General method for identifying abnormal state of equipment based on power quality data

Abstract

The application relates to a device abnormal state general identification method and device based on power quality data, a computer device, a storage medium and a computer program product. The method comprises the following steps: acquiring a trained identification model, and a labeling accuracy of each state label corresponding to the trained identification model, and determining a model accuracy corresponding to the trained identification model. The trained identification model is used for performing state identification on each target frequency spectrum, and a corresponding predicted state type is obtained. In the case that the predicted state type is an abnormal state, a state label matched with the abnormal state is used as a to-be-verified state type, and based on the labeling accuracy and the model accuracy, an abnormal probability corresponding to the to-be-verified state type is obtained through cumulative probability distribution calculation. The abnormal probability is subjected to effectiveness verification, and whether the target power device is abnormal is determined based on the effectiveness verification result, so that the effectiveness of determining whether the target power device is abnormal is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a device abnormal state general identification method and device based on power quality data, a computer device, a storage medium and a computer program product. BACKGROUND

[0002] With the development of power quality monitoring technology, it is often necessary to detect the state of each power device deployed in the power grid line, so as to ensure the normal operation of the power grid.

[0003] In the traditional technology, the fault type of the faulty power device is determined by analyzing the fault waveform characteristics of the faulty power device. Then, based on the fault type, it is identified whether the power device to be detected has the fault type, and if not, it is determined that the state of the power device to be detected is normal.

[0004] However, the fault type cannot identify different fault states, that is, it cannot effectively identify the state of the power device to be detected. Therefore, there is a problem of low effectiveness of determining the abnormal state of the power device. SUMMARY

[0005] Therefore, it is necessary to provide a device abnormal state general identification method and device based on power quality data, a computer device, a computer readable storage medium and a computer program product to solve the above technical problems.

[0006] In a first aspect, the present application provides a device abnormal state general identification method based on power quality data. The method comprises:

[0007] obtaining a plurality of target data collected from a target power device; wherein the target data is the power quality data of the target power device in a target time period;

[0008] obtaining a trained identification model for state identification, and a label accuracy rate corresponding to each state label in a training label, and determining a model accuracy rate corresponding to the trained identification model;

[0009] determining a target frequency spectrum corresponding to each target data, and performing state identification on each target frequency spectrum through the trained identification model to obtain a corresponding predicted state type;

[0010] in the case where the predicted state type is an abnormal state, taking a state label matched with the abnormal state as a to-be-verified state type, and based on the label accuracy rate and the model accuracy rate, obtaining an abnormal probability corresponding to the to-be-verified state type through cumulative probability distribution calculation;

[0011] The abnormal probability is subjected to validity checking, and whether the target power equipment is abnormal is determined based on a result of the validity checking.

[0012] In a second aspect, the present application further provides a device abnormal state general identification device based on power quality data. The device comprises:

[0013] A first obtaining module is configured to obtain a plurality of target data collected from a target power equipment, wherein the target data is power quality data of the target power equipment in a target time period;

[0014] A second obtaining module is configured to obtain a trained identification model for state identification, and a labeling accuracy corresponding to each state label in a training label, and determine a model accuracy corresponding to the trained identification model;

[0015] A state identification module is configured to determine a target frequency spectrum corresponding to each target data, and perform state identification on each target frequency spectrum through the trained identification model to obtain a corresponding predicted state type;

[0016] A calculating module is configured to, in a case where the predicted state type is an abnormal state, take a state label matched with the abnormal state as a state type to be checked, and obtain an abnormal probability corresponding to the state type to be checked through cumulative probability distribution calculation based on the labeling accuracy and the model accuracy;

[0017] A determining module is configured to perform validity checking on the abnormal probability, and determine whether the target power equipment is abnormal based on a result of the validity checking

[0018] In a third aspect, the present application further provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0019] A plurality of target data collected from a target power equipment are obtained, wherein the target data is power quality data of the target power equipment in a target time period;

[0020] A trained identification model for state identification, and a labeling accuracy corresponding to each state label in a training label are obtained, and a model accuracy corresponding to the trained identification model is determined;

[0021] A target frequency spectrum corresponding to each target data is determined, and state identification is performed on each target frequency spectrum through the trained identification model to obtain a corresponding predicted state type;

[0022] in the presence of a predicted state type being an abnormal state, taking a state label matching the abnormal state as a state type to be verified, and based on the labeling accuracy and the model accuracy, obtaining an abnormal probability corresponding to the state type to be verified through cumulative probability distribution calculation;

[0023] validating the abnormal probability, and determining whether the target power equipment is abnormal based on a validation result.

[0024] In a fourth aspect, the present application also provides a computer readable storage medium. The computer readable storage medium has a computer program stored thereon, and the computer program, when executed by a processor, implements the following steps:

[0025] obtaining a plurality of target data collected from a target power equipment; wherein the target data is power quality data of the target power equipment in a target time period;

[0026] obtaining a trained recognition model for state recognition, and a labeling accuracy corresponding to each state label in a training label, and determining a model accuracy corresponding to the trained recognition model;

[0027] determining a target frequency spectrum corresponding to each target data, and performing state recognition on each target frequency spectrum through the trained recognition model to obtain a corresponding predicted state type;

[0028] in the presence of a predicted state type being an abnormal state, taking a state label matching the abnormal state as a state type to be verified, and based on the labeling accuracy and the model accuracy, obtaining an abnormal probability corresponding to the state type to be verified through cumulative probability distribution calculation;

[0029] validating the abnormal probability, and determining whether the target power equipment is abnormal based on a validation result.

[0030] In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program, and the computer program, when executed by a processor, implements the following steps:

[0031] obtaining a plurality of target data collected from a target power equipment; wherein the target data is power quality data of the target power equipment in a target time period;

[0032] obtaining a trained recognition model for state recognition, and a labeling accuracy corresponding to each state label in a training label, and determining a model accuracy corresponding to the trained recognition model;

[0033] determine target spectrum diagrams corresponding to each target data respectively, and perform state recognition on each target spectrum diagram through the trained recognition model to obtain a corresponding predicted state type;

[0034] In the case where the predicted state type is an abnormal state, a state label matched with the abnormal state is taken as a to-be-verified state type, and an abnormal probability corresponding to the to-be-verified state type is obtained through cumulative probability distribution calculation based on the labeling accuracy and the model accuracy.

