Fault detection method, device, equipment, storage medium and program product

Abstract

The application relates to a fault detection method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring a source current signal of a solid-state circuit breaker; performing feature extraction on the source current signal to determine a source current change rate signal corresponding to the source current signal; inputting the source current change rate signal into a pre-trained target support vector machine (SVM) to determine a fault detection result according to the output of the target SVM, and the target penalty parameter and the target kernel parameter in the target SVM are determined based on an optimization algorithm. The method can improve the reliability of fault detection.

Description

Technical Field

[0001] This application relates to the field of power safety technology, and in particular to a fault detection method, apparatus, equipment, storage medium, and program product. Background Technology

[0002] Protection technology is crucial for the reliable operation of DC distribution networks. When a short-circuit fault occurs in a DC distribution network, the system impedance is low, the fault current rises rapidly, and the fault current does not have a natural zero-crossing point. Typically, power devices in the converter can burn out within milliseconds. This necessitates that DC circuit breakers possess rapid arc-extinguishing and fault detection capabilities. High-voltage DC solid-state circuit breakers are widely used due to their small size, long lifespan, fast breaking speed, and arc-free operation. Their main function is to connect steady-state current without generating additional losses during normal operation; and to promptly disconnect the circuit when an overcurrent or short-circuit fault occurs, thus protecting equipment, lines, and personnel safety.

[0003] The reliability of existing methods for detecting system faults needs to be improved. Summary of the Invention

[0004] Therefore, it is necessary to provide a fault detection method, apparatus, equipment, storage medium, and program product that can improve the accuracy of fault detection in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a fault detection method, including:

[0006] Obtain the source current signal of the solid-state circuit breaker;

[0007] Feature extraction is performed on the source current signal to determine the source current change rate signal corresponding to the source current signal;

[0008] The source current change rate signal is input into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm.

[0009] In one embodiment, the target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. The source current change rate signal is input into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM, including:

[0010] The source current change rate signal is input into the short-circuit detection SVM to determine the short-circuit detection result;

[0011] If the short circuit detection result is no short circuit, the source current change rate signal is input into the overload detection SVM to determine the overload detection result;

[0012] If the overload detection result is no overload, the source current change rate signal is input into the oscillation detection SVM to determine the oscillation detection result.

[0013] In one embodiment, feature extraction is performed on the source current signal to determine the source current change rate signal corresponding to the source current signal, including:

[0014] The source current signal is preprocessed, including normalization and smoothing filtering.

[0015] The source current rate of change signal corresponding to the preprocessed source current signal is determined by performing first-order differential processing on the preprocessed source current signal using the Savitzky-Golay algorithm.

[0016] In one embodiment, the training process of the target support vector machine (SVM) includes:

[0017] Obtain the training dataset, which includes multiple training input data and fault labels corresponding to each training input data. Each training input data is a sample source current change rate signal.

[0018] Based on the training dataset and optimization algorithm, determine the target penalty parameters and target kernel parameters of the initial SVM to obtain the target SVM.

[0019] In one embodiment, the optimization algorithm is an improved particle swarm optimization algorithm, which determines the target penalty parameters and target kernel parameters of the initial SVM based on the training dataset and the optimization algorithm, including:

[0020] A particle swarm is randomly generated within a preset parameter range. The particle swarm includes the positions of multiple particles, and the position of a particle represents a set of penalty parameters and kernel parameters.

[0021] The position update process is executed iteratively multiple times until the iteration deadline is met. The target penalty parameter and target kernel parameter are determined based on the results of the iterative execution.

[0022] The i-th position update process includes:

[0023] An initial SVM is trained based on the position of each particle and the training dataset. The fitness of each particle is calculated according to the fitness function. The historical best position of each particle and the global best position of the particle swarm are determined based on the fitness of each particle. The fitness function includes classification accuracy or precision.

[0024] The positions of each particle are updated based on its historical best position and the global best position of the particle swarm.

[0025] In one embodiment, the position of each particle is updated based on its historical best position and the global best position of the particle swarm, including:

[0026] For each particle, the particle's flight velocity is updated based on the particle's historical best position and the global best position of the particle swarm.

[0027] The particle's position is updated based on its flight speed.

[0028] Secondly, this application also provides a fault detection device, comprising:

[0029] The acquisition module is used to acquire the source current signal of the solid-state circuit breaker;

[0030] The feature extraction module is used to extract features from the source current signal and determine the source current change rate signal corresponding to the source current signal.

[0031] The detection module is used to input the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm.

