Circuit breaker state high-precision detection method based on opening and closing current

Abstract

The invention relates to the technical field of electrical variable measurement, in particular to a circuit breaker state high-precision detection method based on opening and closing current. The method comprises the following steps: collecting a plurality of opening and closing current sequences in different states as a training set; acquiring a fundamental frequency sequence of each opening and closing current sequence, and acquiring a frequency spectrum information coefficient of the opening and closing current sequence through the fundamental frequency sequence; dividing all the opening and closing current sequences into four intervals according to the frequency spectrum information coefficient, and obtaining a sub-sequence of the opening and closing current sequence of each interval; the fault correlation degree is obtained through the frequency spectrum information coefficient, the current and the variation coefficient of the subsequences; an attention weight is formed through the fault correlation degree, and a logits value is combined to serve as an input training model; and completing detection through the model. The detection precision of the state of the circuit breaker is improved.

Description

Technical Field

[0001] This application relates to the technical field of electrical variable measurement, and specifically relates to a high-precision detection method for the state of a circuit breaker based on closing and opening currents. Background Art

[0002] Switchgear is an important device in the power system, and the reliability of its performance is directly related to the safe operation of the power system. As the core component of the switchgear, the reliability of the switchgear depends to a large extent on the stability of the working state of its circuit breaker. During the operation of the device, the circuit breaker may have equipment failures, which may further cause abnormal closing and opening of the switchgear, and even serious accidents such as refusal to open and refusal to close. Therefore, high-precision detection of the circuit breaker state is very important for the operation and maintenance of the switchgear, and is also an important measure to ensure the safe operation of the power system.

[0003] Currently, in the industry, the waveform characteristics of the closing and opening currents are calculated and fixed thresholds are set to detect the state of the circuit breaker. Since the fluctuations of the closing and opening currents of the circuit breaker under different states are different, and the thresholds are set only based on some characteristics of the closing and opening current data or through the comparison between the closing and opening current data, the deep information indicating the circuit breaker state cannot be mined from the closing and opening currents, which affects the detection accuracy of the circuit breaker state. And the published patent CN106443433A discloses a circuit breaker state monitoring system and monitoring method based on the closing and opening coil currents, which identifies the mechanical state of the circuit breaker by calculating the similarity of the current vector sets of the current signals of the circuit breaker under normal conditions and the monitored circuit breaker. On the one hand, this method simply relies on the similarity degree between current signals, and it is difficult to further mine the important state information contained in the closing and opening coil currents during the operation of the circuit breaker, resulting in low detection accuracy for early faults or complex faults. On the other hand, the closing and opening coil currents of the circuit breaker under normal conditions will change due to factors such as voltage fluctuations and environmental vibrations. Detecting the circuit breaker state only based on the similarity of the current vector sets of the current signals of the circuit breaker under normal conditions will lead to misjudgment of the circuit breaker state and a decrease in the detection accuracy of the circuit breaker state. Summary of the Invention

[0004] To solve the technical problem of low detection accuracy of the circuit breaker, this application provides a high-precision detection method for the state of a circuit breaker based on closing and opening currents, and the specific technical solutions adopted are as follows: This application proposes a high-precision detection method for the state of a circuit breaker based on closing and opening currents, and this method includes the following steps: Collect a number of closing and opening current sequences in different states as a data set; the different states include normal state and fault state; For each closing and opening current sequence, obtain its amplitude sequence; form a fundamental frequency sequence with the amplitude values around the maximum amplitude in the amplitude sequence; obtain the spectrum information coefficient of the closing and opening current sequence through the length of the fundamental frequency sequence and all the amplitude values in the amplitude sequence except the fundamental frequency sequence. Evenly divide the closing and opening current sequences in the dataset into four intervals according to the size of the spectrum information coefficient; divide the closing and opening current sequences in different intervals into subsequences with different numbers according to the order of each interval; for each subsequence, obtain the spectrum information coefficient of the subsequence through the length of its fundamental frequency sequence and all the amplitude values in the amplitude sequence except the fundamental frequency sequence; calculate the fault correlation degree of the subsequence according to the spectrum information coefficient, the maximum current value, and the coefficient of variation of the subsequence. Set a preset number of neurons for the input layer, hidden layer, and output layer of the neural network; use the ratio of the fault correlation degree of the subsequence to all the fault correlation degrees of the closing and opening current sequence where it is located as the attention weight; combine the logits value of the input layer neuron with its corresponding attention weight as the neuron input to complete the model training; input the closing and opening current sequence to be detected into the trained model to achieve high-precision detection.