[0035] The abnormal probability is subjected to validity verification, and it is determined whether the target power equipment is abnormal based on the validity verification result.

[0036] The device abnormal state general recognition method, device, computer equipment, storage medium and computer program product based on power quality data, by obtaining a plurality of target data collected from a target power equipment, wherein the target data is power quality data of the target power equipment in a target time period. Obtain a trained recognition model for state recognition, and the labeling accuracy of each state label in the training label respectively, and determine the model accuracy corresponding to the trained recognition model. Determine the target spectrum diagram corresponding to each target data respectively, and perform state recognition on each target spectrum diagram through the trained recognition model to obtain a corresponding predicted state type, which realizes accurate recognition of multiple states. In the case where the predicted state type is an abnormal state, directly take the state label matched with the abnormal state as the to-be-verified state type, and obtain the abnormal probability corresponding to the to-be-verified state type through cumulative probability distribution calculation based on the labeling accuracy and the model accuracy. In this way, the abnormal probability of the to-be-verified state type is determined from two dimensions of recognition model and labeling, which greatly improves the accuracy of the abnormal probability. Through validity verification of the abnormal probability, the effectiveness of the abnormal probability corresponding to each to-be-verified state type is ensured. Based on the validity verification result corresponding to each different to-be-verified state type, it can effectively determine whether the target power equipment is abnormal, which greatly improves the effectiveness of determining whether the target power equipment is abnormal. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is an application environment diagram of the device abnormal state general recognition method based on power quality data in one embodiment;

[0038] Figure 2 It is a flowchart of the device abnormal state general recognition method based on power quality data in one embodiment;

[0039] Figure 3 It is a spectrum diagram in one embodiment;

[0040] Figure 4 Fig. 1 is a schematic diagram of a butterfly network in one embodiment;

[0041] Figure 5 Fig. 2 is a schematic diagram of a procedure for obtaining a trained identification model for state identification in one embodiment;

[0042] Figure 6 Fig. 3 is a schematic diagram of a MobileNets network structure in one embodiment;

[0043] Figure 7 Fig. 4 is a schematic diagram of a procedure for constructing a sample library in one embodiment;

[0044] Figure 8 Fig. 5 is a schematic diagram of a general device abnormal state identification method based on power quality data in another embodiment;

[0045] Figure 9 Fig. 6 is a structural block diagram of a device abnormal state general identification apparatus based on power quality data in one embodiment;

[0046] Figure 10 Fig. 7 is an internal structure diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0047] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0048] The device abnormal state general identification method based on power quality data provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the database 102 communicates with the computer device 104 through the network. The data storage system can store the data required by the computer device 104 to process. The data storage system can be integrated on the computer device 104, or placed on the cloud or other network servers. The computer device 104 obtains a plurality of target data collected by the target power device; wherein the target data is the power quality data of the target power device in the target time period. The computer device 104 obtains a trained identification model for state identification, and the annotation accuracy corresponding to each state label in the training label, and determines the model accuracy corresponding to the trained identification model. The computer device 104 determines the target frequency spectrum corresponding to each target data, and identifies the state of each target frequency spectrum through the trained identification model to obtain the corresponding predicted state type. In the case where the predicted state type is an abnormal state, the computer device 104 matches the state label corresponding to the abnormal state as the to-be-verified state type, and based on the annotation accuracy and the model accuracy, the abnormal probability corresponding to the to-be-verified state type is obtained through the cumulative probability distribution calculation. The computer device 104 performs validity verification on the abnormal probability, and determines whether the target power device is abnormal based on the validity verification result, and uploads the result represented by the target power device abnormality to the database 102. Among them, the database 102 is deployed on the edge, that is, a virtual database. The computer device 104 can be a terminal or a server. Among them, the terminal can be, but not limited to, a personal computer, a notebook computer, a smart phone, a tablet computer, etc. The server can be realized by an independent server or a server cluster composed of multiple servers.

[0049] In one embodiment, as shown in Figure 2 , a device abnormal state general identification method based on power quality data is provided. Taking the computer device 104 in Figure 1 as an example, the method includes the following steps:

[0050] Step S202, obtaining a plurality of target data collected by the target power device; wherein the target data is the power quality data of the target power device in the target time period.

[0051] Among them, the power quality data is three-phase voltage data, which can be regarded as waveform time series data.

[0052] Specifically, the computer equipment determines the target time period for collecting data from the target power equipment, and collects data at each target time to obtain target data corresponding to each target collection time. The computer equipment can be a device with data collection and analysis functions, such as a power quality monitoring platform or power quality monitoring point, etc., and is not specifically limited.

[0053] Step S204: Obtain the completed recognition model for state recognition, the annotation accuracy corresponding to each state label in the training labels, and determine the model accuracy corresponding to the completed recognition model.

[0054] The recognition model employs a deep convolutional neural network to identify image features. Its network architecture uses MobileNets to extract deep features from the spectral energy map, connecting different levels of deep features with upsampled shallow features to generate feature maps for multi-scale target recognition in sample images. The state recognition is used to identify the state of power equipment, which can be normal, three-phase short circuit, single-phase ground fault, two-phase short circuit, two-phase ground fault, or open / closed state. The normal state corresponds to a normal state label, the three-phase short circuit to a three-phase short circuit state label, the single-phase ground fault to a single-phase ground fault state label, the two-phase short circuit to a two-phase short circuit label, the two-phase ground fault to a two-phase ground fault label, and the open / closed state to an open / closed state label. The training label can be one of the following: normal state label, three-phase short circuit state label, single-phase ground fault state label, two-phase short circuit label, two-phase ground fault label, or open / closed state label. This training label is used to identify the state of the sample data. Here, labeling accuracy represents the probability of correctly labeling the status tags corresponding to the sample electrical equipment. Model accuracy represents the probability that the trained recognition model will correctly identify the correct device.

[0055] Specifically, the computer device acquires the trained recognition model used for state recognition, as well as the annotation accuracy corresponding to each state label in the training labels. The computer device determines the model accuracy corresponding to the trained recognition model.

