A feature extraction and recognition method based on field data

Abstract

The application belongs to the technical field of feature engineering, and specifically discloses a feature extraction and identification method based on field data. According to the application, the running parameters of a target device collected in the field are analyzed in multiple dimensions; target key parameters are selected from various parameter categories; the target correlation between the target key parameters and other running parameters is calculated according to a parameter correlation coefficient calculation strategy; key feature parameters are selected from the target key parameters according to the target correlation, and a fault mode of the target device is identified according to the key feature parameters based on an attention mechanism and a generative adversarial network identification model. In this way, in the case of limited isolated environment conditions, the running parameters collected in the field are used to the greatest extent, an identification model based on an attention mechanism and a generative adversarial network suitable for an isolated environment is introduced, and the purpose of learning a generative task without using as much labeled data as possible is achieved, thereby effectively improving the accuracy of identifying fault modes.

Description

Technical Field

[0001] This application belongs to the field of feature engineering technology, and more specifically, relates to a feature extraction and recognition method based on field data. Background Technology

[0002] Equipment operation and maintenance and health management are of paramount importance in various industries and face various practical challenges, especially in application scenarios far from land, such as ocean-going ships and space shuttles, which face objective problems such as environmental isolation and harsh conditions. In order to ensure the normal operation and use of equipment, more comprehensive equipment health management methods are needed. Among them, the extraction and selection of fault characteristic parameters is an important link and foundation for achieving effective health management methods, which determines the overall efficiency and convenience.

[0003] Fault feature extraction and fault mode recognition methods can be summarized into three categories. The first category relies on analytical models, requiring the establishment of an accurate and suitable mathematical model for the specific equipment. Usable feature parameters are obtained by observing the model itself or its internal parameters, thereby determining the equipment's operating status. However, issues such as the accuracy of the mathematical model make it difficult to achieve the desired results in practical engineering applications. The second category relies on signal processing. Addressing the difficulty in constructing accurate mathematical models, it analyzes measurable signals such as the system's input and output signals using signal processing techniques such as correlation functions, spectrum analysis, autoregressive moving averages, and wavelet transforms to extract more feature values, such as variance, amplitude, and frequency, thereby enabling detection. The third category relies on artificial intelligence methods, specifically those used when mathematical models cannot be established or input / output signals cannot be effectively analyzed. These include expert systems, neural networks, and fuzzy mathematics. However, due to insufficient understanding of the "black box" principle and mechanisms, the interpretability of feature parameters requires further research, making fault location-level diagnosis difficult. For complex equipment, its complexity, dynamism, limitations in testing methods, and imprecise information representation lead to randomness and uncertainty in the relationship between fault causes and symptoms. Furthermore, considering isolated environments, the directly obtainable fault samples and observational data are limited, and prior information exhibits multi-source heterogeneity. Therefore, the accuracy of these methods in identifying fault modes is relatively low. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application aims to provide a feature extraction and identification method based on field data. This method addresses the problems of insufficient black-box principle and mechanism analysis, randomness and uncertainty in the relationship between fault causes and symptoms, limited fault samples and observation data that can be directly obtained under isolated environmental conditions, and low accuracy in identifying fault patterns due to the presence of multiple sources and heterogeneous prior information.

[0005] To achieve the above objectives, in a first aspect, this application provides a feature extraction and recognition method based on field data, comprising: The operating parameters of the target equipment are collected on-site, and the operating parameters are analyzed in multiple dimensions based on the equipment principle and domain knowledge base of the target equipment. Based on the results of multi-dimensional analysis, the data of each of the operating parameters are classified, and target key parameters are selected from each parameter category according to the domain knowledge base and operating mechanism. Based on the target key parameters, the target correlation degree between the target key parameters and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy; Based on the target relevance, key feature parameters are selected from each of the target key parameters, and based on the recognition model of attention mechanism and generative adversarial network, the target device is identified according to the key feature parameters to identify the fault mode.

[0006] In one embodiment, the step of collecting the operating parameters of the target device on-site and performing multi-dimensional analysis on each operating parameter based on the device principle and domain knowledge base of the target device includes: Collect operating parameters of the target equipment on-site; A lightweight digital twin model adapted to the target device is constructed, and the operating conditions of the target device are simulated based on the lightweight digital twin model. An initial parameter set is constructed based on the simulated operating parameters and the operating parameters of the target equipment acquired on-site. Based on the device principle and domain knowledge base of the target device, a multi-dimensional analysis is performed on each operating parameter in the initial parameter set.

