Method and device for identifying running trend of equipment component, equipment and medium

Abstract

The invention discloses an equipment component operation trend identification method and device, equipment and a medium, and relates to the technical field of equipment operation trend analysis, and the method comprises the steps: obtaining the real-time monitoring data of a current equipment component; inputting the real-time monitoring data into a target equipment operation state prediction model to predict a state estimation sample of a current equipment part based on the real-time monitoring data and the memory matrix, and obtaining a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding moment; according to the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window, performing normal state or abnormal state judgment on the real-time monitoring data under the current time window to obtain a corresponding state judgment result; and if the state judgment result is an abnormal state result, performing error contribution rate calculation on each abnormal feature data of the abnormal monitoring data, and judging the running trend of the current equipment component according to the error contribution rate. And accurate operation trend identification is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment operation trend analysis, in particular to an operation trend identification method and device for equipment components, equipment and medium. BACKGROUND

[0002] With the continuous advancement of rail transit system construction and the development of digital information, intelligent operation and maintenance platforms are widely used to realize real-time monitoring and data collection of rail transit vehicles / locos. At present, the online monitoring system monitors a large amount of trend data types, including vibration data, impact data, dB value, temperature, alarm, speed, etc. At the same time, the data volume is large, and there are problems such as noise in the data due to working conditions, transmission processing, etc. It is difficult for humans to timely discover the change of a certain monitoring quantity trend of the vehicle / locos in the operation process. Secondly, the abnormal judgment method by traditional threshold value is only effective for obvious deviated abnormal data, and it is difficult to timely discover the gradually changing and hidden abnormal data in a large amount of data. At present, in the existing trend identification method, simple methods such as statistical models are easy to implement but have poor robustness, and complex models such as deep learning have strong performance but depend on data and algorithms, which cannot meet the actual situation requirements. In addition, in the actual operation process, noise or data deviation can easily interfere with trend estimation, and as the component degradation process progresses, some gradually changing trends are difficult to identify, and the current trend change situation cannot be accurately quantified. Secondly, for multi-element data scenarios, trend anomalies may be caused by single variable or multiple variable changes, and the contribution of each variable to the trend change is difficult to explain.

[0003] In summary, how to solve the problems of data redundancy, false positives and false negatives caused by fixed threshold, poor adaptability of multi-dimensional data, and difficulty in quantifying gradual trends in trend identification is a technical problem to be solved in the field. SUMMARY

[0004] Therefore, the purpose of the present application is to provide an operation trend identification method, device, equipment and medium for equipment components, which can solve the problems of data redundancy, false positives and false negatives caused by fixed threshold, poor adaptability of multi-dimensional data, and difficulty in quantifying gradual trends in trend identification. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses an operation trend identification method for equipment components, comprising:

[0006] obtaining real-time monitoring data of a current equipment component;

[0007] inputting the real-time monitoring data into a target equipment operation state prediction model, so that the target equipment operation state prediction model predicts a state estimation sample of the current equipment component based on the real-time monitoring data and a memory matrix, and obtains a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time;

[0008] running state prediction model and according to the fusion similarity result, the dynamic similarity threshold parameter, and the sliding time window, judging the real-time monitoring data under the current time window to be in a normal state or an abnormal state to obtain a corresponding state judgment result;

[0009] If the state judgment result is an abnormal state result, error contribution rates of each abnormal feature data of the abnormal monitoring data are calculated to determine the operation trend of the current equipment component according to the error contribution rates.

[0010] Optionally, the operation trend identification method of the equipment component further includes:

[0011] obtaining historical normal operation data of the current equipment component, wherein the historical normal operation data includes a plurality of feature data associated with changes in the operation state of the current equipment component;

[0012] dividing a numerical range of the historical normal operation data by a preset step size to obtain a plurality of numerical range intervals;

[0013] selecting target feature data with a distance less than a preset distance threshold from a current interval endpoint from each of the numerical range intervals to construct a memory matrix.

[0014] Optionally, the historical normal operation data is multi-dimensional feature data including rotation speed data, temperature data, vibration data, operation mileage data, impact SV data, and impact dB data.

[0015] Optionally, the selecting target feature data with a distance less than a preset distance threshold from a current interval endpoint from each of the numerical range intervals to construct a memory matrix includes:

[0016] constructing all feature data at the same time as a historical monitoring sample, calculating a target distance between each feature data in the historical monitoring sample and a current interval endpoint in a numerical range interval in which the feature data is currently located, and retaining target feature data with a target distance less than a preset distance threshold;

[0017] counting current historical monitoring samples in which all target feature data is located, and deleting duplicate samples in the current historical monitoring samples to obtain target historical monitoring samples;

[0018] constructing a memory matrix based on the target historical monitoring samples; wherein the number of rows of the memory matrix is the number of the target historical monitoring samples, and the number of columns is the feature data dimension of the historical normal operation data.

[0019] Optionally, the obtaining a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time includes:

[0020] calculating a Euclidean distance and a cosine similarity between the state estimation sample and the real-time monitoring data at the corresponding time point;

[0021] determining a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time point based on the Euclidean distance and the cosine similarity.

[0022] Optionally, the determining the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time point based on the Euclidean distance and the cosine similarity comprises:

[0023] determining the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time point based on the Euclidean distance and the cosine similarity.

[0024] wherein, the Euclidean distance is represented as d, the cosine similarity is represented as c.

[0025] Optionally, the determining the state normal or state abnormal of the real-time monitoring data at the current time window based on the target device running state prediction model, the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window to obtain a corresponding state judgment result comprises:

[0026] determining, by the target device running state prediction model, whether the fusion similarity result meets a dynamic similarity threshold parameter condition constructed based on a current first threshold parameter and a current second threshold parameter;

[0027] if yes, determining that the state of the real-time monitoring data at the current time window is normal, and updating the current first threshold parameter and the current second threshold parameter to obtain an updated current first threshold parameter and an updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition at a next time window;

[0028] if no, determining that the state of the real-time monitoring data at the current time window is abnormal, and marking the corresponding real-time monitoring data as abnormal monitoring data.