[0032] In one embodiment, the target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. The detection module is specifically used to input the source current change rate signal into the short-circuit detection SVM to determine the short-circuit detection result; if the short-circuit detection result is no short circuit, the source current change rate signal is input into the overload detection SVM to determine the overload detection result; if the overload detection result is no overload, the source current change rate signal is input into the oscillation detection SVM to determine the oscillation detection result.

[0033] In one embodiment, the feature extraction module is specifically used to preprocess the source current signal, including normalization and smoothing filtering; and to perform first-order differentiation processing on the preprocessed source current signal according to the Savitzky-Golay algorithm to determine the source current rate of change signal corresponding to the source current signal.

[0034] In one embodiment, the fault detection device further includes a training module for acquiring a training dataset, wherein the training dataset includes multiple training input data and fault labels corresponding to each training input data, and each training input data is a sample source current change rate signal; the target penalty parameters and target kernel parameters of the initial SVM are determined according to the training dataset and the optimization algorithm to obtain the target SVM.

[0035] In one embodiment, the optimization algorithm is an improved particle swarm optimization algorithm. The training module is specifically used to randomly generate a particle swarm within a preset parameter range. The particle swarm includes the positions of multiple particles, and each particle's position represents a set of penalty parameters and kernel parameters. The module iteratively executes multiple position update processes until an iteration cutoff condition is met. Based on the results of the iterations, the target penalty parameters and target kernel parameters are determined. The i-th position update process includes:

[0036] An initial SVM is trained based on the position of each particle and the training dataset. The fitness of each particle is calculated according to the fitness function. The historical best position of each particle and the global best position of the particle swarm are determined based on the fitness of each particle. The fitness function includes classification accuracy or precision.

[0037] The positions of each particle are updated based on its historical best position and the global best position of the particle swarm.

[0038] In one embodiment, the training module is specifically used to update the flight velocity of each particle based on the particle's historical best position and the global best position of the particle swarm; and to update the particle's position based on the particle's flight velocity.

[0039] Thirdly, this application also provides a solid-state circuit breaker, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects above.

[0041] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the first aspects above.

[0042] The aforementioned fault detection method, apparatus, equipment, storage medium, and program products acquire the source current signal of a solid-state circuit breaker; extract features from the source current signal to determine the corresponding source current change rate signal; and input the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm. Thus, since the target penalty parameter and target kernel parameter of the target SVM are determined by an optimization algorithm, it is guaranteed that the trained target SVM maximizes the classification margin. The target SVM can distinguish different fault types based on the optimal hyperplane, thereby improving the reliability of fault detection. Furthermore, the lightweight SVM does not require high-dimensional transformation, enabling microsecond-level rapid fault detection. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the circuit structure for the desaturation detection method;

[0045] Figure 2 This is a schematic diagram of the circuit structure for the source inductance detection method;

[0046] Figure 3 This is a schematic diagram of the circuit structure for the current sensor detection method;

[0047] Figure 4 This is a schematic diagram of the circuit structure for the gate voltage detection method.

[0048] Figure 5 This is a diagram illustrating the application environment of a fault detection method in one embodiment;

[0049] Figure 6 This is a schematic diagram of the circuit structure in one embodiment;

[0050] Figure 7 This is a flowchart illustrating a fault detection method in one embodiment;

[0051] Figure 8 This is a flowchart illustrating the steps for determining the source current rate of change signal in one embodiment;

[0052] Figure 9 This is a schematic diagram of the SVM hyperplane in one embodiment;

[0053] Figure 10 This is a flowchart illustrating the steps for determining the fault detection result in one embodiment;

[0054] Figure 11 This is a flowchart illustrating the training steps of the target SVM in one embodiment;

[0055] Figure 12 This is a flowchart illustrating the steps for determining the target penalty parameters and target kernel parameters in one embodiment.

[0056] Figure 13 This is a flowchart illustrating the steps for updating the position of each particle in one embodiment.

[0057] Figure 14 This is a structural block diagram of a fault detection device in one embodiment;

[0058] Figure 15 This is a diagram of the internal structure of a solid-state circuit breaker in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0061] Driven by the demand for clean energy and environmental protection, distributed power sources such as photovoltaic power generation, wind power generation, fuel cells, and energy storage devices have developed rapidly. Distributed power sources mainly produce electricity in DC mode, with small single-unit capacity, requiring DC-AC inverter stages to connect to the AC distribution network. Due to the inherent uncertainties of new energy sources, traditional AC distribution networks can no longer meet the future requirements of power distribution technology. Compared to AC distribution networks, DC distribution networks offer superior performance. DC distribution networks can directly absorb distributed power sources without reactive power compensation, and offer higher reliability and economy. With the upgrading of power devices, DC power electronics technology is undergoing further transformation, further highlighting the advantages of DC distribution. Protection technology for DC distribution networks is crucial for their reliable operation, but currently, it largely remains at the theoretical research stage, becoming a bottleneck restricting the development of DC distribution networks. When a short-circuit fault occurs in a DC distribution network, the system impedance is low, the fault current rises rapidly, and the fault current does not naturally cross zero. Typically, within milliseconds, the power devices of the converter will burn out, requiring DC circuit breakers to have rapid arc-extinguishing and fault detection capabilities.