[0007] In one embodiment, the step of calculating the target correlation between the target key parameter and other operating parameters based on the target key parameter and according to the parameter correlation coefficient calculation strategy includes: Obtain the values ​​of the target key parameters before and after the change, and the values ​​of other operating parameters before and after the change, respectively; Based on the target key parameter, the correlation coefficient between the target key parameter and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy, the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change. Count the number of the target key parameters; When the quantity equals a preset threshold, the target correlation between the target key parameter and other operating parameters is quantitatively calculated based on the correlation coefficient.

[0008] In one embodiment, after the step of counting the number of the target key parameters, the method further includes: When the number exceeds a preset threshold, the importance of each of the target key parameters is determined. The weights of the relevance coefficients of each of the target key parameters are set according to their importance. A comprehensive correlation coefficient is calculated based on the weights, the correlation coefficients of the target key parameters, and the quantity. The target correlation between the target key parameters and other operating parameters is then quantitatively calculated based on the comprehensive correlation coefficient.

[0009] In one embodiment, the recognition model based on attention mechanism and generative adversarial network includes a target generation network, a target attention mechanism, and a target recognition network; The step of selecting key feature parameters from each of the target key parameters based on the target relevance, and performing fault mode identification of the target device based on the key feature parameters using an attention mechanism and a generative adversarial network recognition model, includes: The key parameters of each target are sorted according to the target relevance. The parameter ranking results are analyzed based on a multi-dimensional data analysis strategy, and key feature parameters are selected from each of the target key parameters based on the ranking analysis results. When an abnormal operation of the target device is detected, fault mode recognition features are generated based on the target generation network according to the key feature parameters, and the fault mode recognition samples are enhanced based on the target attention mechanism. The enhanced fault mode recognition features and the key feature parameters are input into the target recognition network, and the fault mode recognition results output by the target recognition network are obtained.

[0010] In one embodiment, before the step of generating fault mode recognition features based on the key feature parameters according to the target generation network when an abnormal operation of the target device is detected, the method further includes: Acquire historical and empirical data on failure modes, and generate network training samples based on the historical and empirical data; The prior distribution of random noise and the real sample data are determined based on the network training samples. The prior distribution of the random noise is input into the generator, and the fault sample data output by the generator is obtained. The real sample data and the fault sample data are input into the discriminator, and the discrimination probability value output by the discriminator is obtained; The target generation network is obtained by training the network based on the discrimination probability values.

[0011] Secondly, this application provides a feature extraction and recognition device based on on-site data, comprising: The analysis module is used to collect the operating parameters of the target equipment on site and perform multi-dimensional analysis on each operating parameter based on the equipment principle of the target equipment and the domain knowledge base. The classification module is used to classify the data of each of the operating parameters according to the multi-dimensional analysis results, and select the target key parameters from each parameter category according to the domain knowledge base and operating mechanism. The calculation module is used to calculate the target correlation degree between the target key parameters and other operating parameters based on the target key parameters and according to the parameter correlation coefficient calculation strategy. The identification module is used to select key feature parameters from each of the target key parameters according to the target relevance, and to identify the fault mode of the target device based on the key feature parameters according to the identification model of attention mechanism and generative adversarial network.

[0012] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0013] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0014] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0015] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0016] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: (1) This application addresses scenarios with limited conditions in isolated environments by maximizing the use of on-site collected operating parameters and constructing a lightweight digital twin model adapted to the target device. Based on the lightweight digital twin model, the operating conditions of the target device are simulated, and a complete initial parameter set is constructed. After selecting target key parameters from each parameter category according to the domain knowledge base and operating mechanism, the target key parameters are used as a benchmark, and the target correlation between the target key parameters and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy. Based on the target correlation, key feature parameters are selected from each target key parameter. Since the target device may have defects such as few real fault samples, poor unknown fault identification effect, and easy overfitting under small sample size of the labeled dataset during operation in isolated environment scenarios, this embodiment introduces a recognition model based on attention mechanism and generative adversarial network to identify fault modes of the target device according to key feature parameters, thereby effectively improving the accuracy of fault mode identification.