[0005] In the above solution, the present application sequentially performs frequency-domain and time-domain analysis on the closing and opening current sequences, divides the closing and opening current sequences according to the information content included in the frequency domain, enhances the role of the closing and opening current subsequences with a higher degree of association with the circuit breaker fault in the subsequent training process of the circuit breaker state detection model, and improves the detection accuracy of the circuit breaker state; in addition, the present application constructs a circuit breaker state detection model based on a neural network. Using the fault correlation degree of the closing and opening current subsequences, an attention mechanism is introduced to adjust the training process of the circuit breaker state detection model. On the one hand, the calculation efficiency is improved, and the model training time is shortened; on the other hand, the interference of invalid information in the closing and opening current data is avoided, the learning ability of the circuit breaker fault information included in the closing and opening current data is enhanced, the performance of the circuit breaker state detection model is improved, and high-precision detection of the circuit breaker state is achieved.

[0006] In one embodiment, the fault states include four typical fault states: low coil voltage, closing spring fatigue, transmission mechanism looseness, and opening tripping device jamming.

[0007] In one embodiment, the method of forming a fundamental frequency sequence with the amplitude values around the maximum amplitude in the amplitude sequence is as follows: Select the frequencies and their corresponding amplitude values within a preset range of the maximum amplitude in the amplitude sequence, and sort the amplitude values according to the frequency size to obtain the fundamental frequency sequence.

[0008] In one embodiment, the method of obtaining the spectrum information coefficient through the length of the fundamental frequency sequence and all the amplitude values in the amplitude sequence except the fundamental frequency sequence is as follows: The ratio of the sum of all amplitude values except the fundamental frequency sequence in the amplitude sequence to the sum of all amplitude values in the amplitude sequence is denoted as the first amplitude ratio; The spectral information coefficient is positively correlated with the length of the fundamental frequency sequence and the first amplitude ratio.

[0009] In one embodiment, the method of evenly dividing the switching-on and switching-off current sequences in the dataset into four intervals according to the spectral information coefficient is as follows: Sort all the spectral information coefficients in the dataset in ascending order to form a spectral information coefficient sequence, and evenly divide the spectral information coefficient sequence into four intervals by quartiles, denoted as the first interval, the second interval, the third interval, and the fourth interval from small to large.

[0010] In one embodiment, the method of dividing the switching-on and switching-off current sequences in different intervals into different numbers of subsequences according to the order of each interval is as follows: Divide the switching-on and switching-off current sequences in the first interval into 10 subsequences; divide the switching-on and switching-off current sequences in the second interval into 20 subsequences; divide the switching-on and switching-off current sequences in the third interval into 40 subsequences; divide the switching-on and switching-off current sequences in the fourth interval into 80 subsequences.

[0011] In one embodiment, the fault correlation degree is positively correlated with the spectral information coefficient of the subsequence, the maximum value of the switching-on and switching-off current data, and the coefficient of variation.

[0012] In one embodiment, the number of neurons in the input layer is the same as the number of switching-on and switching-off current data in each switching-on and switching-off current sequence.

[0013] In one embodiment, the attention weight is the ratio of the fault correlation degree of each subsequence to the sum of the fault correlation degrees of all subsequences in the corresponding switching-on and switching-off current sequence.

[0014] In one embodiment, the input of the hidden layer is the product of the logits value of the input layer neuron and its corresponding attention weight.

