Breaker fault determination method, system and device and storage medium

Abstract

The invention discloses a circuit breaker fault judgment method, system and device and a storage medium. The method comprises the following steps: acquiring current data formed based on the action of an operating mechanism of the circuit breaker; and inputting a current feature vector formed based on the current data into a pre-trained classification model to obtain a classification result which is output by the classification model and represents the fault type of the high-voltage circuit breaker. The classification model is trained by fusing the self-organizing mapping algorithm and the clustering algorithm, and the clustering center point used for the clustering algorithm is determined based on the self-organizing mapping algorithm in the training process, so that multiple times of training caused by multiple times of valuing of the clustering center point during the training of the clustering algorithm is avoided, a better clustering effect can be quickly obtained, and the clustering efficiency is improved. The training efficiency of the classification model is improved, and rapid judgment of circuit breaker faults is facilitated.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment fault diagnosis technology, and specifically to a method, system, device and storage medium for determining circuit breaker faults. Background Technology

[0002] The reliability of circuit breaker performance directly affects the reliable operation of the power system. Among various circuit breaker (especially high-voltage circuit breaker) faults, operating mechanism faults account for the highest proportion. Therefore, condition assessment of the circuit breaker operating mechanism is of great significance. The opening and closing coil current curves can reflect the working status of the opening and closing electromagnets and the mechanism itself. Therefore, by analyzing the opening and closing coil current curves of the operating mechanism, fault signs of the circuit breaker operating mechanism can be detected in a timely manner.

[0003] Currently, methods for determining circuit breaker faults based on the current of the opening and closing coils typically include: performing principal component analysis on all characteristic quantities of the opening and closing coil current signal, reducing the dimensionality of the characteristic quantities, and performing K-means clustering analysis on the dimensionality-reduced data to obtain the circuit breaker's state information.

[0004] Because the K-means clustering algorithm requires randomly setting the initial centroid of each cluster during training, the set initial centroid will affect the clustering effect. In order to ensure a better clustering effect, multiple clustering training sessions need to be performed, which results in a large amount of computation and low model training efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device and storage medium for determining circuit breaker faults. By using a self-organizing map algorithm to determine cluster centers, the invention avoids multiple training iterations caused by taking multiple values ​​for cluster centers, thus solving the problem of low model training efficiency.

[0006] This invention is achieved through the following technical solution:

[0007] The first aspect of this application provides a method for determining circuit breaker faults, the method comprising:

[0008] Acquire current data based on the actions of the circuit breaker's operating mechanism;

[0009] The current feature vector formed based on the current data is input into a pre-trained classification model to obtain the classification result representing the fault type of the high-voltage circuit breaker output by the classification model.

[0010] The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centroids through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centroids to obtain the clustering results.

[0011] In one feasible implementation, the method further includes: the classification model is trained in the following manner:

[0012] A range of values ​​for the number of neurons is set for the self-organizing map algorithm. For each number of neurons within the range, a pre-training process for the hybrid model is performed to obtain the corresponding candidate classification model.

[0013] Based on preset evaluation criteria, a classification model is selected from the candidate classification models as the classification model used for inference.

[0014] In one feasible implementation, for a number of neurons within the range of values, the following pre-training process is performed:

[0015] The training samples are input into the self-organizing map algorithm to obtain the weight vector of each neuron output by the self-organizing map algorithm; wherein the value of the weight vector is determined based on the similarity between the training samples and the neurons;

[0016] The weight vector is used as the center point of the clustering algorithm. The training samples are input into the clustering model to obtain the clustering results of the training samples output by the clustering model.

[0017] In one feasible implementation, the step of selecting a classification model from the candidate classification models based on preset evaluation criteria as the classification model used for inference specifically includes:

[0018] Based on the distance between the centroids of two different clusters in the clustering results, and the dispersion of data points in the same cluster, the similarity levels corresponding to two different clusters in the clustering results are determined.

[0019] Based on the aforementioned similarity levels, the metric values ​​for the clustering results are determined;

[0020] Select the minimum value from the various metrics values ​​corresponding to each clustering result, and use the classification model corresponding to the minimum value as the classification model for inference.

[0021] In one feasible implementation, the metric value of the clustering result is determined based on the aforementioned similarity levels, specifically including:

[0022] For each cluster in the clustering results, select the maximum similarity value between that cluster and other clusters;

[0023] The mean of the maximum similarity values ​​corresponding to each class in the clustering results is used as the metric value of the clustering results.