[0017] (2) When calculating the target correlation between the target key parameter and other operating parameters, this application also considers the number of target key parameters. When the number is equal to a preset threshold, the target correlation between the target key parameter and other operating parameters is quantitatively calculated based on the correlation coefficient. When the number is greater than the preset threshold, the target correlation between the target key parameter and other operating parameters is quantitatively calculated based on the comprehensive correlation coefficient. Unlike the commonly used sensitivity coefficient calculation, the target correlation between parameters can be calculated by one perturbation (corresponding to the actual perturbation process) to complete the correlation of all other operating parameters with respect to the target key parameter, instead of requiring a separate perturbation for each parameter in the parameter list as in the sensitivity coefficient, thereby greatly improving the calculation efficiency.

[0018] In summary, this application collects the operating parameters of the target device on-site and performs multi-dimensional analysis on each operating parameter based on the device's operating principle and domain knowledge base. Based on the multi-dimensional analysis results, the operating parameters are categorized, and target key parameters are selected from each parameter category according to the domain knowledge base and operating mechanism. Using the target key parameters as a benchmark, the target correlation between the target key parameters and other operating parameters is calculated according to a parameter correlation coefficient calculation strategy. Key feature parameters are selected from the target key parameters based on the target correlation, and fault mode identification of the target device is performed based on the key feature parameters using an attention mechanism and a generative adversarial network (GAN) recognition model. Through this method, for scenarios with limited isolated environmental conditions, the on-site collected operating parameters are utilized to the maximum extent. After selecting key feature parameters based on target correlation, a suitable recognition model based on an attention mechanism and GAN for isolated environments is introduced. This achieves the goal of generative task learning with minimal use of labeled data, addressing the practical shortcomings of few real fault samples and difficulties in data labeling, thereby effectively improving the accuracy of fault mode identification. Attached Figure Description

[0019] Figure 1 This is one of the flowcharts illustrating the feature extraction and recognition method based on field data provided in the embodiments of this application; Figure 2 This is a schematic diagram of the overall process of fault mode recognition provided in the embodiments of this application; Figure 3 This is the second flowchart illustrating the feature extraction and recognition method based on field data provided in the embodiments of this application; Figure 4 This is a schematic diagram of the module structure of the feature extraction and recognition device based on field data provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] In this article, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0022] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0023] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0024] Based on this, embodiments of this application provide a feature extraction and recognition method based on on-site data, referring to... Figure 1 , Figure 1 This is one of the flowcharts illustrating the feature extraction and recognition method based on field data provided in this application embodiment. In this embodiment, the feature extraction and recognition method based on field data includes steps S10 to S40: Step S10: Collect the operating parameters of the target equipment on site, and perform multi-dimensional analysis on each operating parameter based on the equipment principle of the target equipment and the domain knowledge base.

[0025] It should be noted that in routine management, on-site data collection can be used to obtain the operating parameters of the target equipment, for example, (Parameter 1) (Parameter 2) (Parameter 3), ... For all operating parameters of the target device, such as (parameter n), it is necessary to conduct multi-dimensional analysis of each operating parameter based on the device principle and domain knowledge base in order to facilitate subsequent data classification.

[0026] Further, step S10 includes: collecting operating parameters of the target device on-site; constructing a lightweight digital twin model adapted to the target device, and simulating the operating conditions of the target device based on the lightweight digital twin model; constructing an initial parameter set based on the simulated operating parameters and the operating parameters of the target device collected on-site; and performing multi-dimensional analysis on each operating parameter in the initial parameter set based on the device principle and domain knowledge base of the target device.

[0027] It should be understood that, in order to construct a complete initial parameter set, in addition to collecting the operating parameters of the target equipment on-site, it is also necessary to construct a lightweight digital twin model adapted to the target equipment, and simulate the operating conditions of the target equipment based on the lightweight digital twin model to obtain all the operating parameters of the target equipment. Then, based on the simulated operating parameters and the operating parameters of the target equipment collected on-site, an initial parameter set is constructed. This initial parameter set can be represented as follows: At this point, the operating parameters in the initial parameter set can be analyzed from multiple dimensions based on the equipment principle of the target device and the domain knowledge base, so as to facilitate subsequent data classification and determine one or more target key parameters.

[0028] Step S20: Classify the data of each of the operating parameters according to the multi-dimensional analysis results, and select the target key parameters from each parameter category according to the domain knowledge base and operating mechanism.

[0029] Understandably, after conducting multi-dimensional analysis of various operating parameters based on the target equipment's operating principles and domain-specific knowledge base, the results of this multi-dimensional analysis can be used to categorize these parameters, such as parameters characterizing functional performance, parameters characterizing operational status, and parameters characterizing operational conditions. It is the parameter that most directly reflects the operating status of the target equipment. Is with Closely related parameters These are the parameters that best reflect the functionality of the target device. It is a parameter that reflects whether the operating conditions of the target equipment deviate.