[0062] High-voltage DC solid-state circuit breakers are widely used due to their small size, long lifespan, fast breaking speed, and arc-free operation. Their main function is to promptly disconnect the circuit when an overcurrent or short-circuit fault occurs in the system. The fault inductance of a DC distribution network varies widely, resulting in a large range of changes in the fault current rise rate di / dt. Therefore, when detecting overcurrent or short-circuit faults in the system, the fault detection circuit should be able to operate reliably over a wide di / dt range, while the losses in the fault detection circuit should be minimized to improve the efficiency of the solid-state circuit breaker. Currently, there are four main fault detection methods: the desaturation detection method based on Vds(on), whose circuit structure is as follows... Figure 1 As shown; the source inductance detection method, its fault detection circuit structure is as follows. Figure 2 As shown; the fault detection circuit structure of the current sensor detection method is as follows. Figure 3 As shown; the gate voltage detection method, its fault detection circuit structure is as follows. Figure 4 As shown.

[0063] All of the above-mentioned fault detection methods are susceptible to changes in the rate of rise of the fault current, and their reliability needs to be improved. Furthermore, the response speed of each of these methods is relatively low, failing to meet the requirements for rapid fault current detection.

[0064] In view of this, embodiments of this application provide a fault detection method that can improve the accuracy of fault detection. The fault detection method provided in embodiments of this application can be applied to, for example... Figure 5The solid-state circuit breaker shown is mainly composed of a main switch 101, a control unit 102, an acquisition unit 103, and a power supply module 104. The control unit 102 acquires the source current signal of the solid-state circuit breaker; performs feature extraction on the source current signal to determine the corresponding source current change rate signal; inputs the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm. The state of the solid-state circuit breaker can be judged based on the fault detection result to determine whether the circuit breaker should disconnect the circuit. The power supply module 104 draws power from the DC input ports DC+ and DC-, providing a stable operating power supply for the acquisition unit 103 and the control unit 102.

[0065] For example, the main switch can be a power device MOSFET, the control unit can be a DSP, and the acquisition unit can be a combination of a resistor and an isolation operational amplifier, which can be used to acquire current signals.

[0066] refer to Figure 6 This is a schematic diagram of the circuit structure of a fault detection circuit provided in an embodiment of this application, which drives a solid-state circuit breaker through the fault detection result output by the target SVM.

[0067] In one exemplary embodiment, such as Figure 7 As shown, a fault detection method is provided, which can be applied to... Figure 5 Taking the detection unit in the diagram as an example, the explanation includes the following steps 201 to 203. Wherein:

[0068] Step 201: Obtain the source current signal of the solid-state circuit breaker.

[0069] The solid-state circuit breaker includes a data acquisition unit that can acquire high-sampling-rate source current signals through voltage divider resistors and current sensors on the solid-state circuit breaker.

[0070] Optionally, the source current signal can be a current sampling signal drawn from the source of the solid-state circuit breaker to reflect the actual magnitude and changes of the current flowing through the main switch.

[0071] Step 202: Extract features from the source current signal to determine the source current change rate signal corresponding to the source current signal.

[0072] Among them, the source current change rate signal is an electrical signal that reflects how fast the source current of the main switching device of the solid-state circuit breaker changes over time. Based on this source current change rate signal, it can be used to determine whether there is a fault in the DC power distribution system where the solid-state circuit breaker is located.

[0073] Understandably, when a DC power distribution system is operating normally, the current changes gradually and the rate of change of the source current is very small. However, once a short circuit occurs, the fault current will rise sharply in a very short time, and the rate of change of the source current will also increase instantaneously.

[0074] Step 203: Input the source current change rate signal into the pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on the optimization algorithm.

[0075] Optionally, a pre-trained target SVM is deployed in the solid-state circuit breaker for real-time current diagnosis to determine the fault detection result. Specifically, the source current change rate signal can be input into the target SVM to obtain the fault detection result.

[0076] The fault detection result includes whether a fault exists or not. Alternatively, depending on the type of fault detection, the output fault detection result may include: short circuit, no short circuit, overload, no overload, oscillation, no oscillation, etc. This application embodiment does not limit the type of fault detection result.