[0024] In one feasible implementation, the method further includes: filtering the acquired current data and forming a current feature vector based on the filtered current data;

[0025] The filtering process for the acquired current data specifically includes:

[0026] Acquire the target current data, as well as the adjacent current data at two sampling times before and after the target current data;

[0027] The target current data and each neighboring current data are sorted according to their numerical values, and the value of the target current data is replaced with the median value in the sorting result.

[0028] A second aspect of this application provides a circuit breaker fault determination system, the system comprising:

[0029] The signal acquisition unit is used to acquire current data generated by the operation of the high-voltage circuit breaker's operating mechanism.

[0030] The fault determination unit is used to input the current feature vector formed based on the current data into a pre-trained classification model to obtain the classification result output by the classification model that characterizes the fault type of the high-voltage circuit breaker.

[0031] The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centroids through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centroids to obtain the clustering results.

[0032] In one feasible implementation, the data acquisition unit includes a current acquisition unit, a signal processing unit, and a data transmission unit connected in sequence.

[0033] The current acquisition unit is used to filter, amplify, and perform analog-to-digital conversion on the current signal generated by the operation mechanism of the piezoelectric sensor to output a stable digital current signal.

[0034] The signal processing unit is used to control the analog-to-digital conversion process through a field-programmable gate array, and to perform digital filtering on the digital current signal to output current data for feature extraction.

[0035] The data transmission unit, based on the control of the field-programmable gate array, transmits the current data to the host computer via a wireless transmission module.

[0036] A third aspect of this application provides an electronic device, comprising: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the above-described method.

[0037] A fourth aspect of this application provides a storage medium, comprising: storing a program or instructions on the storage medium, wherein the program or instructions, when executed by a processor, implement the steps of the above-described method.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] This application embodiment trains a classification model by integrating a self-organizing map algorithm and a clustering algorithm. During the training process, the cluster center points for the clustering algorithm are determined based on the self-organizing map algorithm, which avoids multiple training sessions caused by taking multiple values ​​of the cluster center points during the training of the clustering algorithm. This can quickly obtain better clustering results, improve the training efficiency of the classification model, and facilitate the rapid determination of circuit breaker faults. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0041] Figure 1 A flowchart illustrating a method for determining circuit breaker faults provided in an embodiment of this application;

[0042] Figure 2(a) is a schematic diagram of a portion of the circuit structure of a current acquisition unit for a circuit breaker fault determination method provided in the embodiments of this application;

[0043] Figure 2(b) is a schematic diagram of another part of the circuit structure of the current acquisition unit of a circuit breaker fault determination method provided by an embodiment of this application;

[0044] Figure 3(a) is a schematic diagram of the FPGA pin structure of the signal processing unit of a circuit breaker fault determination method provided in an embodiment of this application;

[0045] Figure 3(b) is a schematic diagram of the peripheral circuit structure of a signal processing unit for a circuit breaker fault determination method provided in an embodiment of this application;

[0046] Figure 4 A schematic diagram of the circuit structure of a data transmission unit for a circuit breaker fault determination method provided in an embodiment of this application;

[0047] Figure 5 This is a flowchart illustrating the filtering process of current data in a circuit breaker fault determination method provided in this application embodiment.

[0048] Figure 6(a) is a waveform diagram of the raw current data;

[0049] Figure 6(b) shows the process. Figure 5 The diagram shows the waveforms of the current data before and after filtering.

[0050] Figure 7 A schematic diagram of the neural network model of the self-organizing map algorithm used in the circuit breaker fault determination method provided in this application embodiment;

[0051] Figure 8 A flowchart illustrating a specific implementation of a method for training a classification model in a circuit breaker fault determination method provided in this application embodiment;

[0052] Figure 9 A schematic diagram showing the result of fault determination using a circuit breaker fault determination method provided in this application embodiment;

[0053] Figure 10 A schematic diagram of the structure of a circuit breaker fault determination system provided in an embodiment of this application;

[0054] Figure 11 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for explanation only and are not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0056] As will be known to those skilled in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0057] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, product, or apparatus.

[0058] Example 1

[0059] Embodiment 1 of this application provides a method for determining circuit breaker faults, which solves the problem of low model training efficiency used in circuit breaker fault determination.

[0060] The subject executing this method can be any computing device capable of implementing the method, such as a server, mobile phone, personal computer, smart wearable device, smart robot, etc.

[0061] Furthermore, the embodiments of this application do not limit the execution order of different steps. When using the method provided in the embodiments of this application, the execution order of different steps can be adjusted according to actual needs.

[0062] To more clearly illustrate the embodiments of this application, the circuit breaker structure is first explained as follows:

[0063] The operating mechanism of a circuit breaker refers to the mechanical transmission device that enables the moving contacts of the circuit breaker to reliably "open", "close" and "maintain state" in a specified manner and speed; in short, it is the mechanical system that drives the opening and closing of the circuit breaker contacts.