[0030] It should be understood that, based on the data classification of various operating parameters according to the results of multi-dimensional analysis, target key parameters are selected in each parameter category according to the domain knowledge base and operating mechanism. For example, parameters are selected from the parameter category that characterizes the operating status. Parameters are selected from the categories of key parameters of the target and parameters that characterize the function. As key parameters of the target, a set of key parameters for the target device can be constructed from these key parameters, which can be specifically represented as follows: .

[0031] Step S30: Using the target key parameter as a benchmark, calculate the target correlation degree between the target key parameter and other operating parameters according to the parameter correlation coefficient calculation strategy.

[0032] It should be understood that after selecting the target key parameters, the correlation between the operating parameters and the target key parameters is analyzed based on the target key parameters. That is, the target correlation between the target key parameters and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy. The target correlation represents the degree of mutual influence between the target key parameters and other operating parameters. For this reason, it is necessary to introduce the parameter correlation coefficient calculation strategy to quantitatively calculate the correlation between parameters.

[0033] Step S40: Select key feature parameters from each of the target key parameters according to the target relevance, and perform fault mode identification on the target device based on the key feature parameters using an attention mechanism and a generative adversarial network identification model.

[0034] It should also be emphasized that key feature parameters refer to the feature parameters used to identify failure modes. These key feature parameters can be selected from the key parameters of various targets based on the relevance of the target, and can be expressed as follows: Because target devices operating in isolated environments may suffer from drawbacks such as a limited number of real fault samples, poor identification of unknown faults, and overfitting during training with small sample sizes in labeled datasets, this embodiment introduces a recognition model based on attention mechanisms and generative adversarial networks to address these shortcomings. This model identifies fault patterns in the target device based on key feature parameters, effectively improving the accuracy of fault pattern identification, enabling fault location-level diagnosis, and determining the fault category. Under conditions of limited equipment operating environments, this embodiment emphasizes information from on-site operating status data, enhancing the equipment's rapid diagnostic and repair capabilities, strengthening intelligent operation and maintenance capabilities, shortening the time required to diagnose equipment faults in isolated environments, and reducing equipment maintenance costs. Furthermore, this embodiment also has the advantage of improving information processing and utilization efficiency under small sample conditions.

[0035] Further, the recognition model based on attention mechanism and generative adversarial network includes a target generation network, a target attention mechanism, and a target recognition network; step S40 includes: sorting each of the target key parameters according to the target relevance; analyzing the parameter sorting results based on a multi-dimensional data analysis strategy, and selecting key feature parameters from each of the target key parameters according to the sorting analysis results; when an abnormal operation of the target device is detected, generating fault mode recognition features based on the target generation network and the key feature parameters, and performing feature enhancement on the fault mode recognition samples based on the target attention mechanism; inputting the enhanced fault mode recognition features and the key feature parameters into the target recognition network, and obtaining the fault mode recognition result output by the target recognition network.

[0036] Understandably, target relevance serves as a crucial basis for ranking key parameters of each target. After calculating the target relevance between key parameters and other operating parameters based on the parameter relevance coefficient calculation strategy, the key parameters of each target are ranked according to their target relevance. At this point, it is necessary to analyze the parameter ranking results based on a multi-dimensional data analysis strategy and select key feature parameters to characterize the equipment status. This multi-dimensional data analysis strategy can be a data analysis strategy covering dimensions such as time domain, frequency domain, time-frequency domain, and statistical domain. In the process of extracting key feature parameters, it is also necessary to combine signal processing and machine learning algorithms.