[0077] Alternatively, when deploying the target support vector machine in a solid-state circuit breaker, it is understood that it is deployed in the control unit of the solid-state circuit breaker.

[0078] For example, when deploying a target support vector machine in a solid-state circuit breaker, support vectors, dual coefficients, intercepts, kernel function parameters, and standardized parameters can be extracted from the trained target SVM. Support vectors with minimal contribution to decision-making (corresponding coefficients close to 0) are removed to compress the model size and accelerate online inference. All extracted parameters are then embedded into a header or source file using C language format. The support vector array is declared as a `const float` type and stored in Flash memory. Corresponding C functions are then used to implement feature standardization, kernel function calculation, and decision functions. This code is compiled and run on a DSP to complete the deployment of the target SVM.

[0079] Optionally, the target SVM can be trained by the control unit of the solid-state circuit breaker itself; alternatively, in order to save the computing resources of the control unit in the solid-state circuit breaker, the target SVM can also be trained by the terminal or cloud server and sent to the control unit of the solid-state circuit breaker after training.

[0080] Optionally, the choice of kernel function has a significant impact on the performance of the SVM model. In this embodiment, the kernel function can be a radial basis function (RBF) kernel, which adopts the form of a Gaussian radial basis function. This kernel can map linearly inseparable current feature samples in a low-dimensional space to a high-dimensional feature space, enabling nonlinear classification and identification of different operating conditions and fault types. The RBF kernel can be expressed as:

[0081]

[0082] in, It's a kernel function. It is a kernel parameter, x i and x j These are the feature vectors of two samples.

[0083] Multiple initial SVM models can be trained first, with each initial SVM corresponding to a different combination of kernel parameters and penalty parameters. Then, the kernel parameters and penalty parameters are continuously optimized using an optimization algorithm until the trained SVM model has the best performance. At this point, the SVM with the best performance can be used as the target SVM, with its corresponding kernel parameters as the target kernel parameters and its corresponding penalty parameters as the target penalty parameters.

[0084] The aforementioned fault detection method acquires the source current signal of the solid-state circuit breaker; extracts features from the source current signal to determine the corresponding source current change rate signal; and inputs the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm. Thus, since the target penalty parameter and target kernel parameter of the target SVM are determined by the optimization algorithm, it ensures that the trained target SVM maximizes the classification margin. The target SVM can distinguish different fault types based on the optimal hyperplane, thereby improving the reliability of fault detection. Furthermore, the lightweight SVM does not require high-dimensional transformation, enabling microsecond-level rapid fault detection.

[0085] In one exemplary embodiment, such as Figure 8 As shown, feature extraction is performed on the source current signal to determine the corresponding source current change rate signal, including steps 301 to 302. Wherein:

[0086] Step 301: Preprocess the source current signal, including normalization and smoothing filtering.

[0087] Optionally, the source current signal can be normalized so that the current at each moment is within the value range of [0,1]. Taking the min-max normalization method as an example:

[0088]

[0089] Where x represents the original current collected, x min and x max These are the maximum and minimum values ​​of the current, x. n It is the normalized current.

[0090] Optionally, smoothing the acquired source current signal can eliminate the influence of outliers or noise.

[0091] For example, the source current signal can be filtered using an SG filter, wherein the SG filter employs the Savitzky-Golay algorithm.

[0092] Specifically, when the SG filter has a window length of 5 and a polynomial order of 2, the data window can be formed by the current sampling point and two points before and after it. The quadratic polynomial y(t)=a0+a1t+a2t is fitted by the least squares method. 2 The fitted value y(t) (i.e. a0) of the polynomial at the center point t=0 is used as the filter output to achieve smooth denoising of the current sampling signal.

[0093] Where a0, a1, and a2 are polynomial parameters, each taking values ​​in the range [0,1].

[0094] Step 302: Perform first-order differential processing on the preprocessed source current signal according to the Savitzky-Golay algorithm to determine the source current rate of change signal corresponding to the source current signal.

[0095] Optionally, based on the above introduction to the SG filter, the SG filter can perform polynomial fitting on the data within each sliding window. After fitting the polynomial, not only can the value at the center point be taken as the smoothed output, but the first derivative of the polynomial at the center point can also be taken as the corresponding source current change rate.

[0096] For example, for the acquired source current signal I ds [k] can be processed by traversing a sliding window of length 5. Within each window, 5 consecutive data points are taken, denoted as (t). l I ds [l]), where l = -2, -1, 0, 1, 2, t l Given an equally spaced integer sequence {-2,-1,0,1,2}, with the window center index l=0.