[0064] The circuit breaker's opening and closing coils are control switches that trigger the operation of the operating mechanism. By energizing and de-energizing the opening and closing coils, the opening and closing actions of the operating mechanism are controlled.

[0065] For ease of description, the following uses a circuit breaker fault determination device as the execution subject of this method to provide a detailed description of the method provided in this application embodiment.

[0066] like Figure 1 The diagram shown is a flowchart illustrating the specific implementation of a circuit breaker fault determination method provided in this application, including the following steps 11-12:

[0067] Step 11: Obtain current data based on the operation of the circuit breaker's operating mechanism.

[0068] The current data in this embodiment can be obtained through hardware circuit acquisition and software processing. The hardware circuit is used to acquire the action signals of the operating mechanism to form an analog current signal. After amplification, filtering, analog-to-digital conversion, and other processing of the analog current signal, the processed current signal is sent to the software (host computer) for processing through the communication module to obtain current data that can be used for feature extraction.

[0069] The hardware circuit includes a current acquisition unit, a signal processing unit, and a data transmission unit connected in sequence.

[0070] The current acquisition unit collects the action signal of the circuit breaker operating mechanism based on the piezoelectric sensor, and converts the action signal into a weak analog current signal. The analog current signal is then filtered, amplified, and converted from analog to digital to output a digital current signal.

[0071] As shown in Figures 2(a) and 2(b), the current acquisition unit includes a signal acquisition circuit, a signal conditioning circuit, a power supply circuit, and an analog-to-digital conversion circuit.

[0072] The signal acquisition circuit includes a piezoelectric sensor (specific circuit components not shown in the figure), an analog signal input interface J4 (BNC), and peripheral circuitry. The voltage sensor can be installed on the circuit breaker's operating mechanism. Excited by a 5mA current source, it acquires the operating mechanism's action signal (such as pressure or vibration), generating a weak current signal signl1 proportional to this action signal. This weak current signal is then input into the circuit via the BNC interface.

[0073] The signal conditioning circuit includes a filter circuit, a dual-channel differential amplifier circuit, and a differential drive circuit.

[0074] The filtering circuit includes a JFET input operational amplifier U55B and its peripheral circuitry, which is used to perform low-pass filtering and preliminary amplification on the weak input current signal signl1.

[0075] The dual-channel differential amplifier circuit includes differential amplifier U22A (OPA21971DR) and differential amplifier U22B (OPA21971DR) and their peripheral circuits. It is used to further amplify weak current signals, increase the signal amplitude to the range that subsequent circuits can acquire, and use the differential amplification characteristics to suppress common-mode interference of the input signal.

[0076] The differential drive circuit includes a fully differential amplifier U24 (THS4531) and its peripheral circuitry, which receives the output of the dual differential amplifiers and converts the signal into a high-quality differential signal for transmission to the analog-to-digital converter circuit.

[0077] The analog-to-digital conversion circuit includes an analog-to-digital converter U25 (ADS8867IDGSR) for converting received analog signals from a fully differential amplifier into digital signals.

[0078] The signal acquisition circuit also includes a voltage reference circuit, specifically U26 (REF5050AIDR) and its peripheral circuits, which provide a high-precision, low-drift voltage reference source for the analog-to-digital converter U25; in addition, the circuit also includes a zero-drift operational amplifier U27 (OPA333AIDBVR) to further ensure the stability of the output voltage of the voltage reference circuit.

[0079] In addition, the ferrite beads FB44, FB6, FB7, FB8, FB10-FB13, FB45, etc. in the signal acquisition unit are distributed in the power paths of each chip to isolate the power supplies of different functional modules and prevent noise coupling. High-frequency decoupling capacitors C102, C105, C106, C112, C114, C124, C129, C134, C340, C388, C389, C390, etc. are used to filter out high-frequency noise for the chip; mid-to-low frequency decoupling or energy storage capacitors C115 (10μF), C341 (10μF), C391 (4.7μF), C387 (4.7μF), etc. are used to filter out low-frequency ripple and provide current support when the circuit load changes instantaneously (such as when the ADC starts conversion) to prevent the power supply voltage from dropping. The signal acquisition unit also contains a high-speed switching diode D5 to form a clamping protection circuit at the input end, and a transient voltage suppression diode D7 to absorb instantaneous high-voltage pulses such as surges and electrostatic discharges, protecting the entire circuit board of the signal acquisition unit.