[0056] It should be noted that this annotation accuracy can characterize the accuracy of the label annotation dimension, that is, whether the labels on the training data used for training are correct, and whether the labels on the test data used for testing are correct. This model accuracy can characterize the accuracy of the model dimension, that is, whether the model's results are correct.

[0057] Step S206: Determine the target spectrogram corresponding to each target data, and use the trained recognition model to perform state recognition on each target spectrogram to obtain the corresponding predicted state type.

[0058] The target spectrum is a spectrum obtained by merging the frequency spectrum of the sampling points of the three-phase voltage A, B and C. The predicted state type represents the state of the target power equipment, and can be a normal state type, a three-phase short-circuit type, a single-phase ground short-circuit type, a two-phase short-circuit type, a two-phase ground short-circuit type, an opening and closing type, etc.

[0059] Specifically, the computer device performs short-term fast Fourier processing on each target data to obtain a target spectrum corresponding to each target data. The computer device inputs each target spectrum into the trained recognition model to obtain a predicted state type corresponding to each target spectrum.

[0060] For example, for the target data X at t1, the target data X is composed of sampling points of the three-phase voltage A, B and C. The computer device performs short-term fast Fourier processing on each phase voltage to obtain a frequency spectrum a, b and c corresponding to each phase voltage. The frequency spectrum a, the frequency spectrum b and the frequency spectrum c are all plotted according to one cycle, and the three frequency spectrums are in the same coordinate system. The computer device sequentially splices the frequency spectrum a, the frequency spectrum b and the frequency spectrum c to generate a target spectrum, as shown in the target spectrum. Figure 3 The target spectrum shown in the target spectrum is obtained by splicing the frequency spectrums corresponding to the three-phase voltage.

[0061] The short-term fast Fourier is a time domain extraction algorithm based on the Cooley-Tuckey fast Fourier transform algorithm, that is, the sampling signal x(n) (the sampling length is N (2,…, n) is subjected to short-term discrete Fourier transform:

[0062]

[0063] Wherein, k = 0 to N-1.

[0064] Then, the sampling signal is subjected to the following steps: first, the sampling information x(n) with a length of N is decomposed into two short sequences of N1 and N2, wherein N1 is the odd items of the original sampling signal, and N2 is the even items of the original sampling signal. Then, assuming that the sampling points of each cycle of the computer device are 8 points, the butterfly network fast Fourier calculation method is used for calculation, so as to obtain the frequency spectrum corresponding to each phase voltage. The butterfly network calculation schematic diagram is shown in the following figure: Figure 4 The rotation factor is

[0065] It should be noted that each target spectrum corresponds to a predicted state type, and each target data corresponds to a target spectrum. Obviously, each target data corresponds to a predicted state type.

[0066] Step S208, in the case where the prediction state type is an abnormal state, a state label matched with the abnormal state is taken as a state type to be verified, and based on the labeling accuracy and the model accuracy, an abnormal probability corresponding to the state type to be verified is obtained through cumulative probability distribution calculation.

[0067] The prediction state type can be a normal state type, a three-phase short-circuit type, a single-phase ground short-circuit type, a two-phase short-circuit type, a two-phase ground short-circuit type, an opening and closing type, etc. For the prediction state type that is not a prediction state type representing a normal state, it is directly regarded as an abnormal state. The state label is used to identify the data state.

[0068] Specifically, the computer device obtains the prediction state type corresponding to each target frequency spectrum, and determines the abnormal state of each prediction state type. In the case where at least two prediction state types are abnormal states, the computer device takes a state label matched with the abnormal state as a state type to be verified, and determines the frequency spectrum number of the target frequency spectrum corresponding to the same state type to be verified. For each state type to be verified, the computer device obtains an abnormal probability corresponding to the corresponding state type to be verified based on the frequency spectrum number corresponding to the corresponding state type to be verified, the labeling accuracy, the model accuracy, and cumulative probability distribution calculation. In the case where one prediction state type is an abnormal state, the computer device takes a state label matched with the abnormal state as a state type to be verified, and determines the frequency spectrum number of the state type to be verified. The computer device obtains an abnormal probability corresponding to the state type to be verified based on the frequency spectrum number corresponding to the state type to be verified, the labeling accuracy, the model accuracy, and cumulative probability distribution calculation.

[0069] It should be noted that, since the prediction state type representing an abnormal state is the state type to be verified, the case where at least two prediction state types are abnormal states is equivalent to the case where at least two state types to be verified exist. The case where one prediction state type is an abnormal state is equivalent to the case where one state type to be verified exists.

[0070] For example, there are 100 target spectrograms, and each target spectrogram corresponds to an abnormal state type. Among them, there are 3 predicted state types, that is, the first predicted state type is the normal state type (the number of spectrograms corresponding to the normal state type is 20), and the state label matched therewith is the normal label; the second predicted state type is the three-phase short-circuit type (the number of spectrograms corresponding to the three-phase short-circuit type is 60), and the state label matched therewith is the three-phase short-circuit label; and the third predicted state type is the single-phase ground short-circuit type (the number of spectrograms corresponding to the single-phase ground short-circuit type is 20), and the state label matched therewith is the single-phase ground short-circuit label. Therefore, the three-phase short-circuit type and the single-phase ground short-circuit type are both to-be-verified state types. For the case where the to-be-verified state type is the three-phase short-circuit type, the computer device determines the number of spectrograms corresponding to the three-phase short-circuit type, that is, 60. The computer device obtains an abnormal probability P1 corresponding to the to-be-verified state type being the three-phase short-circuit type through cumulative probability distribution calculation based on the number of spectrograms corresponding to the three-phase short-circuit type, the annotation accuracy and the model accuracy. For the case where the to-be-verified state type is the single-phase ground short-circuit type, the computer device determines the number of spectrograms corresponding to the single-phase ground short-circuit type, that is, 20. The computer device obtains an abnormal probability P2 corresponding to the to-be-verified state type being the single-phase ground short-circuit type through cumulative probability distribution calculation based on the number of spectrograms corresponding to the single-phase ground short-circuit type, the annotation accuracy and the model accuracy.

[0071] In step S210, the abnormal probability is verified for validity, and it is determined whether the target power device is abnormal based on the validity verification result.

[0072] The validity verification is used to verify whether the abnormal probability is valid, that is, the validity verification result represents validity, and it is determined that the abnormal probability is valid, and the target power device is in an abnormal state, that is, the target power device is abnormal.