[0037] It should be noted that time-domain analysis directly extracts statistical features from time series data, providing the most intuitive way to understand the macroscopic state of operating parameters. It assesses the overall operating level, fluctuation magnitude, and presence of abnormal shocks by using parameters such as mean, variance, peak value, and kurtosis. For example, the mean of liquid pressure reflects the system load, while an increase in pressure variance may indicate flow instability, forming the basis for condition assessment and threshold alarms. Frequency-domain analysis decomposes time-domain signals (such as vibrations and pressure pulsations) into a spectrum using Fourier transform, revealing the periodic components hidden in the time waveform. It focuses on the distribution of signal energy at different frequencies, used to identify dominant frequencies or harmonics that match the fault characteristic frequencies of rotating parts (such as bearings and gears). For example, under stable fan speed conditions, specific high-frequency peaks in the spectrum may be early signs of bearing wear. Time-frequency analysis aims to capture the changes in frequency components of non-stationary signals over time. Using a method similar to wavelet transform, it provides both time and frequency information simultaneously. It is suitable for analyzing the parameter behavior of equipment under transient conditions such as startup, shutdown, or sudden load changes. For example, it can analyze the transient pressure response triggered by rapid valve opening adjustments, thus pinpointing the precise moment of a fault. Statistical features characterize equipment status by describing the shape and properties of the distribution of operating parameters. This surpasses the advantages of mean and variance, using indicators such as skewness (measuring distribution asymmetry) and percentiles to sensitively capture changes in data distribution. For instance, a change in the skewness of the valve opening signal distribution may indicate some kind of stagnation or deviation in its operating characteristics, providing clues for early fault diagnosis.

[0038] It should also be noted that, in addition to the aforementioned time-domain analysis, frequency-domain analysis, time-frequency analysis, and statistical characteristics, this also includes model-based characteristics, nonlinear characteristics, and derivative-based characteristics. Model-based characteristics treat a parameter sequence over a period of time as the output of a dynamic system and establish a mathematical model for it using autoregressive models, etc. The coefficients or residuals of this model serve as features, characterizing the inherent dynamic properties of the equipment. When the equipment evolves from a normal state to a fault state, the aforementioned model parameters will change significantly, thus enabling predictive maintenance and state assessment of complex systems. Nonlinear characteristics are used to describe the chaotic and unpredictable dynamic behavior behind complex systems. They quantify the complexity and irregularity of the operating sequence by calculating indicators such as approximate entropy and fractal dimension. For example, when the entropy value of the flow signal of an energy device continuously decreases, it may mean that its flow is shifting from turbulence to a more regular, constrained state, which is usually a sensitive indicator of performance degradation or early failure. The derivative feature focuses on the rate of change of operating parameters, namely the first derivative (rate of change) and the second derivative (acceleration of change) of the parameters. It directly amplifies and reveals the dynamic trend of parameter change, and is extremely sensitive to detecting abrupt changes and trend inflection points. For example, monitoring the rate of change of temperature can detect the problem of decreased efficiency of cooling system earlier than monitoring the absolute value of temperature itself.

[0039] It should be understood that, reference Figure 2 , Figure 2 The diagram illustrates the overall process of fault mode recognition. Specifically, after selecting key feature parameters from the key parameters of each target based on the ranking analysis results, these key feature parameters are used as inputs to the target generation network and the target recognition network. For the target generation network, fault mode recognition features are generated based on the key feature parameters. The fault mode recognition features are then enhanced based on the target attention mechanism. Finally, the enhanced fault mode recognition features and the key feature parameters are combined and input into the target recognition network to achieve fault mode recognition. For example, it can be determined that the fault generated during the operation of the target device belongs to the Nth type of fault.

[0040] It should also be emphasized that when constructing a recognition model for identifying fault modes, the distributed learning and data generation advantages of generative adversarial networks should be utilized to design and implement generators and discriminators suitable for identifying scenarios in isolated environments. This aims to achieve the goal of learning generation tasks with minimal use of labeled data, thereby solving practical problems such as the scarcity of real fault samples and the difficulty of data labeling, and enhancing the robustness and adaptability of the pattern recognition model.

[0041] Furthermore, before the step of generating fault mode recognition features based on the target generation network and the key feature parameters when an abnormal operation of the target device is detected, the method further includes: acquiring historical data and empirical data of fault modes, and generating network training samples based on the historical data and the empirical data; determining the prior distribution of random noise and real sample data based on the network training samples; inputting the prior distribution of random noise into the generator, and acquiring the fault sample data output by the generator; inputting the real sample data and the fault sample data into the discriminator, and acquiring the discrimination probability value output by the discriminator; and training the network based on the discrimination probability value to obtain the target generation network.

[0042] It should be understood that for recognition models based on attention mechanisms and generative adversarial networks, there are target generation networks, target attention mechanisms, and target recognition networks. The target generation network can be a GAN generator network. Before generating fault mode recognition features based on key feature parameters, the target generation network needs to be trained. After processing based on historical and empirical data of fault modes, it is processed by a generator G and a discriminator D. The input of the generator G is the prior distribution of random noise, and the input of the discriminator D is the fault sample data and real sample data output by the generator, which are used to accurately distinguish between real and fake. The output discrimination probability value is between 0 and 1. At this time, the network can be trained based on the discrimination probability value, that is, the network parameters are updated to obtain the target generation network.