[0080] To ensure the reliability of the signal acquisition unit, the electromagnetic interference resistance of the current acquisition unit was tested, and its data acquisition accuracy under strong electromagnetic interference was measured. The results are shown in Table 1. In addition, the fast response capability of the circuit acquisition unit was tested. Given a current signal within 100ms as an analog input, its fast response capability was measured. The results are shown in Table 2.

[0081]

[0082] Table 1

[0083] As can be seen from Table 1, the data acquisition accuracy of the current acquisition unit under strong electromagnetic interference is 99.8%.

[0084]

[0085] Table 2

[0086] As shown in Table 2, when a current signal within 100ms is given as an analog input, the current acquisition unit has a fast response capability and an acquisition accuracy of 99.9%.

[0087] The signal processing unit is used to control the analog-to-digital conversion process through a field-programmable gate array, and to perform digital filtering on the digital current signal to output current data for feature extraction.

[0088] As shown in Figures 3(a) and 3(b), the signal processing unit includes a field-programmable gate array (FPGA) U47 (GW1N-4K-QFN88) and peripheral circuits.

[0089] In the FPGA's SPI interface (SPI0-SPI4), PSI2 is used to communicate with the analog-to-digital converter of the current acquisition unit and read the digital signal output by the analog-to-digital converter; SPI1 is used to communicate with the wireless communication module of the data transmission unit, sending data processed by the FPGA to the data transmission module or sending instructions to the wireless communication module.

[0090] The peripheral circuitry includes a clock circuit, which specifically includes a crystal oscillator OSC1 (25MHz) to provide a reference clock signal for the FPGA.

[0091] In addition, the peripheral circuitry includes power management, filtering, and external interface (connector) circuits. Ferrite beads (FB41, FB42, FB43, QT2012RL600HC, etc.) are connected in series in the power path to isolate the power supplies of different circuit modules; capacitors C260, C262, C263, C264, C265, C267, C268, C269, C270, C271, C272, etc., are distributed near the power pins to filter out high-frequency noise; resistors (RN1, RN2, RN3, RN4, R219, R220, R221, R222, R224, R225, R226, R232, R235, etc.) are used for impedance matching and signal damping; and the JTAG interface is used for downloading FPGA programs, in-circuit debugging, and boundary scan testing.

[0092] To ensure the accuracy of signal processing by the signal processing unit, the unit was tested under strong electromagnetic interference to measure its electromagnetic interference resistance. The test results are shown in Table 3 below. In addition, the time taken from signal input to signal processing system processing to signal preparation for transmission, i.e., response time, was measured, and the results are shown in Table 4.

[0093]

[0094] Table 3

[0095] As shown in Table 3, the signal processing unit achieves a signal processing accuracy of 99.7% under strong electromagnetic interference.

[0096]

[0097] Table 4

[0098] As can be seen from Table 4, the average response time of the signal processing unit is 0.15s.

[0099] The data transmission unit, based on the control of the field-programmable gate array, transmits the current data to the host computer via a wireless transmission module.

[0100] like Figure 4As shown, the data transmission unit includes a wireless communication circuit and its peripheral circuits.

[0101] The wireless communication circuit includes a wireless transceiver module U54 (VG4142S433X0M1), which communicates with the signal processing unit (FPGA) via a standard SPI interface and digital control signals. Its core task is to modulate digital data packets sent by the FPGA onto a radio frequency carrier and transmit them through an antenna; conversely, it can also receive radio frequency signals in the air, demodulate them, and send them back to the FPGA via SPI.

[0102] The peripheral circuitry includes status indicator circuitry and power supply, filtering, and other circuitry to ensure the normal operation of the wireless communication circuitry. The status indicator circuitry includes status indicator lights D23 and D24 (LEDs), which are used to indicate the data reception and transmission status, respectively.

[0103] To ensure the accuracy of the data transmission unit, the data transmission unit was tested under strong electromagnetic interference to measure its electromagnetic interference resistance. The measurement results are shown in Table 5 below. In addition, the data transmission unit was tested for transmission delay, and the results are shown in Table 6.

[0104]

[0105] Table 5

[0106] As can be seen from Table 5, the data transmission accuracy of the data transmission unit is 99.6% under strong electromagnetic interference.

[0107]

[0108] Table 6

[0109] As can be seen from Table 6, the average transmission delay of the data transmission unit is 49.4ms.

[0110] Based on Figure 2 above~ Figure 4 The hardware circuitry processes the analog signal obtained from the circuit breaker's operating mechanism, converting it into a stable digital signal and sending it to the host computer.

[0111] To ensure the stability of the digital signal, this embodiment performs further filtering on the acquired digital signal to form current data for feature extraction.