[0073] Specifically, the computer device verifies the abnormal probability corresponding to the to-be-verified state type for validity, and obtains a validity verification result corresponding to the to-be-verified state type. In the case where the validity verification result represents validity, it is determined that the target power device is abnormal, and the computer device uploads the result represented by the abnormality of the target power device to a database deployed at the edge end to realize edge detection.

[0074] It should be noted that the computer device of the present application can real-time acquire target data, and based on the target data, it can real-time identify whether the target power device is abnormal, and send the result represented by whether the target power device is abnormal to the database at the edge end, avoiding uploading massive data to the database at the edge end for processing, greatly reducing the computing pressure of the database at the edge end, and enabling faster and better identification of the state of the target power device.

[0075] For example, there are two types of states to be checked, which are three-phase short-circuit type and single-phase ground short-circuit type. Among them, the abnormal probability P1 of the three-phase short-circuit type, and the abnormal probability P2 of the single-phase ground short-circuit type. If at least one of the validity check results corresponding to the abnormal probability P1 and the abnormal probability P2 is valid, it is determined that the target power equipment is abnormal.

[0076] In the above-mentioned general identification method of equipment abnormal state based on power quality data, a plurality of target data collected by the target power equipment is obtained; wherein the target data is the power quality data of the target power equipment in the target time period. A trained identification model for state identification and the annotation accuracy of each state label corresponding to the training label are obtained, and the model accuracy corresponding to the trained identification model is determined. The target frequency spectrum corresponding to each target data is determined, and each target frequency spectrum is subjected to state identification by the trained identification model to obtain the corresponding predicted state type, thereby achieving accurate identification of multiple states. In the case where the predicted state type is an abnormal state, the state label matched with the abnormal state is directly taken as the state type to be checked, and based on the annotation accuracy and the model accuracy, the abnormal probability corresponding to the state type to be checked is obtained through cumulative probability distribution calculation. In this way, the abnormal probability of the state type to be checked is determined from two dimensions of identification model and annotation, which greatly improves the accuracy of the abnormal probability. Through the validity check of the abnormal probability, the effectiveness of the abnormal probability corresponding to each state type to be checked is ensured. Based on the validity check results of each different state type to be checked, the abnormality of the target power equipment can be effectively determined, which greatly improves the effectiveness of determining whether the target power equipment is abnormal.

[0077] In one embodiment, as shown in FIG. 5, the method comprises the following steps: Figure 5 The step of obtaining a trained identification model for state identification comprises:

[0078] In step S502, training data collected by a sample power equipment is obtained, and the training data is power quality data of the sample power equipment in an abnormal state.

[0079] The training data carries a training label, which can be a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, and an opening and closing label.

[0080] Specifically, the computer device collects multiple sample power devices multiple times to obtain training data. All the training data represent data in which the sample power devices are in an abnormal state. Each training data carries a training label corresponding to the respective training data.

[0081] Step S504, determine the training spectrum corresponding to each training data, and construct a sample library based on the training spectrum.

[0082] Specifically, the computer device performs short-term fast Fourier processing on each training data to obtain a training spectrum corresponding to each training data. The computer device determines an augmented spectrum, which can be generated based on a normal waveform or based on noise data, and the like, and is not limited in particular. The computer device determines a sample spectrum based on the training spectrum and the augmented spectrum, and constructs a sample library based on each sample spectrum. The sample library stores multiple sample spectrums. The augmented spectrum also has a training label corresponding thereto. Therefore, each sample spectrum corresponds to a training label.

[0083] Step S506, construct an initial recognition model, and train the initial recognition model through the sample library to obtain a trained recognition model.

[0084] Specifically, the computer device determines a MobileNets network matching the data characteristics based on the data characteristics of the power device. The terminal constructs an initial recognition model based on the MobileNets network matching the data characteristics, and trains the initial recognition model based on each sample training spectrum in the sample library and the training label corresponding to each sample spectrum to obtain a trained recognition model.

[0085] The MobileNets network structure matching the data characteristics is specifically as shown in Table 1, which includes two structures involving convolution layers (conv), normalization processing (BN), activation functions (ReLU), and separable convolution layers (DepthwiseConv). The structure parameters are as shown in Table 1: Figure 6

[0086] Table 1 Structure Parameters

[0087]

[0088] ​In the embodiment, based on the training data representing that the sample power equipment is in an abnormal state, a sample library for training an initial identification model is constructed, ensuring that the sample spectrum graph contains various states, greatly increasing the information amount of the sample spectrum graph. In this way, the identification model based on the sample library with rich information can be used for detection of various states, so as to ensure the effectiveness of the determination of the target power equipment state.

[0089] In one embodiment, as shown in Figure 7 The sample library is constructed based on the training spectrum graph, including:

[0090] In step S702, waveform recognition is performed on each training data to obtain a waveform recognition result, and the training data with a normal waveform represented by the waveform recognition result is taken as the to-be-truncated data.

[0091] The training data is three-phase voltage data, and the expression of each phase voltage can be represented by the expression of an electromagnetic wave. The three-phase voltage can be represented by a waveform.

[0092] Specifically, the computer device performs waveform recognition on each training data to obtain a waveform recognition result corresponding to each training data. The computer device filters the training data based on each waveform recognition result to obtain training data with a normal waveform represented by the waveform recognition result. The computer device takes the training data with a normal waveform represented by the waveform recognition result as the to-be-truncated data.

[0093] In step S704, a normal waveform is truncated from the to-be-truncated data, and a standard training spectrum graph corresponding to the normal waveform is determined.

[0094] Specifically, the computer device truncates the to-be-truncated data to obtain a normal waveform corresponding to each to-be-truncated data. The computer device performs short-term fast Fourier transform on each normal waveform to obtain a standard training spectrum graph corresponding to each normal waveform.

[0095] It should be noted that according to the technical requirements of the industry standard data acquisition equipment, the waveform before the fault is retained when the acquisition function is triggered. The waveform before the fault includes a normal waveform. Further, in order to ensure the effectiveness of the sample library, the number and proportion of the standard training spectrum graphs can be controlled to be 3:7.

[0096] In step S706, noise data is obtained, and new power quality data is constructed based on the noise data.