[0029] Optionally, the determining, by the target device running state prediction model, whether the fusion similarity result meets the dynamic similarity threshold parameter condition constructed based on the current first threshold parameter and the current second threshold parameter comprises:

[0030] determining, by the target device running state prediction model, whether the fusion similarity result meets the dynamic similarity threshold parameter condition constructed based on the current first threshold parameter and the current second threshold parameter.

[0031] wherein,​​ This indicates the fusion similarity result. This represents the current first threshold parameter. Indicates the target adjustment factor. This represents the current second threshold parameter. This indicates the preset standard similarity threshold.

[0032] Optionally, the step of updating the current first threshold parameter and the current second threshold parameter to obtain updated current first threshold parameters and updated current second threshold parameters for constructing dynamic similarity threshold parameter conditions under the next time window includes:

[0033] based on The current first threshold parameter is updated to obtain the updated current first threshold parameter used to construct the dynamic similarity threshold parameter conditions for the next time window; wherein, This represents the updated current first threshold parameter. This indicates the fusion similarity result. This indicates the fusion similarity result in the next time window;

[0034] based on The current second threshold parameter is updated to obtain the updated current second threshold parameter used to construct the dynamic similarity threshold parameter conditions for the next time window; wherein, This represents the updated current second threshold parameter. This represents the current second threshold parameter. This indicates the fusion similarity result. This indicates the fusion similarity result for the next time window.

[0035] Optionally, after determining that the status of the real-time monitoring data under the current time window is abnormal, the method further includes:

[0036] The current first threshold parameter and the current second threshold parameter under the current time window are respectively set as the updated current first threshold parameter and the updated current second threshold parameter to construct the dynamic similarity threshold parameter conditions under the next time window.

[0037] Optionally, the step of calculating the error contribution rate of each abnormal feature data in the abnormal monitoring data, and judging the current operating trend of the equipment component based on the error contribution rate, includes:

[0038] based on Calculate the error contribution rate of each abnormal feature data in the anomaly detection data;

[0039] in, Indicates the error contribution rate. represents the first sample of the first estimated feature data of the i-th sample, represents the first monitoring data of the i-th monitoring data abnormal feature data;

[0040] Determine the relative ratio based on the real-time monitoring value of the target abnormal feature data corresponding to the maximum error contribution rate and the state estimation value, and determine the running trend of the current equipment component based on the size relationship between the relative ratio and the preset relative ratio threshold.

[0041] Optionally, determining the running trend of the current equipment component based on the size relationship between the relative ratio and the preset relative ratio threshold comprises:

[0042] When the relative ratio is greater than zero and the relative ratio is greater than the preset relative ratio threshold, it is determined that the trend of the target abnormal feature data is abnormally rising;

[0043] When the relative ratio is greater than zero and the relative ratio is less than the preset relative ratio threshold, it is determined that the trend of the target abnormal feature data is gradually rising;

[0044] When the relative ratio is less than zero and the absolute value of the relative ratio is greater than the preset relative ratio threshold, it is determined that the trend of the target abnormal feature data is abnormally decreasing;

[0045] When the relative ratio is less than zero and the absolute value of the relative ratio is less than the preset relative ratio threshold, it is determined that the trend of the target abnormal feature data is gradually decreasing.

[0046] In a second aspect, the application discloses a device component running trend identification device, comprising:

[0047] A data acquisition module is configured to acquire real-time monitoring data of a current equipment component.

[0048] A result acquisition module is configured to input the real-time monitoring data into a target equipment running state prediction model, so that the target equipment running state prediction model predicts a state estimation sample of the current equipment component based on the real-time monitoring data and a memory matrix, and acquires a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time.

[0049] A judgment module is configured to perform state normal or state abnormal judgment on the real-time monitoring data at the current time window through the target equipment running state prediction model and according to the fusion similarity result, a dynamic similarity threshold parameter and a sliding time window, to obtain a corresponding state judgment result.

[0050] a trend prediction module, configured to, if the state judgment result is an abnormal state result, perform error contribution rate calculation on each abnormal feature data of the abnormal monitoring data, so as to judge the operation trend of the current equipment component according to the error contribution rate.

[0051] In a third aspect, the present application discloses an electronic device, comprising:

[0052] a memory, configured to save a computer program;

[0053] a processor, configured to execute the computer program to implement the steps of the operation trend identification method of the device component disclosed in the foregoing.

[0054] In a fourth aspect, the present application discloses a computer readable storage medium, configured to store a computer program; wherein the computer program is executed by a processor to implement the steps of the operation trend identification method of the device component disclosed in the foregoing.

[0055] It can be seen that the application discloses a device component operation trend identification method, which comprises the following steps: acquiring real-time monitoring data of a current device component; inputting the real-time monitoring data into a target device operation state prediction model, so that the target device operation state prediction model predicts a state estimation sample of the current device component based on the real-time monitoring data and a memory matrix, and acquires a fusion similarity result between the state estimation sample and the real-time monitoring data at a corresponding time; performing state normal or state abnormal judgment on the real-time monitoring data at a current time window by the target device operation state prediction model and according to the fusion similarity result, a dynamic similarity threshold parameter and a sliding time window, so as to obtain a corresponding state judgment result; if the state judgment result is an abnormal state result, performing error contribution rate calculation on each abnormal feature data of abnormal monitoring data, so as to judge the operation trend of the current device component according to the error contribution rate. It can be seen that the state estimation sample is generated based on the memory matrix and real-time data, the absolute difference and directional difference of data are captured at the same time through the fusion similarity, the accuracy of similarity calculation in a multi-dimensional data scene is improved, further, the dynamic threshold parameter is added, the experience dependence of the fixed threshold is avoided, and the false alarm / miss alarm caused by equipment aging and working condition fluctuation is reduced; the sliding time window mechanism makes the threshold adapt to the time sequence change of data, and the sensitivity to gradual trend is improved. Finally, the feature with the greatest abnormal influence is located through the error contribution rate, the abnormal source is traced, the problem that the traditional method is difficult to distinguish between gradual and sudden abnormality is solved, and clear trend judgment basis is provided for operation and maintenance personnel. Specifically, the sliding window mode is adopted, the threshold is dynamically updated based on the similarity change, the residual statistical characteristics can be reflected in time, the influence of random factors can be eliminated, the more accurate dynamic threshold of the key features of the component is calculated, and the error of the predicted component operation state is reduced; and the error contribution rate of each feature reflects the influence of each feature on the threshold from the comparison angle of relative state, which is more consistent with the actual application. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0057] Figure 1 A device component operation trend identification method flow chart is disclosed in the present application.