[0112] like Figure 5 As shown, the filtering process includes steps 51 and 52:

[0113] Step 51: Obtain the target current data, as well as the adjacent current data at two sampling times before and after the target current data.

[0114] If the target current data is represented as at Where t represents the sampling time corresponding to the target current data; then each neighboring current data can be represented as a t-2 a t-1 a t+1 a t+2 .

[0115] Step 52: Sort the target current data and each neighboring current data according to their numerical values, and replace the value of the target current data with the median value in the sorting result.

[0116] Following the procedures in steps 51 and 52, a set of data was used to test the processing time and filtering accuracy of the process. The processing time test structure is shown in Table 7 below. It can be seen that the average processing time is 0.15s.

[0117]

[0118] Table 7

[0119] The results of the filtering accuracy are shown in Table 8 and Figures 6(a) and 6(b). As can be seen from Table 8, the filtering accuracy is 99.5%, which is greater than 99%. As can be seen from Figure 6, the data waveform is smoother after filtering is stopped.

[0120]

[0121] Table 8

[0122] Step 12: Input the current feature vector formed based on the current data into the pre-trained classification model to obtain the classification result representing the fault type of the high-voltage circuit breaker output by the classification model;

[0123] The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centers through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centers to obtain the clustering results.

[0124] The clustering algorithm used in this embodiment is the K-means clustering algorithm.

[0125] K-means is a distance-based iterative algorithm that classifies N data samples into k predefined clusters, such that each sample is closer to the centroid of its own cluster than to the centroids of any other cluster. The K-means algorithm uses Euclidean distance as the similarity comparison standard and the sum of squared errors as the clustering criterion function. Its basic calculation process is as follows:

[0126] 1) Initialization: Set the number of clusters k, and randomly select k samples from the sample data as the initial cluster centers.

[0127] 2) Class division: Calculate the distance between N sample data and k initial cluster centers according to formula (1), and assign the N sample data to the nearest center point according to the distance to form k clusters.

[0128] ; (1);

[0129] In the formula, n is the dimension of x.

[0130] 3) Cluster center point solution: Calculate the average value of all objects in each class, and use this as the new cluster center for k clusters.

[0131] 4) Convergence judgment: Use equation (2) to judge whether the clustering is converged.

[0132] ; (2);

[0133] In the formula: C i For the i-th cluster, m i C i Cluster centers; Z q C i The samples in.

[0134] Repeat steps 2) to 4) until E reaches the limiting condition or the fluctuation is very small, indicating that the algorithm tends to be stable and the clustering ends.

[0135] The self-organizing map (SOM) algorithm is a clustering algorithm that automatically classifies input data in an unsupervised manner, learns self-organizing patterns from input data, repeatedly adjusts connection weights, and finally represents the classification results in a competitive layer.

[0136] like Figure 7 As shown, the SOM network can be divided into one-dimensional or two-dimensional topologies, capable of generating different responses to different input signals, and achieving clustering of neurons with the same characteristics. During neuron training, in addition to adjusting the weights of the winning neurons, the weights of adjacent neurons also need to be adjusted accordingly, with the adjustment range decreasing from near to far. The nearest neurons stimulate each other, while the more distant neurons inhibit each other, thus achieving the goal of clustering.

[0137] The competitive learning of SOM neural networks mainly consists of the following four steps:

[0138] 1) Initialize the network. Set the network connection weights {w} ij Assign a random value within the interval [0,1]. Determine the learning rate. initial value ( ); Define the field initial value Determine the total number of learning iterations T. N .

[0139] 2) Find the winning neuron. For each input data, find the node that best matches it. The calculation formula is shown in (3).

[0140] ; (3);

[0141] In the formula, The formula (3) represents the value of the input data and indicates the calculation of the input data. With each weight The distance is determined relative to the input data. The nearest weight.

[0142] 3) Adjust the connection weights of the output nodes. Adjust the connection weight vector of the neuron using the weight update formula:

[0143] ;(4);

[0144] ;(5);

[0145] ;(6);

[0146] Among them, learning efficiency It decreases monotonically with time t, ensuring the convergence of the learning process. It is the domain function surrounding the winning neuron. This represents the distance between node j and the winning neuron i(x). The neuron corresponding to the weight closest to the input data is the winning neuron.

[0147] 4) Increase the number of learning steps. Let t = t + 1, and repeat steps 2) and 3) until the learning efficiency is achieved. The clustering results of the input vectors are obtained when the number of training iterations approaches zero or reaches the set total number of training iterations TN.