[0097] Specifically, the computer device determines a noise type and obtains noise data corresponding to the noise type. The computer device constructs new power quality data based on the noise data.

[0098] For example, the computer device simulates at 30 cycles, samples 1024 points in each cycle, and determines that the noise type is white noise. The computer device determines noise data corresponding to the white noise, determines each phase voltage data based on the noise data, and constructs new power quality data based on the each phase voltage data. For example, the new power quality data X A，B，C The expression of (t) is as follows:

[0099]

[0100] Wherein, t is time, r * is white noise.

[0101] Step S708, determine the new training frequency spectrum corresponding to the new power quality data.

[0102] Specifically, the computer device respectively performs short-term fast Fourier transform on each new power quality data to obtain a new training frequency spectrum corresponding to each new power quality data.

[0103] Step S710, the training frequency spectrum, the standard training frequency spectrum, and the new training frequency spectrum are all used as sample frequency spectrums, and a sample library is constructed based on the sample frequency spectrums.

[0104] Specifically, the computer device uses the training frequency spectrum, the standard training frequency spectrum, and the new training frequency spectrum as sample frequency spectrums. The computer device respectively copies the training frequency spectrum by different preset multiples to obtain each copy frequency spectrum, and uses the copy frequency spectrum as a sample frequency spectrum. The computer device constructs a sample library based on each sample frequency spectrum. For example, the preset multiple can be 6 times, 4 times, 2 times, etc.

[0105] In this embodiment, the standard training frequency spectrum corresponding to each normal waveform is determined by each normal waveform, and the new training frequency spectrum corresponding to each noise data is determined by each noise data. In this way, the sample library can be greatly expanded, ensuring the diversity of sample frequency spectrums in the sample library, which is conducive to improving the accuracy of the identification model in identifying the state of the target power equipment.

[0106] In an embodiment, the determining the model accuracy corresponding to the trained identification model comprises: obtaining test data collected from the sample power equipment, the test data being power quality data when the sample power equipment is in an abnormal state; determining a test label corresponding to the test data, the test label representing a state type of the sample power equipment; determining a test spectrum corresponding to the test data, and performing state identification on the test spectrum by using the trained identification model to obtain a training state type corresponding to the test data; and determining the model accuracy corresponding to the trained identification model based on the test label and the training state type.

[0107] The test label can be a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, or an opening and closing label.

[0108] Specifically, the computer device obtains test data collected from the plurality of sample power equipment multiple times, and determines a test label corresponding to each test data. The computer device performs short-term fast Fourier transform on each test data to obtain a test spectrum corresponding to each test data. The computer device obtains a trained identification model, and determines a training state type corresponding to each test spectrum based on each test spectrum and the trained identification model, and determines a to-be-verified label corresponding to the training state type. The to-be-verified label can be a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, or an opening and closing label. The computer device determines the model accuracy corresponding to the trained identification model based on the to-be-verified label and the test label corresponding to each test spectrum.

[0109] For example, for each test spectrum, the computer device compares the to-be-verified label corresponding to the test spectrum with the test label to obtain a comparison result, and determines a first number of comparison results. The computer device determines a target result representing that the to-be-verified label and the test label are consistent based on the comparison result corresponding to each test spectrum, and determines a second number of target results. The computer device determines an initial accuracy based on the first number and the second number. For example, the initial accuracy can be obtained by dividing the second number by the first number. In a case where the initial accuracy is greater than or equal to a predetermined threshold, the initial accuracy is taken as the model accuracy.

[0110] The ratio of the test data and the training data can be configured according to actual needs, such as 1:9. The predetermined threshold is used to detect whether the trained identification model reaches an identification standard. The greater the initial accuracy is greater than the predetermined threshold, the more standard the trained identification model is.

[0111] In the embodiment, the trained recognition model is subjected to state recognition through test data, and the training state type corresponding to the training data can be determined. In this way, the accuracy and effectiveness of the trained recognition model can be further verified based on each training state type and test label, so as to facilitate effective and high-accuracy recognition of the abnormal state of the target power equipment.

[0112] In one embodiment, based on the annotation accuracy and the model accuracy, the abnormal probability corresponding to the to-be-verified state type is obtained through cumulative probability distribution calculation, including: taking the product of the model accuracy and the annotation accuracy as a state probability, the state probability representing the probability of correctly identifying the state. Determining the target number of target frequency spectrum corresponding to the to-be-verified state type. Based on the state probability and the target number, cumulative probability distribution calculation is performed to determine the abnormal probability corresponding to the to-be-verified state type.

[0113] Specifically, the computer device determines the number of to-be-verified state types and the total number of target frequency spectrums. The computer device determines the number of frequency spectrums of the target frequency spectrum corresponding to each to-be-verified state type. The computer device takes the product of the annotation accuracy and the model accuracy as a state probability, and subtracts a unit quantity from the state probability to obtain a state error probability. For each to-be-verified state type, the computer device obtains the abnormal probability corresponding to the corresponding to-be-verified state type through cumulative probability distribution calculation based on the number of frequency spectrums corresponding to the corresponding to-be-verified state type, the total number of frequency spectrums, the state probability, and the state error rate. The cumulative probability distribution is considered as a binomial distribution.

[0114] For example, let the to-be-verified state type be i, the state probability be Pi, the number of frequency spectrums of the frequency spectrum type corresponding to the to-be-verified state type be m, and the total number of frequency spectrums be n. Then, the cumulative probability distribution calculation is performed according to the following formula:

[0115]

[0116] Wherein, the total number of frequency spectrums n can be the sampling times, which can be preset in advance according to the needs.

[0117] In the embodiment, the probability for cumulative probability distribution is determined through the model accuracy and the annotation accuracy, so that the abnormal probability of the to-be-verified state type is determined from two dimensions of the recognition model and the annotation, which greatly improves the accuracy of the abnormal probability.

[0118] In one embodiment, the effectiveness verification of the abnormal probability includes: comparing the abnormal probability corresponding to the to-be-verified state type with a probability threshold to obtain an effectiveness verification result.

[0119] Specifically, the computer device obtains the abnormal probability corresponding to the state type to be verified, and compares the abnormal probability with the probability threshold to obtain the effectiveness verification result. If the abnormal probability is greater than the probability threshold, it is determined that the effectiveness verification result represents passing. If the abnormal probability is less than or equal to the probability threshold, it is determined that the effectiveness verification result represents failing. In the case where the effectiveness verification result represents passing, it is determined that the target power device has an abnormality.