[0043] This embodiment collects the operating parameters of the target device on-site and performs multi-dimensional analysis on each operating parameter based on the device's operating principle and domain knowledge base. Based on the multi-dimensional analysis results, the operating parameters are categorized, and target key parameters are selected from each parameter category according to the domain knowledge base and operating mechanism. Using the target key parameters as a benchmark, the target correlation coefficient is calculated between the target key parameters and other operating parameters according to a parameter correlation coefficient calculation strategy. Key feature parameters are selected from the target key parameters based on the target correlation coefficient, and a fault mode identification of the target device is performed based on the key feature parameters using an attention mechanism and a generative adversarial network (GAN) recognition model. Through this method, for scenarios with limited isolated environmental conditions, the on-site collected operating parameters are utilized to the maximum extent. After selecting key feature parameters based on target correlation, a suitable recognition model based on an attention mechanism and GAN for isolated environments is introduced. This achieves the goal of generative task learning with minimal use of labeled data, addressing the practical shortcomings of few real fault samples and difficulties in data labeling, thereby effectively improving the accuracy of fault mode identification.

[0044] In one specific implementation, this application provides steps for calculating the target correlation between key target parameters and other operational parameters. Please refer to... Figure 3 , Figure 3 This is the second flowchart illustrating the feature extraction and recognition method based on field data provided in this application embodiment. Step S30 includes steps S301 to S304: Step S301: Obtain the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change.

[0045] Step S302: Based on the target key parameter, calculate the correlation coefficient between the target key parameter and other operating parameters according to the parameter correlation coefficient calculation strategy, the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change.

[0046] Understandably, unlike the commonly used sensitivity coefficient calculation, the calculation of the target correlation between parameters can be completed with a single perturbation (corresponding to the actual perturbation process) to determine the correlation between all other operating parameters and the target key parameter (corresponding to the j-th parameter in the formula), instead of requiring a separate perturbation for each operating parameter in the parameter list as with the sensitivity coefficient. This greatly improves computational efficiency. After determining the target key parameter, the correlation coefficient can be used to quantitatively calculate the correlation between parameters. Therefore, it is necessary to calculate the correlation coefficient between the target key parameter and other operating parameters according to the parameter correlation coefficient calculation strategy, specifically: .

[0047] in, Indicates the first i Other operating parameters and the first j The number of correlations between key parameters of each target , The first i The values ​​of other operating parameters before and after the change. , The first j The values ​​of key parameters of each target before and after the change.

[0048] Step S303: Count the number of the target key parameters.

[0049] It should be understood that during the operation of the target equipment, there may be more than one key target parameter. Therefore, it is necessary to count the number of key target parameters to improve the accuracy of calculating the target correlation between key target parameters and other operating parameters.

[0050] Furthermore, after step S303, the method further includes: when the number is greater than a preset threshold, obtaining the importance of each of the target key parameters; setting the weight of the correlation coefficient of each of the target key parameters according to the importance; calculating a comprehensive correlation coefficient according to each weight, the correlation coefficient of each of the target key parameters and the number, and quantitatively calculating the target correlation between the target key parameters and other operating parameters according to the comprehensive correlation coefficient.

[0051] It should be noted that when the number of key target parameters exceeds a preset threshold, indicating multiple key target parameters, it is necessary to set the weight of the correlation coefficient for each key target parameter based on its importance. For key target parameters directly related to the task objective or endangering system security, this weight can even be set to the maximum value to reflect its importance in the parameter evaluation process. To simplify the process of setting the weight coefficient, a hierarchical parameter setting method can be adopted, where key target parameters at the same level have the same weight. In this case, a comprehensive correlation coefficient can be calculated by combining the correlation coefficients of each key target parameter with the aforementioned quantity, specifically as follows: .

[0052] in, This represents the overall correlation coefficient. Indicates the first j The weights of the correlation coefficients of the key parameters of each target. This indicates the number of key parameters of the target.

[0053] Step S304: When the quantity equals a preset threshold, the target correlation between the target key parameter and other operating parameters is quantitatively calculated based on the correlation coefficient.