[0015] The beneficial effects of this application are: This application sequentially performs frequency-domain and time-domain analyses on the opening and closing current sequences, divides the opening and closing current sequences according to the amount of information contained in the frequency domain, enhances the role of the opening and closing current subsequences with a higher degree of association with breaker faults in the subsequent training process of the breaker state detection model, and improves the detection accuracy of the breaker state. In addition, based on a neural network, this application constructs a breaker state detection model. Using the fault correlation degree of the opening and closing current subsequences, an attention mechanism is introduced to adjust the training process of the breaker state detection model. On the one hand, the calculation efficiency is improved and the model training time is shortened. On the other hand, the interference of invalid information in the opening and closing current data is avoided, the learning ability of the breaker fault information contained in the opening and closing current data is enhanced, the performance of the breaker state detection model is improved, and the high-precision detection of the breaker state is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of a high-precision breaker state detection method based on opening and closing currents provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To further elaborate on the technical means and effects adopted by the present application to achieve the intended invention purpose, the following, in combination with the drawings and preferred embodiments, details the specific implementation manner, structure, features, and effects of a high-precision breaker state detection method based on opening and closing currents proposed according to the present application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0020] An embodiment of a high-precision breaker state detection method based on opening and closing currents: The following specifically describes the specific solution of a high-precision breaker state detection method based on opening and closing currents provided by the present application in combination with the drawings.

[0021] Please refer to Figure 1, which shows a flowchart of a high-precision detection method for circuit breaker status based on switching-on and switching-off currents provided by an embodiment of the present application. The method includes the following steps: Step S001, collect a number of switching-on and switching-off current sequences in different states as a data set.

[0022] The switching-on and switching-off current data of the circuit breaker contains rich information. According to the change characteristics of the switching-on and switching-off current data, the operation of the switching-on and switching-off mechanism of the circuit breaker can be obtained, so as to judge whether the circuit breaker has faults such as low voltage, fatigue of the closing spring, looseness of the transmission mechanism, etc., in order to detect the status of the circuit breaker.

[0023] Directly using the collected switching-on and switching-off current data to judge the fault status of the circuit breaker has strong subjectivity, and there are deviations in the constructed data set labels, which directly affect the subsequent training of the circuit breaker status detection model.

[0024] To avoid the deviation introduced by the artificial labels of the data set, the present invention constructs the data set through multiple repeated experiments. The specific process is as follows: Install the switching-on and switching-off current acquisition device on the switching-on and switching-off operation power line to collect the switching-on and switching-off current data of the circuit breaker. In this embodiment, the test object for constructing the data set is a 35kV outdoor high-voltage SF6 circuit breaker, and the switching-on and switching-off current acquisition device uses a Hall current sensor, and the acquisition frequency of the switching-on and switching-off current data is 10kHz.

[0025] Collect the switching-on and switching-off current data for a preset time and sort it in chronological order to form a switching-on and switching-off current sequence. In this embodiment, considering the action time of the circuit breaker for switching off and switching on, the acquisition time is set to 200ms, that is, the number of switching-on and switching-off current data in each switching-on and switching-off current sequence is 2000.

[0026] The present invention constructs the data set by using the simulated fault method, and collects the switching-on and switching-off current sequences in the normal working state and the fault state respectively. In this embodiment, the fault states include four typical fault states: low coil voltage, fatigue of the closing spring, looseness of the transmission mechanism, and jamming of the switching-off tripping device, and the proportion of the collected data is 3:2:2:2:1.

[0027] Specifically, the low coil voltage fault of the circuit breaker in the test is achieved by lowering the coil voltage; the fatigue of the closing spring is achieved by loosening the fixing bolt of the closing spring; the looseness of the transmission mechanism is achieved by loosening the base bolt; and the jamming of the switching-off lock device is achieved by using a card to block the switching-off lock device.

[0028] The opening and closing current sequences of the circuit breaker in different states are collected by the above method, and the opening and closing current sequences obtained by the acquisition device and their corresponding circuit breaker states are jointly composed into a dataset sample for subsequent training of the circuit breaker state detection model. The dataset contains N opening and closing current sequences, where N is 1000 in this embodiment.

[0029] Thus, the opening and closing current sequences in the normal state and different fault states are obtained.