[0120] For example, in the case where there are at least two state types to be verified, if there is at least one effective verification result representing passing, it is determined that the target power device has an abnormality. For example, the state type to be verified is i, the threshold probability is 0.95, the model accuracy is 0.9, and the label accuracy P label(i) Pi = 0.9 x P label(i) Therefore, the effective verification result representing passing can be represented as follows:

[0121]

[0122] In this embodiment, by performing effectiveness verification on the abnormal probability, the effectiveness of the abnormal probability corresponding to each state type to be verified is ensured. Then, based on the effectiveness verification result, it is determined whether the target power device has an abnormality, which greatly improves the effectiveness of determining whether the target power device has an abnormality.

[0123] In order to more clearly understand the technical solutions of the present application, a more detailed embodiment is provided for description. As shown in Figure 8 Specifically, the implementation steps are as follows:

[0124] Step one: the computer device collects multiple sample power equipment multiple times to obtain training data (i.e. corresponding waveform time series data in the figure), the voltage spectrum energy module in the computer device respectively performs short-term fast Fourier transform on each training data to obtain a training spectrum corresponding to each training data. The abnormal waveform sample library module in the computer device performs waveform identification on each training data to obtain a waveform identification result corresponding to each training data. The abnormal waveform sample library module in the computer device filters the training data based on each waveform identification result to obtain training data whose waveform identification result represents the presence of normal waveform. The abnormal waveform sample library module in the computer device takes the training data whose waveform identification result represents the presence of normal waveform as the to-be-truncated data. The abnormal waveform sample library module in the computer device obtains normal waveforms corresponding to each to-be-truncated data by truncating the to-be-truncated data. The voltage spectrum energy module in the computer device respectively performs short-term fast Fourier transform on each normal waveform to obtain a standard training spectrum corresponding to each normal waveform (i.e. historical data in the figure). The abnormal waveform sample library module in the computer device determines the noise type and obtains noise data corresponding to the noise type. Based on the noise data, new power quality data is constructed. The voltage spectrum energy module in the computer device respectively performs short-term fast Fourier transform on each new power quality data to obtain a new training spectrum corresponding to each new power quality data. The abnormal waveform sample library module in the computer device takes the training spectrum, the standard training spectrum, and the new training spectrum as sample spectrums, and constructs a sample library based on the sample spectrums. The convolutional neural network module in the computer device determines a MobileNets network matched with the data characteristics of the power equipment based on the data characteristics of the power equipment. The convolutional neural network module in the computer device constructs an initial identification model based on the MobileNets network matched with the data characteristics, and trains the initial identification model based on each sample training spectrum in the sample library and the training label corresponding to each sample spectrum (i.e. machine learning in the figure), to obtain a trained identification model.

[0125] Step two: The computer device determines the test data obtained by multiple sample power devices through multiple acquisitions, and determines the test labels corresponding to each test data. The abnormal waveform sample library module in the computer device performs short-term fast Fourier processing on each test data to obtain a test spectrum corresponding to each test data. The convolutional neural network module in the computer device obtains a trained identification model, and determines the training state type corresponding to each test spectrum based on each test spectrum and the trained identification model, and determines the to-be-verified label corresponding to the training state type. The to-be-verified label can be a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, and an opening and closing label. The convolutional neural network module in the computer device determines the model accuracy corresponding to the trained identification model based on the to-be-verified label and the test label corresponding to each test spectrum.

[0126] Step three: The computer device determines the target time period for collecting the target power device, and collects each target time to obtain target data corresponding to each target collection time (i.e., real-time data in the corresponding figure). The abnormal recognition module in the computer device obtains the trained identification model for state recognition sent by the convolutional neural network module (i.e., the pre-trained model in the corresponding figure), and the model accuracy corresponding to the trained identification model. The abnormal recognition module in the computer device obtains the annotation accuracy corresponding to each state label in the training label sent by the abnormal waveform sample library module (i.e., the sample annotation accuracy probability in the corresponding figure). The voltage spectrum energy graph module in the computer device performs short-term fast Fourier processing on each target data to obtain a target spectrum corresponding to each target data, and sends it to the abnormal recognition module. The abnormal recognition module in the computer device inputs each target spectrum into the trained identification model to obtain a predicted state type corresponding to each target spectrum.

[0127] Step four: the anomaly identification module in the computer device obtains the predicted state type corresponding to each target spectrum respectively, and judges the anomaly state of each predicted state type, and takes the state label matched with the anomaly state as the state type to be verified. The anomaly identification module in the computer device determines the number of categories of the state type to be verified, and determines the total number of target spectrums. The anomaly identification module in the computer device determines the number of target spectrums corresponding to each state type to be verified. The anomaly identification module in the computer device takes the product of the annotation accuracy and the model accuracy as the state probability, and takes one unit quantity minus the state probability as the state error probability. For each state type to be verified, the anomaly identification module in the computer device obtains the anomaly probability corresponding to the corresponding state type to be verified based on the number of spectrums corresponding to the corresponding state type to be verified and the total number of spectrums, the state probability and the state error rate, and obtains the anomaly probability corresponding to the corresponding state type to be verified through cumulative probability distribution calculation. The anomaly identification module in the computer device obtains the anomaly probability corresponding to the state type to be verified, and compares the anomaly probability with the probability threshold to obtain the effectiveness verification result. If the anomaly probability is greater than the probability threshold, it is determined that the effectiveness verification result represents pass. If the anomaly probability is less than or equal to the probability threshold, it is determined that the effectiveness verification result represents fail. In the case that the effectiveness verification result represents pass, it is determined that the target power equipment exists anomaly, and anomaly early warning is performed.