[0054] Understandably, the preset threshold can be set to 1. When the number of target key parameters is equal to the preset threshold, it means that there is one and only one target key parameter. At this time, the target correlation between the target key parameter and other operating parameters can be directly calculated quantitatively based on the target correlation coefficient.

[0055] This embodiment obtains the values ​​of the target key parameter before and after the change, as well as the values ​​of other operating parameters before and after the change. Using the target key parameter as a benchmark, it calculates the correlation coefficient between the target key parameter and other operating parameters according to a parameter correlation coefficient calculation strategy, the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change. It then counts the number of target key parameters. When the number equals a preset threshold, it quantitatively calculates the target correlation degree between the target key parameter and other operating parameters based on the correlation coefficient. By using this method, the correlation coefficient between the target key parameter and other operating parameters is calculated based on the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change. A single perturbation can complete the calculation of the correlation coefficients of all other operating parameters for the target key parameter. Then, combined with the number of target key parameters, the target correlation degree between the target key parameter and other operating parameters is quantitatively calculated, thereby effectively improving the accuracy of the target correlation calculation.

[0056] The following describes the feature extraction and recognition device based on field data provided in this application. The feature extraction and recognition device based on field data described below can be referred to in conjunction with the feature extraction and recognition method based on field data described above. Please refer to... Figure 4 , Figure 4 This is a schematic diagram of the module structure of the feature extraction and recognition device based on field data provided in this application embodiment, including: The analysis module T10 is used to collect the operating parameters of the target equipment on site and perform multi-dimensional analysis on each operating parameter based on the equipment principle of the target equipment and the domain knowledge base.

[0057] The classification module T20 is used to classify the data of each of the operating parameters according to the multi-dimensional analysis results, and select the target key parameters from each parameter category according to the domain knowledge base and operating mechanism.

[0058] The calculation module T30 is used to calculate the target correlation between the target key parameters and other operating parameters based on the target key parameters and according to the parameter correlation coefficient calculation strategy.

[0059] The identification module is used to select key feature parameters from each of the target key parameters according to the target relevance, and to identify the fault mode of the target device based on the key feature parameters according to the identification model of attention mechanism and generative adversarial network.

[0060] This embodiment collects the operating parameters of the target device on-site and performs multi-dimensional analysis on each operating parameter based on the device's operating principle and domain knowledge base. Based on the multi-dimensional analysis results, the operating parameters are categorized, and target key parameters are selected from each parameter category according to the domain knowledge base and operating mechanism. Using the target key parameters as a benchmark, the target correlation coefficient is calculated between the target key parameters and other operating parameters according to a parameter correlation coefficient calculation strategy. Key feature parameters are selected from the target key parameters based on the target correlation coefficient, and a fault mode identification of the target device is performed based on the key feature parameters using an attention mechanism and a generative adversarial network (GAN) recognition model. Through this method, for scenarios with limited isolated environmental conditions, the on-site collected operating parameters are utilized to the maximum extent. After selecting key feature parameters based on target correlation, a suitable recognition model based on an attention mechanism and GAN for isolated environments is introduced. This achieves the goal of generative task learning with minimal use of labeled data, addressing the practical shortcomings of few real fault samples and difficulties in data labeling, thereby effectively improving the accuracy of fault mode identification.

[0061] It is understood that the detailed functional implementation of each of the above modules can be found in the description of the aforementioned method embodiments, and will not be repeated here.

[0062] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0063] Based on the methods in the above embodiments, this application provides an electronic device, please refer to... Figure 5 , Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application.

[0064] It should be noted that the system may include: a processor 10, a communications interface 20, a memory 30, and a communication bus 40. The processor 10, communications interface 20, and memory 30 communicate with each other via the communication bus 40. The processor 10 can invoke logical instructions stored in the memory 30 to execute the methods described in the above embodiments.

[0065] Furthermore, the logical instructions in the aforementioned memory 30 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0066] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0067] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0068] It is understood that the processor in the embodiments of this application can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0069] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor.

[0070] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. Those skilled in the art will readily understand that the above descriptions are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A feature extraction and recognition method based on field data, characterized in that, include: The operating parameters of the target equipment are collected on-site, and the operating parameters are analyzed in multiple dimensions based on the equipment principle and domain knowledge base of the target equipment. Based on the results of multi-dimensional analysis, the data of each of the operating parameters are classified, and target key parameters are selected from each parameter category according to the domain knowledge base and operating mechanism. Based on the target key parameters, the target correlation degree between the target key parameters and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy; Based on the target relevance, key feature parameters are selected from each of the target key parameters, and based on the recognition model of attention mechanism and generative adversarial network, the target device is identified according to the key feature parameters to identify the fault mode.