[0148] During training, the classification model in this embodiment sets the range of the number of neurons k for the self-organizing map algorithm to be [a, b]. For each number of neurons within the range, a pre-training process for the hybrid model is performed to obtain the corresponding candidate classification model. Based on preset evaluation criteria, a classification model is selected from the candidate classification models as the classification model used for inference.

[0149] For a given number of neurons k, the training process can involve inputting training samples into a self-organizing map algorithm to obtain the weight vector of each neuron output by the self-organizing map algorithm; wherein the value of the weight vector is determined based on the similarity between the training sample and the neuron; using the weight vector as the center point of a clustering algorithm, inputting the training samples into a clustering model to obtain the clustering result of the training samples output by the clustering model.

[0150] Based on preset evaluation criteria, a classification model is selected from the candidate classification models as the classification model used for inference. This includes: determining the similarity degree corresponding to each pair of different clusters in the clustering results based on the distance between the centroids of each pair of different clusters and the dispersion of data points in the same cluster; determining the metric value of the clustering results based on the similarity degree; selecting the minimum value from the metric values ​​corresponding to each clustering result, and using the classification model corresponding to the minimum value as the classification model used for inference.

[0151] Based on the aforementioned similarity levels, the metric value of the clustering result is determined, specifically including: for each class in the clustering result, selecting the maximum similarity value among the similarity values ​​of that class and other classes; and taking the mean of the maximum similarity values ​​corresponding to each class in the clustering result as the metric value of the clustering result.

[0152] like Figure 8 As shown, the specific process of training the classification model includes:

[0153] 1) Determine the range [a, b] for the number of clusters (number of neurons) k.

[0154] 2) Use the SOM neural network to cluster the input samples, set the number of output neurons to k, and obtain the weight vector of each output neuron by performing similarity calculation (such as distance calculation) with the training samples, which is the center point of the cluster.

[0155] 3) Use the center point from 2) to initialize the K-means algorithm, and then use the K-means algorithm to divide the data samples into k clusters.

[0156] 4) Calculate the DBI (Davies-Bouldin) index of the K-means clustering results.

[0157] 5) Let k = k + 1.

[0158] 6) Repeat steps 2) to 5) until all k values ​​within the cluster number range [a,b] determined in step 1) have been clustered.

[0159] 7) Compare the obtained DBI indices, and the cluster corresponding to the smallest DBI index is the best cluster.

[0160] The DBI index is a metric used to evaluate clustering algorithms, assessing clustering effectiveness through inter-cluster distance and intra-cluster distance. The DBI index can be calculated as follows:

[0161] 1) Dispersion Si: This is an index that measures the degree of dispersion of data points in the i-th cluster.

[0162] ;(7);

[0163] X j A represents the j-th data point in the i-th class; i T represents the center point of the i-th class; i The i-th class represents the number of data points; q represents the distance calculation method, where q=1 represents the mean distance from each point to the center, and q=2 represents the standard deviation of the distance from each point to the center. Both are used to measure the degree of dispersion. In this embodiment, q=1.

[0164] 2) Distance value M ij : Represents the distance between the center points of the i-th class and the j-th class.

[0165] ;(8);

[0166] a ki This represents the value of the k-th attribute of the center point of the i-th class.

[0167] 3) Similarity R ij : Measures the similarity between class i and class j.

[0168] ;(9);

[0169] 4) From R ij Select the maximum value R from the middle i =max(R ij The mean of the maximum similarity scores between the i-th class and other classes is calculated. This refers to the DBI index.

[0170] ;(10);

[0171] Different numbers of categories k will lead to different results value, The smaller the value, the better the classification effect.

[0172] The model is trained through the above training process to obtain a classification model that can be used for reasoning. In practical reasoning applications, the current feature vector formed based on circuit data is input into the classification model to obtain the classification result representing the fault type of the circuit breaker.

[0173] To verify the fault determination time and accuracy of the classification model, the fault determination time was tested using test data, and the test results are shown in Table 9. In addition, the fault determination accuracy was tested 20 times using 200 sets of data in Table 10, and the test results are shown in Table 11.

[0174]

[0175] Table 9

[0176] As can be seen from Table 9, the average fault determination time is 0.59s, which is less than 1s.

[0177]

[0178] Table 10

[0179] Table 10 shows the characteristic data of the closing coil current of the operating mechanism and the status of the circuit breaker operating mechanism.

[0180]

[0181] Table 11

[0182] As shown in Table 11, the average accuracy rate of fault diagnosis was 98.7%, which is greater than 95%. Furthermore, from... Figure 9 From this, we can conclude that there were three data classification errors in the 20th fault assessment result, meaning the fault assessment accuracy was (200-3) / 200=98.5%.