[0128] In the embodiment, a plurality of target data collected by the target power equipment is obtained, wherein the target data is power quality data of the target power equipment in a target time period. A trained recognition model for state recognition and a label accuracy of each state label in a training label corresponding to each state label are obtained, and a model accuracy corresponding to the trained recognition model is determined. A target spectrum corresponding to each target data is determined, and each target spectrum is subjected to state recognition by the trained recognition model to obtain a corresponding predicted state type, thereby achieving accurate recognition of multiple states. In the case where the predicted state type is an abnormal state, a state label matched with the abnormal state is directly taken as a to-be-verified state type, and an abnormal probability corresponding to the to-be-verified state type is obtained through cumulative probability distribution calculation based on the label accuracy and the model accuracy. In this way, the abnormal probability of the to-be-verified state type is determined from two dimensions of the recognition model and the label, thereby greatly improving the accuracy of the abnormal probability. The effectiveness of the abnormal probability corresponding to each to-be-verified state type is ensured through effectiveness verification of the abnormal probability. Based on the effectiveness verification results of each different to-be-verified state type, whether the target power equipment is abnormal can be effectively determined, thereby greatly improving the effectiveness of determining whether the target power equipment is abnormal. In addition, through self-learning of sample spectrum in a sample library, an engineering applicable recognition rate can be achieved, and strong general performance and generalization performance are achieved, which can be widely promoted to engineering applications at low cost.

[0129] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0130] Based on the same inventive concept, the embodiments of the present application also provide a device abnormal state general identification apparatus based on power quality data for implementing the device abnormal state general identification method based on power quality data described above. The implementation scheme for solving the problem provided by the apparatus is similar to the implementation scheme described in the above method, so the specific limitations in one or more device abnormal state general identification apparatus based on power quality data embodiments provided below can be referred to the limitations of the device abnormal state general identification method based on power quality data described above, which will not be repeated here.

[0131] In one embodiment, as shown in Figure 9 a device abnormal state general identification apparatus based on power quality data is provided, comprising: a first acquisition module 902, a second acquisition module 904, a state identification module 906, a calculation module 908 and a determination module 910, wherein:

[0132] The first acquisition module 902 is configured to acquire a plurality of target data collected from a target power device; wherein the target data is power quality data of the target power device in a target time period.

[0133] The second acquisition module 904 is configured to acquire a trained identification model for state identification, and a labeling accuracy corresponding to each state label in a training label, and determine a model accuracy corresponding to the trained identification model.

[0134] The state identification module 906 is configured to determine a target frequency spectrum corresponding to each target data, and perform state identification on each target frequency spectrum through the trained identification model to obtain a corresponding predicted state type.

[0135] The calculation module 908 is configured to, in the case that the predicted state type is an abnormal state, take a state label matched with the abnormal state as a to-be-verified state type, and based on the labeling accuracy and the model accuracy, obtain an abnormal probability corresponding to the to-be-verified state type through cumulative probability distribution calculation.

[0136] The determination module 910 is configured to perform validity verification on the abnormal probability, and determine whether the target power device is abnormal based on the validity verification result.

[0137] In one embodiment, the second acquisition module 904 is configured to acquire training data collected from a sample power device, the training data being power quality data of the sample power device in an abnormal state. Determine a training frequency spectrum corresponding to each training data, and based on the training frequency spectrum, construct a sample library. Construct an initial identification model, and train the initial identification model through the sample library to obtain a trained identification model.

[0138] In an embodiment, the second obtaining module 904 is configured to perform waveform recognition on each training data to obtain a waveform recognition result, and take the training data with a normal waveform as the to-be-intercepted data according to the waveform recognition result. A normal waveform is intercepted from the to-be-intercepted data, and a standard training spectrogram corresponding to the normal waveform is determined. Noise data is obtained, and new power quality data is constructed based on the noise data. A new training spectrogram corresponding to the new power quality data is determined. The training spectrogram, the standard training spectrogram, and the new training spectrogram are all taken as sample spectrograms, and a sample library is constructed based on the sample spectrograms.

[0139] In an embodiment, the second obtaining module 904 is configured to obtain test data collected from a sample power device, the test data being power quality data when the sample power device is in an abnormal state, determine a test label corresponding to the test data, the test label representing a state type of the sample power device, determine a test spectrogram corresponding to the test data, and perform state recognition on the test spectrogram by using the trained recognition model to obtain a training state type corresponding to the test data. Based on the test label and the training state type, a model accuracy corresponding to the trained recognition model is determined.

[0140] In an embodiment, the calculating module 908 is configured to take a product of the model accuracy and the labeling accuracy as a state probability, the state probability representing a probability of correctly identifying a state. A target number of target spectrograms corresponding to the to-be-verified state type is determined. The state probability and the target number are subjected to cumulative probability distribution calculation to determine an abnormal probability corresponding to the to-be-verified state type.

[0141] In an embodiment, the determining module 910 is configured to compare the abnormal probability corresponding to the to-be-verified state type with a probability threshold to obtain an effectiveness verification result.

[0142] The above various modules in the device abnormal state universal recognition apparatus based on power quality data can be all or partially implemented by software, hardware, or a combination thereof. The above various modules can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to the above various modules.

[0143] In an embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in FIG. 8. Figure 10As shown in the figure. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store general identification data of device abnormal state based on power quality data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a general identification method of device abnormal state based on power quality data.

[0144] Those skilled in the art can understand that, Figure 10 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0145] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in each of the above method embodiments.

[0146] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps in each of the above method embodiments.

[0147] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by the processor to implement the steps in each of the above method embodiments.