[0058] Figure 2 A device component operation trend identification device structure schematic diagram is disclosed in the present application.

[0059] Figure 3A structure diagram of an electronic device is disclosed. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0061] With the continuous advancement of rail transit system construction and the development of digital information, intelligent operation and maintenance platform is widely used to realize real-time monitoring and data collection of the state of rail transit vehicles / locos. At present, the online monitoring system monitors various trend data types, including vibration data, impact data, dB value, temperature, alarm, speed, etc. Meanwhile, the data volume is large, and there are problems such as noise in the data due to working conditions, transmission processing, etc. It is difficult for manual work to find the change of a certain monitoring trend of the vehicle / loco in the operation process in time. Secondly, the abnormal judgment method based on traditional threshold is only effective for obvious abnormal data, and it is difficult to find the gradually changing and hidden abnormal data in a large amount of data in time. If the algorithm automatic analysis and the change of trend data can be realized, the analysis personnel will pay attention to the change of the measuring point in time, and the risk of fault analysis leakage will be reduced.

[0062] At present, the data trend identification methods are mainly divided into the following three categories:

[0063] 1) Statistical measurement-based method: data trend anomaly is judged by data statistical distribution. For example, Z-score method, based on the standard deviation of data to measure the deviation between a data point and the average value. Statistical hypothesis testing method, including z-test, t-test, etc. The abnormality is identified by comparing the test statistic with the critical value. Moving average method, the observation values in the continuous window of time series data are arithmetically averaged, and the long-term trend and periodic change of data are highlighted by smoothing noise and short-term fluctuations.

[0064] 2) Machine learning-based methods: By learning the rules and patterns from the data to identify anomalies and other related tasks. For example, the K-Nearest Neighbors algorithm classifies by measuring the distance between samples, and finds the K nearest known category samples to determine the category of the sample to be classified. Isolation Forest algorithm, by constructing a binary tree method to isolate each abnormal sample, abnormal data due to the number is small and the difference with normal data is large, so when isolated, it needs fewer steps, suitable for continuous data anomaly detection; Principal Component Analysis (PCA) is usually used for anomaly detection in PCA algorithm, which involves projecting data onto PCA carriers and determining whether a point is abnormal based on a certain threshold. In addition, advanced representations of data can be learned to capture complex contextual feature associations in trend data, thereby better identifying trend anomalies in data. For example, the Variational Autoencoder-Long Short-Term Memory Network hybrid model extracts local features in short periods and estimates long-term correlations in sequences, enabling the identification of anomalies across multiple time scales. Graph attention mechanism-based models use graph attention mechanisms to view each feature as a node in a graph, and describe the correlation between features through the connection between nodes, thereby better describing sequence anomalies. Graph deviation learning is performed based on the derived graph attention scores to output graph deviation scores for anomaly judgment.

[0065] 3) State estimation-based methods: Identify data trend anomalies by estimating the residual changes between the estimated value and the observed value. The main method is the Multivariate State Estimation Technique (MSET). The traditional MSET method is based on healthy data during normal operation to obtain the estimated value of the actual operation, and evaluates the data trend changes by the difference between the estimated value and the actual value.

[0066] Correspondingly, the above existing methods have the following technical problems:

[0067] 1) Statistical measure-based methods are difficult to extract time features of data as they require the assumption that data follows a specific probability distribution, and often perform poorly in handling high-order multi-dimensional time series anomalies. In addition, when dealing with a large amount of redundant data, statistical methods face performance bottlenecks and cannot be easily expanded.

[0068] 2) Machine learning-based methods depend on data quality for their application results, and noise or bias in the data can directly lead to inaccurate model results. In addition, complex models usually require a large amount of data for training, which is costly and difficult to collect and label. At the same time, if the length of the input data is directly increased, the dimension of the input data increases, and the model training complexity increases greatly, which requires higher hardware conditions and takes a long time to train, not meeting the low-latency detection requirements of vehicle-mounted real-time calculation. When the data distribution is unbalanced, the model may overfit, and the performance is unstable.

[0069] 3) Traditional multi-element state estimation method needs a large amount of historical health data, which is easy to produce data redundancy, increase the complexity of calculation and computing resources. Secondly, the threshold for judging residual usually depends on experience and statistical method, and the threshold is too loose to miss detection, and too strict to misreport, and data interference and noise are easy to cause model result difference.

[0070] In the existing trend identification method, the simple method such as statistical model is easy to implement but poor in robustness, and the complex model such as deep learning has strong performance but depends on data and algorithm, which cannot meet the demand of actual situation. In addition, in the actual operation process, noise or data deviation is easy to interfere with the trend estimation, and with the process of component degradation, some gradual trend changes are difficult to identify, and the current trend change situation cannot be accurately quantified, and secondly, for multi-element data scene, trend anomaly may be caused by single variable or multiple variable cooperative change, and the contribution degree of each variable to trend change is difficult to explain.

[0071] Therefore, the present application provides a device component operation trend identification scheme, which can solve the problems of data redundancy, false alarm and missed alarm caused by fixed threshold, poor multi-dimensional data adaptability and difficulty in quantifying gradual trend.

[0072] Reference Figure 1 The embodiment of the present application discloses a device component operation trend identification method, comprising:

[0073] Step S11: acquiring real-time monitoring data of the current device component.