[0183] This application embodiment trains a classification model by integrating a self-organizing map algorithm and a clustering algorithm. During the training process, the cluster center points for the clustering algorithm are determined based on the self-organizing map algorithm, which avoids multiple training sessions caused by taking multiple values ​​of the cluster center points during the training of the clustering algorithm. This can quickly obtain better clustering results, improve the training efficiency of the classification model, and facilitate the rapid determination of circuit breaker faults.

[0184] Example 2

[0185] To address the problem of low model training efficiency in the prior art for circuit breaker fault determination, and based on the same inventive concept as Embodiment 1, this application also provides a circuit breaker fault determination system.

[0186] The specific structural diagram of the system is as follows: Figure 10As shown, it includes the following functional units 1001~1002:

[0187] The data acquisition unit 1001 is used to acquire current data generated by the operation of the circuit breaker's operating mechanism.

[0188] The data acquisition unit 1001 includes a current acquisition unit, a signal processing unit, and a data transmission unit connected in sequence.

[0189] The current acquisition unit is used to filter, amplify, and perform analog-to-digital conversion on the current signal generated by the piezoelectric sensor's operating mechanism to output a stable digital current signal. The signal processing unit is used to control the analog-to-digital conversion process through a field-programmable gate array (FPGA) and to perform digital filtering on the digital current signal to output current data for feature extraction. The data transmission unit, based on the control of the FPGA, transmits the current data to the host computer via a wireless transmission module.

[0190] The data acquisition unit 1001 further includes a data processing unit, which is used to filter the acquired current data and form a current feature vector based on the filtered current data. The filtering of the acquired current data specifically includes: acquiring target current data and neighboring current data at two sampling times before and after the target current data; sorting the target current data and neighboring current data according to their numerical values, and replacing the value of the target current data with the median value in the sorting result.

[0191] The fault determination unit 1001 is used to input the current feature vector formed based on the current data into a pre-trained classification model to obtain the classification result output by the classification model that characterizes the fault type of the high-voltage circuit breaker.

[0192] The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centroids through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centroids to obtain the clustering results.

[0193] The circuit breaker fault determination system also includes a model training unit. This unit trains a hybrid model incorporating self-organizing map and clustering algorithms to obtain a classification model suitable for reasoning. Specifically, it is used for:

[0194] A range of values ​​for the number of neurons is set for the self-organizing map algorithm. For each number of neurons within the range, a pre-training process is performed for the hybrid model to obtain the corresponding candidate classification model. Based on preset evaluation criteria, a classification model is selected from the candidate classification models as the classification model used for inference.

[0195] The model training unit is also used to perform the following pre-training process for a number of neurons within the range of values:

[0196] The training samples are input into a self-organizing map algorithm to obtain the weight vector of each neuron output by the self-organizing map algorithm; wherein the value of the weight vector is determined based on the similarity between the training samples and the neurons; the weight vector is used as the center point of a clustering algorithm, and the training samples are input into a clustering model to obtain the clustering result of the training samples output by the clustering model.

[0197] The step of selecting a classification model from the candidate classification models based on preset evaluation criteria as the classification model for inference specifically includes: determining the similarity degree corresponding to each pair of different clusters in the clustering results based on the distance between the centroids of each pair of different clusters in the clustering results and the dispersion of data points in the same cluster; determining the metric value of the clustering results based on the similarity degree; selecting the minimum value from the metric values ​​corresponding to each clustering result, and using the classification model corresponding to the minimum value as the classification model for inference.

[0198] Based on the aforementioned similarity levels, the metric value of the clustering result is determined, specifically including: for each class in the clustering result, selecting the maximum similarity value among the similarity values ​​of that class and other classes; and taking the mean of the maximum similarity values ​​corresponding to each class in the clustering result as the metric value of the clustering result.

[0199] This application embodiment trains a classification model by integrating a self-organizing map algorithm and a clustering algorithm. During the training process, the cluster center points for the clustering algorithm are determined based on the self-organizing map algorithm, which avoids multiple training sessions caused by taking multiple values ​​of the cluster center points during the training of the clustering algorithm. This can quickly obtain better clustering results, improve the training efficiency of the classification model, and facilitate the rapid determination of circuit breaker faults.

[0200] Based on the same inventive concept as the foregoing embodiments of this application, this application also provides a computing device.

[0201] like Figure 11As shown, the computing device includes a memory 1101 and a processor 1102. The memory 1101 can be configured to store various other data to support operation on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device. The memory 1101 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0202] Processor 1102, coupled to memory 1101, is used to execute a computer program stored in memory 1101 for performing a method for aggregating feasible domains of a virtual power plant as described in the foregoing embodiments.