[0148] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0149] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0150] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0151] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A general method for identifying abnormal states of equipment based on power quality data, characterized in that, The method comprises: acquiring a plurality of target data collected from a target power device; wherein the target data is power quality data of the target power device in a target time period; acquiring a trained recognition model for state recognition, and an annotation accuracy of each state label in a training label corresponding to each state label, and determining a model accuracy corresponding to the trained recognition model; wherein acquiring the trained recognition model for state recognition comprises: acquiring training data collected from a sample power device, the training data being power quality data of the sample power device in an abnormal state; the training data carries a training label, and the training label comprises a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, and an opening and closing label; determining a training frequency spectrum corresponding to each training data, and constructing a sample library based on the training frequency spectrum; wherein each training data is subjected to short-term fast Fourier processing to obtain a training frequency spectrum corresponding to each training data; an expanded frequency spectrum is determined, wherein the expanded frequency spectrum is generated based on normal waveforms and noise data; a standard training frequency spectrum corresponding to each normal waveform is determined through each normal waveform, and a new training frequency spectrum corresponding to each noise data is determined through each noise data; a sample frequency spectrum is determined based on the training frequency spectrum and the expanded frequency spectrum, and a sample library is constructed based on each sample frequency spectrum; an initial recognition model is constructed, and the initial recognition model is trained through the sample library to obtain the trained recognition model; wherein test data collected from the sample power device is acquired, the test data being power quality data of the sample power device in an abnormal state; a test label corresponding to the test data is determined, the test label representing a state type of the sample power device; a test frequency spectrum corresponding to the test data is determined, and the test frequency spectrum is subjected to state recognition through the trained recognition model to obtain a training state type corresponding to the test data; a model accuracy corresponding to the trained recognition model is determined based on the test label and the training state type; a target frequency spectrum corresponding to each target data is determined, and each target frequency spectrum is subjected to state recognition through the trained recognition model to obtain a corresponding predicted state type; wherein the target frequency spectrum is obtained by short-term fast Fourier processing of the target data, and is generated by merging frequency spectrums of three-phase voltages; in a case where the predicted state type is an abnormal state, a state label matching the abnormal state is taken as a to-be-verified state type, and an abnormal probability corresponding to the to-be-verified state type is obtained through cumulative probability distribution calculation based on the annotation accuracy and the model accuracy; wherein a product of the model accuracy and the annotation accuracy is taken as a state probability, and the state probability represents a probability of correctly recognizing a state; determining a target number of target spectrograms corresponding to the to-be-verified state type; performing cumulative probability distribution calculation on the state probability and the target number to determine an abnormal probability corresponding to the to-be-verified state type; performing validity verification on the abnormal probability, and determining whether the target power equipment is abnormal based on a validity verification result.

2. The method of claim 1, wherein, The method further includes: performing waveform recognition on each training data to obtain a waveform recognition result, and taking training data with a normal waveform as to-be-intercepted data; intercepting a normal waveform from the to-be-intercepted data, and determining a standard training spectrogram corresponding to the normal waveform; obtaining noise data, and constructing new power quality data based on the noise data; determining a new training spectrogram corresponding to the new power quality data; taking the training spectrogram, the standard training spectrogram, and the new training spectrogram as sample spectrograms, and constructing a sample library based on the sample spectrograms.

3. The method of claim 1, wherein, The validity verification on the abnormal probability includes: comparing the abnormal probability corresponding to the to-be-verified state type with a probability threshold to obtain a validity verification result.

4. A device abnormal state general recognition apparatus based on power quality data, characterized by, The device includes: a first obtaining module configured to obtain a plurality of target data collected from a target power equipment; the target data is power quality data of the target power equipment in a target time period; a second obtaining module configured to obtain a trained recognition model for state recognition, a labeling accuracy of each state label in a training label, and a model accuracy corresponding to the trained recognition model; the trained recognition model for state recognition is obtained by obtaining training data collected from a sample power equipment; the training data is power quality data of the sample power equipment in an abnormal state; the training data carries a training label, and the training label includes a normal state label, a three-phase short-circuit state label, a single-phase ground short-circuit state label, a two-phase short-circuit label, a two-phase ground short-circuit label, and an opening and closing label; a training spectrogram corresponding to each training data is determined, and a sample library is constructed based on the training spectrogram; each training data is subjected to short-term fast Fourier processing to obtain a training spectrogram corresponding to each training data; an extended spectrogram is determined, wherein the extended spectrogram is generated based on a normal waveform and based on noise data; a standard training spectrogram corresponding to each normal waveform is determined through each normal waveform, and a new training spectrogram corresponding to each noise data is determined through each noise data; a sample spectrogram is determined based on the training spectrogram and the extended spectrogram, and a sample library is constructed based on each sample spectrogram; an initial recognition model is constructed, and the initial recognition model is trained through the sample library to obtain a trained recognition model. The test data is power quality data when the sample power equipment is in an abnormal state; a test label corresponding to the test data is determined, and the test label represents a state type of the sample power equipment; a test spectrum corresponding to the test data is determined, and state recognition is performed on the test spectrum by using the trained recognition model to obtain a training state type corresponding to the test data; and a model accuracy corresponding to the trained recognition model is determined based on the test label and the training state type. The state recognition module is configured to determine a target spectrum corresponding to each target data, and perform state recognition on each target spectrum by using the trained recognition model to obtain a corresponding predicted state type; the target spectrum is obtained by performing short-term fast Fourier processing on the target data, and is obtained by merging the spectrum of three-phase voltage. The calculation module is configured to, in a case where the predicted state type is an abnormal state, take a state label matched with the abnormal state as a to-be-verified state type, and obtain an abnormal probability corresponding to the to-be-verified state type by using cumulative probability distribution calculation based on the labeling accuracy and the model accuracy. The product of the model accuracy and the labeling accuracy is taken as a state probability, and the state probability represents a probability of correctly recognizing a state; a target number of target spectrums corresponding to the to-be-verified state type is determined; and cumulative probability distribution calculation is performed on the state probability and the target number to determine an abnormal probability corresponding to the to-be-verified state type. The determination module is configured to perform effectiveness verification on the abnormal probability, and determine whether the target power equipment is abnormal based on an effectiveness verification result.

5. The apparatus of claim 4, wherein, The second acquisition module is further configured to acquire training data collected from a sample power equipment, the training data being power quality data when the sample power equipment is in an abnormal state. Each training spectrum corresponding to each training data is determined, and a sample library is constructed based on the training spectrums; an initial recognition model is constructed, and the initial recognition model is trained by using the sample library to obtain a trained recognition model.

6. The apparatus of claim 4, wherein, The second acquisition module is further configured to perform waveform recognition on each training data to obtain a waveform recognition result, and take training data with a normal waveform as to-be-intercepted data; a normal waveform is intercepted from the to-be-intercepted data, and a standard training spectrum corresponding to the normal waveform is determined. Noise data is acquired, and new power quality data is constructed based on the noise data; a new training spectrum corresponding to the new power quality data is determined. The training spectrum, the standard training spectrum, and the new training spectrum are all taken as sample spectrums, and a sample library is constructed based on the sample spectrums. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 3.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which when executed by a processor, implements the steps of the method of any one of claims 1 to 3.

9. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by a processor, implements the steps of the method of any one of claims 1 to 3.

Images