[0030] Step S002: For each opening and closing current sequence, obtain its fundamental frequency sequence, and obtain the spectral information coefficient of the opening and closing current sequence through the fundamental frequency sequence.

[0031] The time-varying nature of the opening and closing currents of the circuit breaker in different states results in differences in the distribution positions of the opening and closing current data containing the key information of the circuit breaker state in the same state. Due to the complex time-domain characteristics of the opening and closing current sequences, if they are directly input into the neural network, it will be difficult for the model to effectively extract the key state information, thereby reducing the detection accuracy.

[0032] To capture the circuit breaker state information contained in the opening and closing current data, it is necessary to obtain the correlation degree between each opening and closing current data and the circuit breaker fault, and estimate the circuit breaker fault information contained in different opening and closing current data.

[0033] Take each opening and closing current sequence as the input, and output its amplitude sequence through the FFT algorithm to estimate the circuit breaker state information contained in the opening and closing current sequence.

[0034] The fundamental frequency of the opening and closing current sequence is the main frequency component, corresponding to the maximum amplitude frequency of the amplitude sequence. The fundamental frequency is usually closely related to the working state of the circuit breaker. When a mechanical fault occurs in the circuit breaker, high-frequency oscillations will appear in the opening and closing current sequence, and the high-frequency components deviating from the fundamental frequency in the amplitude sequence can accurately characterize the corresponding fault.

[0035] Select the frequencies within the preset range of the maximum amplitude in the amplitude sequence and their corresponding amplitude values, and sort the amplitude values according to the frequency size to obtain the fundamental frequency sequence. In this embodiment, the preset range size is 3 dB. By screening the fundamental frequency sequence, the influence of noise and irrelevant information can be reduced, and the frequency domain features closely related to the circuit breaker state can be focused on.

[0036] On the one hand, the frequency components around the fundamental frequency are more concentrated in the normal state of the circuit breaker. The longer the fundamental frequency sequence, the more dispersed the fundamental frequency distribution, the more state information contained in the switching current sequence, the greater the probability of phenomena such as vibration during the switching operation of the circuit breaker, and the greater the possibility of a fault state, so it is necessary to strengthen its detection. On the other hand, since the fluctuations of the switching current caused by the circuit breaker fault are usually in the high-frequency part of the amplitude sequence, when the proportion of the amplitude energy outside the fundamental frequency sequence in the amplitude sequence is larger, the probability that the corresponding switching current sequence contains the circuit breaker fault state information is greater, and the greater the role in extracting the key information of the circuit breaker state.

[0037] Based on the above analysis, the spectral information coefficient is obtained through the length of the fundamental frequency sequence and all the amplitude values except the fundamental frequency sequence in the amplitude sequence. The ratio of the sum of all the amplitude values except the fundamental frequency sequence in the amplitude sequence to the sum of all the amplitude values in the amplitude sequence is denoted as the first amplitude ratio.

[0038] The spectral information coefficient is positively correlated with the length of the fundamental frequency sequence and the first amplitude ratio.

[0039] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large; the specific relationship is determined by the actual application, and this application does not make special restrictions.

[0040] Preferably, in this embodiment, the expression of the spectral information coefficient is: , B represents the length of the fundamental frequency sequence; R represents the first amplitude ratio; F represents the spectral information coefficient of the amplitude sequence. The spectral information coefficient reflects the circuit breaker fault state information contained in the corresponding switching current sequence.

[0041] The larger the spectral information coefficient, the more circuit breaker fault information contained in the corresponding switching current sequence, the greater the probability that the circuit breaker is in a fault state, and the greater the role in subsequent detection of the circuit breaker state.

[0042] It should be noted that negative correlation means that when one variable increases, the other variable decreases, and the two variables change in opposite directions. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small; the specific relationship can be a ratio relationship, a subtraction relationship, etc., which is determined by the actual application, and this application does not make special restrictions.

[0043] So far, the spectral information coefficient of each switching current sequence has been obtained.