[0074] In the embodiment, the trend data of the current device component monitored by the running part vehicle-mounted fault diagnosis system is acquired to obtain the real-time monitoring data, wherein the device component can be specifically a industrial device component with multi-parameter monitoring demand, and can specifically include but not limited to locomotive running part device component, bogie, pantograph, bearing and other rail transit device components, and no specific limitation is made thereto; the real-time monitoring data includes rotation speed, temperature data, vibration data, running mileage data, impact SV data and impact dB data. The features sensitive to the device state change are selected, for example, the six-dimensional features of SV data corresponding speed, SV trend data, vibration acceleration effective value, vibration acceleration average value, vibration acceleration peak value and measuring point temperature data are selected as the basis for subsequent research.

[0075] Step S12: inputting the real-time monitoring data into the target device operation state prediction model, so that the target device operation state prediction model predicts the state estimation sample of the current device component based on the real-time monitoring data and the memory matrix, and acquires the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time.

[0076] In the embodiment, the historical normal operation data of the current device component is acquired, wherein the historical normal operation data contains a plurality of characteristic data associated with the change of the running state of the current device component; the numerical range of the historical normal operation data is divided by a preset step to obtain a plurality of numerical range interval segments; target characteristic data with a distance less than a preset distance threshold from the current interval endpoint is selected from each numerical range interval segment to construct a memory matrix. The historical normal operation data is multi-dimensional characteristic data including rotation speed data, temperature data, vibration data, running mileage data, impact SV data, and impact dB data. It can be understood that the memory matrix is constructed before the target device state prediction model is predicted. First, the historical normal operation data of the current device component is acquired, wherein the historical normal operation data is characteristic data when the device normally operates under different working conditions. It should be noted that the historical normal operation data is acquired by using the above-mentioned real-time monitoring data acquisition method, and only the corresponding time information is a past time before the current time, so the acquisition process is not described again. The value range and order of magnitude of the historical data of different characteristics are usually quite different. In order to avoid the influence of individual parameters with large absolute values on the overall estimation effect of the model, the obtained characteristic data needs to be processed, that is, the dimensionless processing of each characteristic data of the historical normal operation data is performed. Common methods include maximum and minimum value normalization and z-score standardization. For example, for the six-dimensional characteristics selected for the tread, the maximum and minimum values of each type of characteristic data are calculated for data normalization calculation, and the expression is as follows:

[0077] ;

[0078] wherein, represents the normalized characteristic data of the historical normal operation data, represents the original characteristic data of the historical normal operation data, represents the minimum value of the original characteristic data, represents the maximum value of the original characteristic data.

[0079] After the above normalization processing, each normalized feature data is obtained, and then normalized historical normal operation data is obtained, and then the normalized historical normal operation data is used to construct a memory matrix. Specifically, the normal operation trend data (normalized historical normal operation data) of the equipment under different working conditions is used, and the equidistant sampling method is used to select data points that meet the distance principle for each normalized feature data to construct a memory matrix, that is, for each normalized feature data, a series of slot points are uniformly divided in the value range of the feature data, and these slots are defined by a preset step length step, which is not limited. In each slot, find the data point closest to the center point of the slot and not exceeding the preset distance threshold δ, collect these data points that meet the condition, and gradually fill the memory matrix. The specific operation is as follows:

[0080] All feature data at the same time is constructed into a historical monitoring sample, the target distance between each feature data in the historical monitoring sample and the current interval endpoint in the current numerical range interval is calculated, and the target feature data with a target distance less than a preset distance threshold is retained; the current historical monitoring sample in which all target feature data is located is counted, and the repeated samples in the current historical monitoring sample are deleted to obtain a target historical monitoring sample; a memory matrix is constructed based on the target historical monitoring sample; wherein the number of rows of the memory matrix is the number of the target historical monitoring samples, and the number of columns is the feature data dimension of the historical normal operation data. It can be understood that according to the numerical range of each feature data, a step length threshold (such as 0.01) is set to divide the data range into a series of intervals to obtain the interval endpoints; all feature data points at the same time are constructed into a sample, the distance between each data point and each interval endpoint is calculated for each feature, and the sample corresponding to the data point with a distance less than a given threshold δ (δ is a very small positive number) is retained; the retained repeated samples are deleted, and the retained samples are used to construct a memory matrix D, and the expression form of the memory matrix is as follows:

[0081] ;

[0082] Wherein, The data value of the nth target feature data in the mth target historical monitoring sample, m represents the feature data dimension, and n represents the number of target historical monitoring samples.

[0083] In this way, the equidistant sampling method uniformly divides the value range of the feature by a preset step, and only selects data points meeting the distance threshold δ in each interval, thereby avoiding the retention of a large number of repeated or approximate samples in the traditional method, improving the data redundancy problem caused by a large number of repeated or approximate data points in the data set, significantly reducing the sample amount of the memory matrix, and reducing the complexity of subsequent calculation and the consumption of computing resources, thereby adapting to the low-latency demand of vehicle-mounted real-time calculation. Based on the distance principle, the data points with a distance ≤ δ from the interval endpoints are selected to ensure that the samples in each interval can cover the typical features of the interval, and repeated samples are deleted to avoid information redundancy. The finally constructed memory matrix contains not only the normal operation data features of the equipment under different working conditions, but also the key information retained through screening, thereby providing a reliable normal state benchmark for the generation of subsequent estimation samples. The intervals are independently divided for each feature, and the value range differences of different features are taken into account, for example, the order of magnitude difference between temperature and vibration acceleration, so that the memory matrix can adapt to the normal state distribution of multi-dimensional features, thereby laying a foundation for subsequent fusion similarity calculation and dynamic threshold judgment.

[0084] In the embodiment, before the target equipment operation state prediction model is used, a target equipment operation state prediction model based on a multivariate state estimation technology is built, the state estimation sample (estimated value) of the actual operation of the equipment is obtained through processing of the real-time monitoring data of the equipment and the memory matrix, and then the correlation degree between the estimated value and the real-time monitoring data (actual value), i.e., the similarity, is calculated to measure the difference between the data. The specific steps are as follows: real-time monitoring data of the equipment running for a period of time is selected, the state estimation sample of the actual operation of the equipment in this period of time is obtained through the MSET method, i.e., through the operation of the actual value and the state memory matrix D, and the calculation formula is as follows:

[0085]

[0086] Among them, is the real-time monitoring data, represents the state estimation sample, is the memory matrix, represents the Euclidean distance operator.

