an engine thermal testing device according to claim 1, characterized in that: The preset coefficient is the ratio of the number of labeled data in the test sample.

7. The operating status monitoring method of an engine thermal testing device according to any one of claims 1 to 6, characterized in that: After obtaining the status monitoring result, the monitoring method further includes: using the average abnormality degree as an abnormality degree label and storing the real-time operation data as labeled data.

Images