[0044] Step S003: Divide all the switching-on and switching-off current sequences into four intervals according to the spectral information coefficient, and obtain the subsequences of the switching-on and switching-off current sequences in each interval; obtain the fault correlation degree through the spectral information coefficient, current and coefficient of variation of the subsequences.

[0045] Through the above steps, the spectral information coefficients of each switching-on and switching-off current sequence are obtained. Sort all the spectral information coefficients in the dataset in ascending order to form a spectral information coefficient sequence. For the spectral information coefficient sequence, divide it into four intervals evenly by quartiles, and denote them as the first interval, the second interval, the third interval, and the fourth interval in ascending order.

[0046] Since each spectral information coefficient corresponds to a switching-on and switching-off current sequence, that is, all the switching-on and switching-off current sequences are divided into four intervals. Based on the different intervals where the switching-on and switching-off current sequences are located, the switching-on and switching-off current sequences are evenly divided into subsequences with different numbers.

[0047] On the one hand, for the switching-on and switching-off current sequences with smaller spectral information coefficients, longer subsequences are generated by division to avoid the influence of their redundant information on the breaker state detection, that is, the number of subsequences divided from the switching-on and switching-off current sequences with smaller spectral information coefficients is smaller. On the other hand, the larger the spectral information coefficient of the switching-on and switching-off current sequence, the shorter the obtained subsequence by division, which is convenient for more precisely capturing local information, that is, the number of subsequences divided from the switching-on and switching-off current sequences with larger spectral information coefficients is larger.

[0048] Therefore, in this embodiment, the switching-on and switching-off current sequences in the first interval are divided into 10 subsequences; the switching-on and switching-off current sequences in the second interval are divided into 20 subsequences; the switching-on and switching-off current sequences in the third interval are divided into 40 subsequences; the switching-on and switching-off current sequences in the fourth interval are divided into 80 subsequences.

[0049] The breaker state information contained in the subsequences at different positions of the switching-on and switching-off current sequences is different. When the subsequence is near the inflection point of the switching-on and switching-off current data, the internal mechanical mechanism of the breaker is in a moving state, and the change of the coil current in the breaker is large, and the corresponding switching-on and switching-off current data contains more state information. At the same time, when the coil current in the breaker is large, the internal mechanical mechanism of the breaker is close to the start and stop stages of the movement, which is a high-incidence stage of faults, and the switching-on and switching-off current data at this point is more likely to characterize the subtle fault state of the breaker. Therefore, the more breaker state information is contained in the corresponding subsequence.

[0050] In addition, considering that the spectral information coefficient reflects the breaker fault state information contained in the corresponding switching current data from the frequency domain perspective of the sequence, the spectral information coefficients of different subsequences are calculated and obtained in the same way to characterize the breaker fault state information contained in their switching current data from the frequency domain analysis perspective, thereby reflecting the correlation degree between the subsequence and the breaker fault state. That is, the spectral information coefficient of the subsequence is obtained through the length of the fundamental frequency sequence corresponding to the subsequence and the first amplitude ratio.

[0051] Moreover, considering the differences in the magnitudes of the switching current data of different subsequences, the coefficient of variation is used to reflect the fluctuation degree of the subsequence in the time domain, eliminating the influence of the differences in the magnitudes of the switching current data on characterizing the data fluctuation degrees of different subsequences.

[0052] Based on the above analysis, the fault correlation degree of the subsequence is calculated through the spectral information coefficient, the maximum current value, and the coefficient of variation of the subsequence.

[0053] The fault correlation degree is positively correlated with the spectral information coefficient of the subsequence, the maximum value of the switching current data, and the coefficient of variation.

[0054] Preferably, in this embodiment, the expression of the fault correlation degree is: , represents the spectral information coefficient of the k-th subsequence in the switching current sequence; represents the maximum value of the switching current data in the k-th subsequence, represents the coefficient of variation of the k-th subsequence, represents the fault correlation degree of the k-th subsequence.