[0087] In the embodiment, the Euclidean distance and the cosine similarity between the state estimation sample and the real-time monitoring data at the corresponding time are calculated, and the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time is determined through the Euclidean distance and the cosine similarity.

[0088] Specifically, the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time is determined through wherein, represents the Euclidean distance,​ represents the cosine similarity. It can be understood that the state estimation sample is generated based on a memory matrix constructed based on historical normal data, representing normal operation data of the equipment under the current working condition; the real-time monitoring data is the current real-time monitoring data of the equipment, representing the real operation data of the equipment. By calculating the fusion similarity result between the two, the essence is to measure the consistency of the actual state and the normal state, the higher the similarity, the closer the actual operation state to the normal state; the lower the similarity, the more obvious the deviation from the normal state. Moreover, the above fusion similarity calculation method considers the difference in absolute position and direction of the vector space, combines the Euclidean distance and the cosine similarity, and proposes a similarity calculation method based on fusion distance to calculate the similarity of the estimation sample and the actual sample, which is suitable for the multi-dimensional data scene of trend data. In this way, the Euclidean distance and the cosine similarity are combined to evaluate the similarity between the data, which takes the advantages of the two distance measurement methods and solves the inapplicability of a single distance calculation method, and adapts to the multi-dimensional data scene of trend data.

[0089] Step S13: performing state normal or state abnormal judgment on the real-time monitoring data under the current time window according to the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window through the target equipment operation state prediction model, to obtain a corresponding state judgment result.

[0090] In this embodiment, it is judged through the target equipment operation state prediction model whether the fusion similarity result meets the dynamic similarity threshold parameter condition constructed based on the current first threshold parameter and the current second threshold parameter; specifically, it is judged through the target equipment operation state prediction model whether the fusion similarity result meets ; wherein, represents the fusion similarity result, represents the current first threshold parameter, represents the target adjustment coefficient, represents the current second threshold parameter, represents the preset standard similarity threshold. It can be understood that the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window are used to judge the state normal or state abnormal of the real-time monitoring data under the current time window, specifically, if the fusion similarity result is greater than or equal to the constructed according to the current first threshold parameter and the current second threshold parameter, and the fusion similarity result is greater than the preset standard similarity threshold, it is determined that the real-time monitoring data is normal, otherwise, it is determined that the real-time monitoring data is abnormal. The target adjustment coefficient can be set to 4, and the preset standard similarity threshold can be set to 0.8.

[0091] In the embodiment, if yes, it is determined that the state of the real-time monitoring data under the current time window is normal, and the current first threshold parameter and the current second threshold parameter are updated to obtain updated current first threshold parameter and updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition under a next time window; specifically, based on The current first threshold parameter is updated to obtain an updated current first threshold parameter used for constructing a dynamic similarity threshold parameter condition under a next time window; wherein, represents the updated current first threshold parameter, represents the fusion similarity result, represents a fusion similarity result under a next time window; based on The current second threshold parameter is updated to obtain an updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition under a next time window; wherein, represents the updated current second threshold parameter, represents the current second threshold parameter, represents the fusion similarity result, represents a fusion similarity result under a next time window. It can be understood that through the above process, the threshold parameters under different time windows are updated in the case of determining that the running state of the real-time monitoring data is normal, and the update condition is to update the threshold parameters under the current time window according to the threshold parameters of the previous time window. In this way, the threshold values under each time window are dynamically updated. Since the equipment is in long-term operation, the normal state is not absolutely fixed: due to natural aging of components, working condition fluctuations and the like, the statistical characteristics (such as mean value and fluctuation range) of normal data slowly drift; for example, the normal range of vibration data of a new device may be different from the normal range after years of operation. If a fixed threshold value is used, it will cause misjudgment (mistakenly judging the normal state after aging as abnormal, or missing the real abnormality) due to the out-of-date normal benchmark. The dynamic threshold value of the embodiment updates the mean value and the standard deviation in real time, so that the normal benchmark is adaptively adjusted according to the current health status of the equipment, and ensures that the threshold value always matches the current normal state.

[0092] In the embodiment, if no, it is determined that the state of the real-time monitoring data under the current time window is abnormal, and the corresponding real-time monitoring data is marked as abnormal monitoring data.

[0093] Step S14: If the state judgment result is an abnormal state result, error contribution rate calculation is performed on each abnormal feature data of the abnormal monitoring data to determine the running trend of the current equipment component according to the error contribution rate.

[0094] In this embodiment, after determining that the real-time monitoring data under the current time window is abnormal, the method further includes: setting the current first threshold parameter and the current second threshold parameter under the current time window to the updated current first threshold parameter and the updated current second threshold parameter for constructing the dynamic similarity threshold parameter conditions under the next time window, respectively. It can be understood that if the above judgment conditions are not met, the real-time monitoring data under the current time window is determined to be abnormal, and there is no need to update the threshold parameters. , .

[0095] In this embodiment, after determining the abnormal monitoring data, in order to determine the current operating trend of the equipment components, the error contribution rate of each abnormal feature data of the abnormal monitoring data is further calculated. Specifically, based on... Calculate the error contribution rate of each abnormal feature data in the anomaly detection data; where, Indicates the error contribution rate. In the state estimation sample, the first... The first sample One estimated feature data, Indicates the first in the abnormal monitoring data The first monitoring data The system identifies several abnormal feature data points. Based on the real-time monitoring value and state estimate of the target abnormal feature data corresponding to the maximum error contribution rate, a relative ratio is determined. The operating trend of the current equipment component is then judged based on the relationship between this relative ratio and a preset relative ratio threshold. It is understood that the error contribution rate is calculated for each abnormal feature data point at each time an anomaly is determined. The specific calculation formula is shown above. In this way, by calculating the error contribution rate of each abnormal feature data point, the target abnormal feature data with the largest error contribution rate is determined. Then, the actual value and estimated value of the target abnormal feature data are retrieved, and the relative ratio between the actual value and the estimated value is calculated. The calculation formula is as follows:

[0096] ;

[0097] in, This represents the actual value of the target anomaly feature data corresponding to the maximum error contribution rate. This represents the estimated value of the feature corresponding to the maximum error contribution rate.