[0203] When the processor 1102 executes the computer program to perform the circuit breaker fault determination method, it trains the classification model by integrating the self-organizing map algorithm and the clustering algorithm. During the training process, the cluster center points used for the clustering algorithm are determined based on the self-organizing map algorithm, which avoids the multiple training caused by the multiple values ​​of the cluster center points during the training of the clustering algorithm. This can quickly obtain better clustering results, improve the training efficiency of the classification model, and facilitate the rapid determination of circuit breaker faults.

[0204] When the processor 1102 executes the computer program in the memory 1101, in addition to the functions described above, it can also perform other functions, as detailed in the descriptions of the preceding embodiments.

[0205] Furthermore, such as Figure 11 As shown, the computing device also includes other components such as a display 1104, a communication component 1103, a power supply component 1105, and an audio component 1106. Figure 11 The diagram only shows some components and does not mean that the computing device includes only these components. Figure 11 The components shown.

[0206] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the methods provided in the above embodiments.

[0207] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0208] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0209] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining circuit breaker faults, characterized in that, The method includes: Acquire current data based on the actions of the circuit breaker's operating mechanism; The current feature vector formed based on the current data is input into a pre-trained classification model to obtain the classification result representing the fault type of the circuit breaker output by the classification model. The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centroids through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centroids to obtain the clustering results.

2. The method according to claim 1, characterized in that, The method further includes: the classification model is trained in the following manner: A range of values ​​for the number of neurons is set for the self-organizing map algorithm. For each number of neurons within the range, a pre-training process for the hybrid model is performed to obtain the corresponding candidate classification model. Based on preset evaluation criteria, a classification model is selected from the candidate classification models as the classification model used for inference.

3. The method according to claim 2, characterized in that, For a given number of neurons within the specified value range, the following pre-training process is performed: The training samples are input into the self-organizing map algorithm to obtain the weight vector of each neuron output by the self-organizing map algorithm; wherein the value of the weight vector is determined based on the similarity between the training samples and the neurons; The weight vector is used as the center point of the clustering algorithm. The training samples are input into the clustering model to obtain the clustering results of the training samples output by the clustering model.

4. The method according to claim 2, characterized in that, The selection of a classification model from the candidate classification models based on preset evaluation criteria as the classification model used for inference specifically includes: Based on the distance between the centroids of two different clusters in the clustering results, and the dispersion of data points in the same cluster, the similarity levels corresponding to two different clusters in the clustering results are determined. Based on the aforementioned similarity levels, the metric values ​​for the clustering results are determined; Select the minimum value from the various metrics values ​​corresponding to each clustering result, and use the classification model corresponding to the minimum value as the classification model for inference.

5. The method according to claim 4, characterized in that, Based on the aforementioned similarity levels, the metric values ​​for the clustering results are determined, specifically including: For each cluster in the clustering results, select the maximum similarity value between that cluster and other clusters; The mean of the maximum similarity values ​​corresponding to each class in the clustering results is used as the metric value of the clustering results.

6. The method according to claim 1, characterized in that, The method further includes: filtering the acquired current data and forming a current feature vector based on the filtered current data; The filtering process for the acquired current data specifically includes: Acquire the target current data, as well as the adjacent current data at two sampling times before and after the target current data; The target current data and each neighboring current data are sorted according to their numerical values, and the value of the target current data is replaced with the median value in the sorting result.

7. A circuit breaker fault determination system, characterized in that, The system includes: The data acquisition unit is used to acquire current data based on the operation of the circuit breaker's operating mechanism. The fault determination unit is used to input the current feature vector formed based on the current data into a pre-trained classification model to obtain the classification result of the classification model that characterizes the fault type of the circuit breaker. The classification model is a hybrid model that includes a self-organizing map algorithm and a clustering algorithm. During the training process, the classification model determines the cluster centroids through the self-organizing map algorithm, and the clustering algorithm performs cluster analysis based on the cluster centroids to obtain the clustering results.

8. The system according to claim 7, characterized in that, The data acquisition unit includes a current acquisition unit, a signal processing unit, and a data transmission unit connected in sequence. The current acquisition unit is used to filter, amplify, and perform analog-to-digital conversion on the current signal generated by the operation mechanism of the piezoelectric sensor to output a stable digital current signal. The signal processing unit is used to control the analog-to-digital conversion process through a field-programmable gate array, and to perform digital filtering on the digital current signal to output current data for feature extraction. The data transmission unit, based on the control of the field-programmable gate array, transmits the current data to the host computer via a wireless transmission module.

9. An electronic device, characterized in that, include: It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1-6.

10. A storage medium, characterized in that, include: The storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1-6.

Images