[0055] The fault correlation degree reflects the correlation degree between the information contained in the switching current data within the subsequence and the breaker fault state. On the one hand, the fault correlation degree characterizes the role of the position of the subsequence in detecting the breaker state. The greater the fault correlation degree, the greater the subsequent role in detecting the breaker state. On the other hand, the fault correlation degree reflects the amount of breaker fault information contained in the current subsequence. The greater the fault correlation degree of the subsequence, the more attention needs to be paid to its switching current data in subsequent detections to improve the detection accuracy of the breaker state.

[0056] So far, the fault correlation degrees of all subsequences in the switching current sequence have been obtained.

[0057] Step S004, constructing an attention weight through the fault correlation degree and using the logits value as the input to train the model; completing the detection through the model.

[0058] This application constructs a breaker state detection model based on a neural network. Specifically, in this embodiment, a BP neural network is selected as the detection model.

[0059] There is a lot of redundant closing and opening current data in each data sample. These data points may be noise, background interference, or current data that has no direct correlation with the circuit breaker status. Under normal conditions of the circuit breaker, due to factors such as vibration during equipment operation and voltage changes, there are certain normal fluctuations in the closing and opening current sequences.

[0060] After the closing and opening current sequence is input into the circuit breaker status detection model, the neural network will attempt to learn all input features, including those that have nothing to do with the circuit breaker status. On the one hand, this leads to an increase in network training time, and the model may overfit irrelevant features, thereby reducing its generalization ability and detection accuracy. On the other hand, during the training process, the model will overly focus on the fluctuation features under normal conditions of the circuit breaker, masking the true fault features, reducing the model's ability to extract effective features, and affecting the detection performance of the model.

[0061] To overcome the above defects, this application introduces an attention mechanism into the circuit breaker status detection model, enabling the model to focus on the parts of the input sequence that contain more circuit breaker status information, thereby improving the performance and robustness of the model.

[0062] Specifically, the BP neural network in this application is divided into an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is the same as the number of closing and opening current data in each closing and opening current sequence; then a hidden layer and an output layer with a preset number of neurons are set. In this embodiment, the number of neurons is 2000, the number of neurons in the hidden layer is 512, and the number of neurons in the output layer is 1; the neuron in the output layer represents the two states of normal and faulty of the circuit breaker.

[0063] Considering that the fault correlation degree of each subsequence represents the amount of circuit breaker fault information it contains, the larger its value, the greater its role in detecting the circuit breaker status. Therefore, the ratio of the fault correlation degree of each subsequence to the total fault correlation degrees in the closing and opening current sequence is used as the attention weight.

[0064] In this embodiment, the expression of the attention weight is: , where K represents the number of subsequences in the closing and opening current sequence; and respectively represent the fault correlation degrees of the i-th and k-th subsequences in the closing and opening current sequence; represents the attention weight of the closing and opening current data within the k-th subsequence in the closing and opening current sequence.

[0065] The core function of the attention mechanism is to dynamically allocate weights to different input features, thereby highlighting important features and suppressing irrelevant features. Subsequences with a high fault correlation degree have a greater impact on detecting the circuit breaker state. Therefore, a higher weight is assigned to them through the attention mechanism to enhance the model's learning ability of the circuit breaker state information and reduce the interference of irrelevant information.

[0066] The product of the logits value of the input layer neuron and its corresponding attention weight is used as the input of the hidden layer neuron, and the BP neural network model is trained based on the constructed dataset according to the error backpropagation algorithm.

[0067] The to-be-detected opening and closing current sequences obtained by the acquisition device are input into the trained circuit breaker state detection model, and the corresponding circuit breaker state is output to achieve high-precision detection of the circuit breaker state.