[0098] Specifically, when the relative ratio is greater than zero and the relative ratio is greater than the preset relative ratio threshold, it is determined that the target abnormal feature data has an abnormal upward trend; when the relative ratio is greater than zero and the relative ratio is less than the preset relative ratio threshold, it is determined that the target abnormal feature data has a gradual upward trend; when the relative ratio is less than zero and the absolute value of the relative ratio is greater than the preset relative ratio threshold, it is determined that the target abnormal feature data has an abnormal downward trend; and when the relative ratio is less than zero and the absolute value of the relative ratio is less than the preset relative ratio threshold, it is determined that the target abnormal feature data has a gradual downward trend. It can be understood that the preset relative ratio threshold is set, when r>0, if |r| is greater than the preset relative ratio threshold, it is considered that the corresponding feature data under the corresponding sample has an abnormal upward trend; if |r| is less than the preset relative ratio threshold, it is considered that the corresponding feature data under the corresponding sample has a gradual upward trend. When r≤0, if |r| is greater than the preset relative ratio threshold, it is considered that the corresponding feature data under the corresponding sample has an abnormal downward trend; if |r| is less than the preset relative ratio threshold, it is considered that the corresponding feature data under the corresponding sample has a gradual downward trend.

[0099] In this way, the data trend judgment method based on interval statistical dynamic threshold can dynamically update the threshold based on similarity according to the equipment operation condition and the self health state by using the sliding window method, monitor abnormal data, and judge the upward and downward changes of the data trend through the feature error contribution rate in all abnormal states. This method is not only suitable for the locomotive field, but also universal in the operation trend identification field of other equipment components, and has a wide application range.

[0100] It can be seen that the application discloses a device component operation trend identification method, which comprises the following steps: acquiring real-time monitoring data of a current device component; inputting the real-time monitoring data into a target device operation state prediction model, so that the target device operation state prediction model predicts a state estimation sample of the current device component based on the real-time monitoring data and a memory matrix, and acquires a fusion similarity result between the state estimation sample and the real-time monitoring data at a corresponding time; performing state normal or state abnormal judgment on the real-time monitoring data at a current time window by the target device operation state prediction model and according to the fusion similarity result, a dynamic similarity threshold parameter and a sliding time window, so as to obtain a corresponding state judgment result; if the state judgment result is an abnormal state result, performing error contribution rate calculation on each abnormal feature data of abnormal monitoring data, so as to judge the operation trend of the current device component according to the error contribution rate. It can be seen that the state estimation sample is generated based on the memory matrix and the real-time data, the absolute difference and the direction difference of the data are captured at the same time through the fusion similarity, the accuracy of the similarity calculation in the multi-dimensional data scene is improved, further, the dynamic threshold parameter is added, the experience dependence of the fixed threshold is avoided, and the false alarm / miss alarm caused by the equipment aging and the working condition fluctuation is reduced; the sliding time window mechanism makes the threshold adapt to the time sequence change of the data, and the sensitivity to the gradual change trend is improved. Finally, the feature with the greatest abnormal influence is located through the error contribution rate, the abnormal source is traced, the problem that the traditional method is difficult to distinguish the gradual change and the sudden abnormality is solved, and clear trend judgment basis is provided for the operation and maintenance personnel. Specifically, the sliding window mode is adopted, the threshold is dynamically updated based on the similarity change, the residual statistical characteristics can be reflected in time, the influence of random factors can be eliminated, the more accurate dynamic threshold of the key features of the component is calculated, and the error of the predicted component operation state is reduced; and the error contribution rate of each feature reflects the influence of each feature on the threshold from the comparison angle of the relative state, which is more consistent with the actual application.

[0101] Referring to Figure 2 The application also discloses a device component operation trend identification device, which comprises the following components:

[0102] A data acquisition module 11 is configured to acquire real-time monitoring data of a current device component.

[0103] A result acquisition module 12 is configured to input the real-time monitoring data into a target device operation state prediction model, so that the target device operation state prediction model predicts a state estimation sample of the current device component based on the real-time monitoring data and a memory matrix, and acquires a fusion similarity result between the state estimation sample and the real-time monitoring data at a corresponding time.

[0104] The judging module 13 is configured to perform state normal or state abnormal judgment on the real-time monitoring data in the current time window by the target equipment running state prediction model and according to the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window, to obtain a corresponding state judgment result.

[0105] The trend prediction module 14 is configured to, if the state judgment result is an abnormal state result, perform error contribution rate calculation on each abnormal feature data of the abnormal monitoring data, to judge the running trend of the current equipment component according to the error contribution rate.

[0106] Therefore, the real-time monitoring data of the current equipment component is obtained, the real-time monitoring data is input into a target equipment running state prediction model, the target equipment running state prediction model predicts a state estimation sample of the current equipment component based on the real-time monitoring data and a memory matrix, and a fusion similarity result between the state estimation sample and the real-time monitoring data at a corresponding time is obtained. The real-time monitoring data in the current time window is judged by the target equipment running state prediction model and according to the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window, to obtain a corresponding state judgment result. If the state judgment result is an abnormal state result, error contribution rate calculation is performed on each abnormal feature data of the abnormal monitoring data, to judge the running trend of the current equipment component according to the error contribution rate. Therefore, the state estimation sample is generated based on the memory matrix and the real-time data, the absolute difference and the directional difference of the data are captured simultaneously through the fusion similarity, the accuracy of the similarity calculation in the multi-dimensional data scene is improved, further, the dynamic threshold parameter is added, the experience dependence of the fixed threshold is avoided, and the false alarm / miss alarm caused by equipment aging and working condition fluctuation is reduced. The sliding time window mechanism makes the threshold adapt to the time sequence change of the data, and the sensitivity to the gradual change trend is improved. Finally, the feature with the greatest impact on the anomaly is located through the error contribution rate, the anomaly is traced, the problem that the traditional method is difficult to distinguish the gradual change and the sudden abnormality is solved, and clear trend judgment basis is provided for the operation and maintenance personnel. Specifically, the sliding window method is adopted, the threshold is dynamically updated based on the similarity change, the residual statistical characteristics can be reflected in time, the influence of random factors can be eliminated, the more accurate dynamic threshold of the key feature of the component is calculated, and the error of the predicted component running state is reduced. The error contribution rate of each feature reflects the influence of each feature on the threshold from the relative state comparison angle, and is more consistent with the actual application.