2. The feature extraction and recognition method as described in claim 1, characterized in that, The steps of collecting the operating parameters of the target equipment on-site and performing multi-dimensional analysis of each operating parameter based on the equipment principle and domain knowledge base of the target equipment include: Collect operating parameters of the target equipment on-site; A lightweight digital twin model adapted to the target device is constructed, and the operating conditions of the target device are simulated based on the lightweight digital twin model. An initial parameter set is constructed based on the simulated operating parameters and the operating parameters of the target equipment acquired on-site. Based on the device principle and domain knowledge base of the target device, a multi-dimensional analysis is performed on each operating parameter in the initial parameter set.

3. The feature extraction and recognition method as described in claim 1, characterized in that, The step of calculating the target correlation between the target key parameter and other operating parameters based on the target key parameter and according to the parameter correlation coefficient calculation strategy includes: Obtain the values ​​of the target key parameters before and after the change, and the values ​​of other operating parameters before and after the change, respectively; Based on the target key parameter, the correlation coefficient between the target key parameter and other operating parameters is calculated according to the parameter correlation coefficient calculation strategy, the values ​​of the target key parameter before and after the change, and the values ​​of other operating parameters before and after the change. Count the number of the target key parameters; When the quantity equals a preset threshold, the target correlation between the target key parameter and other operating parameters is quantitatively calculated based on the correlation coefficient.

4. The feature extraction and recognition method as described in claim 3, characterized in that, After the step of counting the number of the target key parameters, the method further includes: When the number exceeds a preset threshold, the importance of each of the target key parameters is determined. The weights of the relevance coefficients of each of the target key parameters are set according to their importance. A comprehensive correlation coefficient is calculated based on the weights, the correlation coefficients of the target key parameters, and the quantity. The target correlation between the target key parameters and other operating parameters is then quantitatively calculated based on the comprehensive correlation coefficient.

5. The feature extraction and recognition method according to any one of claims 1 to 4, characterized in that, The recognition model based on attention mechanism and generative adversarial network includes a target generation network, a target attention mechanism, and a target recognition network; The step of selecting key feature parameters from each of the target key parameters based on the target relevance, and performing fault mode identification of the target device based on the key feature parameters using an attention mechanism and a generative adversarial network recognition model, includes: The key parameters of each target are sorted according to the target relevance. The parameter ranking results are analyzed based on a multi-dimensional data analysis strategy, and key feature parameters are selected from each of the target key parameters based on the ranking analysis results. When an abnormal operation of the target device is detected, fault mode recognition features are generated based on the target generation network according to the key feature parameters, and the fault mode recognition samples are enhanced based on the target attention mechanism. The enhanced fault mode recognition features and the key feature parameters are input into the target recognition network, and the fault mode recognition results output by the target recognition network are obtained.

6. The feature extraction and recognition method as described in claim 5, characterized in that, Before the step of generating fault mode recognition features based on the key feature parameters according to the target generation network when an abnormal operation of the target device is detected, the method further includes: Acquire historical and empirical data on failure modes, and generate network training samples based on the historical and empirical data; The prior distribution of random noise and the real sample data are determined based on the network training samples. The prior distribution of the random noise is input into the generator, and the fault sample data output by the generator is obtained. The real sample data and the fault sample data are input into the discriminator, and the discrimination probability value output by the discriminator is obtained; The target generation network is obtained by training the network based on the discrimination probability values.

7. A feature extraction and recognition device based on field data, characterized in that, include: The analysis module is used to collect the operating parameters of the target equipment on site and perform multi-dimensional analysis on each operating parameter based on the equipment principle of the target equipment and the domain knowledge base. The classification module is used to classify the data of each of the operating parameters according to the multi-dimensional analysis results, and select the target key parameters from each parameter category according to the domain knowledge base and operating mechanism. The calculation module is used to calculate the target correlation degree between the target key parameters and other operating parameters based on the target key parameters and according to the parameter correlation coefficient calculation strategy. The identification module is used to select key feature parameters from each of the target key parameters according to the target relevance, and to identify the fault mode of the target device based on the key feature parameters according to the identification model of attention mechanism and generative adversarial network.

8. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, When the computer program product is run on a processor, the processor causes the processor to perform the method as described in any one of claims 1-6.

Images