[0068] It should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

[0069] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A high-precision detection method for circuit breaker status based on opening and closing current, characterized in that: The method comprises the following steps: Collecting a number of opening and closing current sequences in different states as a data set; the different states include a normal state and a fault state; For each opening and closing current sequence, its amplitude sequence is obtained; the amplitude values ​​around the maximum amplitude in the amplitude sequence constitute a fundamental frequency sequence; the spectrum information coefficient of the opening and closing current sequence is obtained through the length of the fundamental frequency sequence and all the amplitude values ​​in the amplitude sequence except the fundamental frequency sequence; The opening and closing current sequence in the data set is evenly divided into four intervals according to the size of the spectrum information coefficient; the opening and closing current sequences in different intervals are divided into different numbers of subsequences according to the order of each interval; for each subsequence, the spectrum information coefficient of the subsequence is obtained through the length of its base frequency sequence and all amplitude values ​​in the amplitude sequence except the base frequency sequence; the fault correlation degree of the subsequence is calculated according to the spectrum information coefficient, current maximum value and variation coefficient of the subsequence; A preset number of neurons is set for the input layer, hidden layer and output layer of the neural network; the ratio of the fault correlation degree of the subsequence and the fault correlation degree of all the faults in the opening and closing current sequence in which it is located is used as the attention weight; the logits value of the input layer neuron and its corresponding attention weight are combined as the neuron input to complete the model training; the opening and closing current sequence to be detected is input into the trained model to achieve high-precision detection.

2. A high-precision detection method for circuit breaker status based on opening and closing current according to claim 1, characterized in that: The fault conditions include four typical fault conditions: low coil voltage, fatigue of closing spring, loose transmission mechanism and jamming of opening release device.

3. A high-precision detection method for circuit breaker status based on opening and closing current as claimed in claim 1, characterized in that: The method of forming a base frequency sequence from amplitude values ​​around the maximum amplitude in the amplitude sequence is: In the amplitude sequence, the frequencies within the preset range of the maximum amplitude and their corresponding amplitude values ​​are selected, and the amplitude values ​​are sorted according to the frequency to obtain the fundamental frequency sequence.

4. A high-precision detection method for circuit breaker status based on opening and closing current as claimed in claim 1, characterized in that: The method for obtaining the spectrum information coefficient by using the length of the baseband sequence and all the amplitude values ​​in the amplitude sequence except the baseband sequence is: The ratio of the cumulative sum of all amplitude values ​​in the amplitude sequence except the fundamental frequency sequence to the cumulative sum of all amplitude values ​​in the amplitude sequence is recorded as the first amplitude ratio; The spectrum information coefficient is positively correlated with the length of the baseband sequence and the first amplitude ratio.

5. A high-precision detection method for circuit breaker status based on opening and closing current as claimed in claim 1, characterized in that: The method for evenly dividing the opening and closing current sequence in the data set into four intervals according to the size of the spectrum information coefficient is: All the spectral information coefficients in the data set are sorted in ascending order to form a spectral information coefficient sequence, and the spectral information coefficient sequence is divided into four intervals by quartiles, which are recorded as the first interval, the second interval, the third interval, and the fourth interval from small to large.

6. A high-precision detection method for circuit breaker status based on opening and closing current according to claim 1, characterized in that: The method of dividing the opening and closing current sequences of different intervals into different numbers of sub-sequences according to the order of each interval is: The opening and closing current sequence in the first interval is divided into 10 sub-sequences; the opening and closing current sequence in the second interval is divided into 20 sub-sequences; the opening and closing current sequence in the third interval is divided into 40 sub-sequences; and the opening and closing current sequence in the fourth interval is divided into 80 sub-sequences.

7. A high-precision detection method for circuit breaker status based on opening and closing current as claimed in claim 1, characterized in that: The fault correlation degree is positively correlated with the frequency spectrum information coefficient of the subsequence, the maximum value of the opening and closing current data, and the coefficient of variation.

8. A high-precision detection method for circuit breaker status based on opening and closing currents as claimed in claim 1, characterized in that: The number of neurons in the input layer is the same as the number of opening and closing current data in each opening and closing current sequence.

9. A high-precision detection method for circuit breaker status based on opening and closing currents as claimed in claim 1, characterized in that: The attention weight is the ratio of the fault correlation degree of each subsequence to the cumulative sum of the fault correlation degrees of all subsequences in the corresponding opening and closing current sequence.

10. A high-precision detection method for circuit breaker status based on opening and closing currents according to claim 1, characterized in that: The input of the hidden layer is the product of the logits value of the input layer neuron and its corresponding attention weight.

Images