[0107] Further, the embodiment of the application further discloses an electronic device, Figure 3 The electronic device 20 shown in the figure is not considered as any limitation on the use range of the application.

[0108] Figure 3 A structural schematic diagram of an electronic device 20 is provided in the embodiments of the present application. The electronic device 20 can specifically include at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25 and a communication bus 26. The memory 22 is configured to store a computer program, and the processor 21 is configured to load and execute the computer program to implement the related steps in the operation trend identification method of the device components disclosed in any of the preceding embodiments. In addition, the electronic device 20 in the embodiments can be specifically an electronic computer.

[0109] In the embodiments, the power supply 23 is configured to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 is capable of creating a data transmission channel between the electronic device 20 and external devices, and the communication protocol followed by the communication interface 24 can be any communication protocol applicable to the technical solutions of the present application, which is not specifically limited herein; the input / output interface 25 is configured to obtain external input data or output data to the outside, and the specific interface type can be selected according to the specific application needs, which is not specifically limited herein.

[0110] The processor 21 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array) and a PLA (Programmable Logic Array). The processor 21 can also include a main processor and a coprocessor. The main processor is a processor for processing data in a wake-up state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 21 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 can further include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0111] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage mode can be temporary storage or permanent storage.

[0112] The operating system 221 is used to manage and control each hardware device on the electronic device 20 and the computer program 222, so as to realize the operation and processing of the processor 21 on the mass data 223 in the memory 22, which can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program capable of completing the operation trend identification method of the device components executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 can further include a computer program capable of completing other specific work. The data 223 can include data transmitted by an external device and received by the electronic device, and data collected by the self input / output interface 25, etc.

[0113] Further, the application also discloses a computer readable storage medium for storing a computer program; wherein the computer program is executed by a processor to realize the operation trend identification method of the device components disclosed in the foregoing. The specific steps of the method can refer to the corresponding content disclosed in the foregoing embodiments, which will not be described here.

[0114] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can refer to the method part.

[0115] Those skilled in the art will further appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or any combination thereof. To clearly illustrate the interchangeability of hardware and software, various components have been described above generally in terms of their functionality, without limitation. The handwiring and software implementations of the examples described herein could be accomplished using any number of microprocessors, microcontrollers, programmable consumption logic devices, application-specific integrated circuits, or general-purpose computers with interconnecting circuits that either run software programs or use opencircuit or other hardware components that are designed to perform the functions described herein. The embodiments described herein can be implemented along with software modules, and the software modules can be stored on any of a variety of non-transitory machine-readable media. A non-transitory machine-readable medium includes any medium that participates in providing instructions to a processor for execution. Such a medium can take many forms, including but not limited to, non-volatile media, volatile media, and transmission media. Non-volatile media includes, for example, optical or magnetic disks and other persistent memory. Volatile media includes dynamic memories, and physical registers. Transmission media includes coaxial cables, copper wires and fiber optic cables, including wires that comprise bus conductors. Transmission media also can also take the form of acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media include, for example, a floppy disk, a flexible disk, hard disk, solid-state drive, magnetic tape, or any other magnetic data storage medium, a Compact Disc - Read Only Memory (CD-ROM), any other optical medium, punch cards, paper tape, any other physical medium with patterns of holes, a RAM, a programmable ROM (PROM), an erasable PROM (EPROM), a FLASH-EPROM, any other memory chip or cartridge, a carrier wave, a

[0116] Finally, it should also be noted that, in the present text, relational terms such as first and second and the like can only be used to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between such entities or operations. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0117] The above has carried on the detailed introduction to the scheme provided by the present application, the principle and implementation mode of the present application are described by applying the specific examples in the present text, the above example explanation is only for helping the understanding of the method and core idea of the present application; simultaneously, for the general technical personnel in the field, according to the idea of the present application, there will be the change in the specific implementation mode and application range, the above-mentioned content should not be understood as the limitation of the present application.

Claims

1. A method of identifying a running trend of a device component, characterized by, The method comprises: obtaining real-time monitoring data of a current device component; inputting the real-time monitoring data into a target device running state prediction model, so that the target device running state prediction model predicts a state estimation sample of the current device component based on the real-time monitoring data and a memory matrix, and obtains a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time; performing state normal or state abnormal judgment on the real-time monitoring data at the current time window through the target device running state prediction model and according to the fusion similarity result, a dynamic similarity threshold parameter and a sliding time window, to obtain a corresponding state judgment result; if the state judgment result is an abnormal state result, performing error contribution rate calculation on each abnormal feature data of the abnormal monitoring data, to judge the running trend of the current device component according to the error contribution rate.

2. The method of claim 1, wherein The method further comprises: obtaining historical normal running data of a current device component, wherein the historical normal running data contains a plurality of feature data associated with the running state change of the current device component; dividing the numerical range of the historical normal running data by a preset step length to obtain a plurality of numerical range interval segments; selecting target feature data with a distance less than a preset distance threshold from the current interval endpoint from each numerical range interval segment to construct a memory matrix.

3. The method of claim 2, wherein The historical normal running data is multi-dimensional feature data including rotation speed data, temperature data, vibration data, running mileage data, impact SV data and impact dB data.

4. The method of claim 2, wherein The method of selecting target feature data with a distance less than a preset distance threshold from the current interval endpoint from each numerical range interval segment to construct a memory matrix comprises: constructing all feature data at the same time as a historical monitoring sample, calculating the target distance between each feature data in the historical monitoring sample and the current interval endpoint in the numerical range interval where the feature data is currently located, and retaining target feature data with a target distance less than a preset distance threshold; counting the current historical monitoring sample where all target feature data is located, and deleting duplicate samples in the current historical monitoring sample to obtain target historical monitoring samples; constructing a memory matrix based on the target historical monitoring samples; wherein the number of rows of the memory matrix is the number of target historical monitoring samples, and the number of columns is the feature data dimension of the historical normal running data.

5. The method of claim 1, wherein The method of obtaining a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time comprises: calculating the Euclidean distance and cosine similarity between the state estimation sample and the real-time monitoring data at the corresponding time; determining the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time through the Euclidean distance and the cosine similarity.

6. The method of claim 5, wherein The method of determining the fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time through the Euclidean distance and the cosine similarity comprises: By determining a fusion similarity result between the state estimation sample and the real-time monitoring data at the corresponding time point; wherein, denotes the Euclidean distance, denotes the cosine similarity.

7. The method of claim 1, wherein The state normal or state abnormal judgment on the real-time monitoring data under the current time window is performed through the target equipment running state prediction model and according to the fusion similarity result, the dynamic similarity threshold parameter and the sliding time window to obtain a corresponding state judgment result, and the state judgment result includes: The target equipment running state prediction model is used to judge whether the fusion similarity result meets a dynamic similarity threshold parameter condition constructed based on a current first threshold parameter and a current second threshold parameter. If yes, it is determined that the state of the real-time monitoring data under the current time window is normal, and the current first threshold parameter and the current second threshold parameter are updated to obtain updated current first threshold parameter and updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition under a next time window. If no, it is determined that the state of the real-time monitoring data under the current time window is abnormal, and the corresponding real-time monitoring data is marked as abnormal monitoring data.

8. The method of claim 7, wherein The target equipment running state prediction model is used to judge whether the fusion similarity result meets a dynamic similarity threshold parameter condition constructed based on a current first threshold parameter and a current second threshold parameter. determining, by the target device, whether the fusion similarity result meets a preset condition according to the target device running state prediction model ; wherein, represents a fusion similarity result, represents a current first threshold parameter, represents a target adjustment coefficient, represents a current second threshold parameter, represents a preset standard similarity threshold.

9. The method according to claim 7, wherein The current first threshold parameter and the current second threshold parameter are updated to obtain updated current first threshold parameter and updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition under a next time window. based on updating the current first threshold parameter to obtain an updated current first threshold parameter used for constructing a dynamic similarity threshold parameter condition in a next time window; wherein, denotes the updated current first threshold parameter, denotes the fusion similarity result, denotes the fusion similarity result in the next time window; Based on updating the current second threshold parameter to obtain an updated current second threshold parameter used for constructing a dynamic similarity threshold parameter condition in a next time window; wherein, denotes the updated current second threshold parameter, denotes the current second threshold parameter, denotes the fusion similarity result, denotes the fusion similarity result in the next time window.

10. The method of claim 7, wherein the operation trend of the device component is identified based on the operation data of the device component. After it is determined that the state of the real-time monitoring data under the current time window is abnormal, the method further includes: The current first threshold parameter and the current second threshold parameter under the current time window are set as the updated current first threshold parameter and the updated current second threshold parameter used for constructing the dynamic similarity threshold parameter condition under the next time window.

11. The method of claim 1 to 10, wherein The error contribution rate calculation is performed on each abnormal feature data of the abnormal monitoring data to determine the running trend of the current equipment component according to the error contribution rate, and the error contribution rate calculation includes: based on calculating an error contribution rate of each abnormal feature data of the abnormal detection data; wherein, represents an error contribution rate, represents the first estimated feature data of the th sample in the state estimation sample, represents the first abnormal feature data of the th monitoring data in the abnormal monitoring data; A relative ratio is determined based on a real-time monitoring value and a state estimation value of target abnormal feature data corresponding to the maximum error contribution rate, and the running trend of the current equipment component is determined based on a size relationship between the relative ratio and a preset relative ratio threshold.

12. The method of claim 11, wherein The running trend of the current equipment component is determined based on a size relationship between the relative ratio and a preset relative ratio threshold, and the determination includes: When the relative ratio is greater than zero and the relative ratio is greater than the preset relative ratio threshold, it is determined that the target abnormal feature data has an abnormal upward trend; When the relative ratio is greater than zero and the relative ratio is less than the preset relative ratio threshold, it is determined that the target abnormal feature data has a gradually increasing upward trend; When the relative ratio is less than zero and an absolute value of the relative ratio is greater than the preset relative ratio threshold, it is determined that the target abnormal feature data has an abnormal downward trend; When the relative ratio is less than zero and the absolute value of the relative ratio is less than the preset relative ratio threshold, it is determined that the target abnormal feature data has a gradually decreasing downward trend.

13. A device for identifying the operating trend of a device component, characterized in that, The method includes: The data acquisition module is configured to acquire real-time monitoring data of a current device component. The result acquisition module is configured to input the real-time monitoring data into a target device operation state prediction model, so that the target device operation state prediction model predicts a state estimation sample of the current device component based on the real-time monitoring data and a memory matrix, and acquires a fusion similarity result between the state estimation sample and the real-time monitoring data at a corresponding time point. The judgment module is configured to perform state normal or state abnormal judgment on the real-time monitoring data at a current time window by the target device operation state prediction model and according to the fusion similarity result, a dynamic similarity threshold parameter and a sliding time window, to obtain a corresponding state judgment result. The trend prediction module is configured to, if the state judgment result is an abnormal state result, perform error contribution rate calculation on each abnormal feature data of the abnormal monitoring data, and judge an operation trend of the current device component according to the error contribution rate.

14. An electronic device, comprising: The memory is configured to save a computer program. The processor is configured to execute the computer program to implement the steps of the device component operation trend identification method according to any one of claims 1 to 12. The computer program is configured to be executed by the processor to implement the steps of the device component operation trend identification method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, ​

Images