[0097] For the 5 data points within the window, the binomial expression y = a0 + a1t + a2t is used. 2Perform a fitting operation to minimize the fitting error. Based on the least squares principle, construct the following error function and solve for its minimum value:

[0098]

[0099] By solving the above optimization problem, we can obtain the fitting coefficients a0, a1, and a2 corresponding to the window, where a0 is the smoothed signal value at the center of the window, and a1 is the first derivative of the polynomial at the center, i.e., the estimated value of the source current change rate.

[0100] Optionally, to avoid repeatedly solving the least squares problem for each window and improve computational efficiency, a dedicated convolution kernel for calculating the first derivative is obtained in advance by solving the least squares problem in the form of a Vandermonde matrix, when the window length is 5 and the polynomial order is 2. Based on this convolution kernel, the calculation of the source current change rate can be transformed into a convolution operation, as shown in the following formula:

[0101]

[0102] Optionally, refer to Figure 9 The horizontal axis represents the source current amplitude, and the vertical axis represents the rate of change of current. Green squares represent normal load and overload current samples, and yellow dots represent short-circuit fault current samples. Combined with... Figure 9 It is known that current amplitude (horizontal axis) alone cannot distinguish between overload and fault. However, by combining the current change rate (vertical axis) feature, SVM can find the optimal hyperplane, maximize the classification interval between the two types of samples, and achieve accurate classification of normal and fault currents.

[0103] The above-mentioned source current signal is preprocessed, including normalization and smoothing filtering. The preprocessed source current signal is then subjected to first-order differentiation according to the Savitzky-Golay algorithm to determine the source current rate of change signal. The source current rate of change signal is then used as the input signal of the target SVM. By identifying the rate of change of current, fault detection in the power system can be performed more accurately and reliably.

[0104] The target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. When determining the fault detection result based on the output of the target SVM, the source current change rate signal can be sequentially input into the short-circuit detection SVM, the overload detection SVM, and the oscillation detection SVM to obtain the fault detection result. In an exemplary embodiment, such as... Figure 10 As shown, the source current change rate signal is input into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM, including steps 401 to 403. Wherein:

[0105] Step 401: Input the source current change rate signal into the short-circuit detection SVM to determine the short-circuit detection result.

[0106] Optionally, a short-circuit detection SVM can be used to detect the source current change rate signal to determine whether a short circuit has occurred.

[0107] The short-circuit detection SVM can be a binary classifier, whose short-circuit detection result is either short-circuit or not short-circuit. When the short-circuit detection result is short-circuit, a first indication message is output, which can be used to trigger hardware protection / shutdown. For example, the first indication message can contain the code [1,1]. When the short-circuit detection result is not short-circuit, the source current change rate signal is input into the overload detection SVM.

[0108] Step 402: If the short circuit detection result is no short circuit, input the source current change rate signal into the overload detection SVM to determine the overload detection result.

[0109] Optionally, an overload detection SVM can be used to detect the source current change rate signal to determine whether an overload has occurred.

[0110] Similarly, the overload detection SVM can be a binary classifier, with its overload detection result being either overload or not overload. When the overload detection result is overload, a second indication message is output, which is used for PWM current limiting / warning. For example, the second indication message can contain the encoding [1,0]. When the overload detection result is not overload, the source current change rate signal is input into the oscillation detection SVM.

[0111] Step 403: If the overload detection result is no overload, input the source current change rate signal into the oscillation detection SVM to determine the oscillation detection result.

[0112] Optionally, the source current change rate signal can be detected by an oscillation detection SVM to determine whether an overload has occurred.

[0113] Similarly, the oscillation detection SVM can be a binary classifier, with its oscillation detection result being either oscillation or no oscillation. When the oscillation detection result is oscillation, a third indication message is output, which is used to adjust the control loop / provide a prompt. For example, the third indication message can contain the code [0,1]. When the oscillation detection result is no oscillation, the oscillation detection result will be output, or a fourth indication message will be output, which is used to indicate no fault. The fourth indication message can contain the code [0,0].

[0114] The training process of the target SVM is illustrated below. In one exemplary embodiment, such as... Figure 11 As shown, the training process of the target SVM includes the following steps 501 to 502. Wherein:

[0115] Step 501: Obtain the training dataset.

[0116] The training dataset includes multiple training input data and fault labels corresponding to each training input data. Each training input data is a sample source current change rate signal.

[0117] Optionally, the corresponding sample source current change rate signal can be obtained by acquiring multiple sample source current signals, and then preprocessing and extracting features from the multiple sample source current signals.

[0118] The methods of preprocessing and feature extraction will not be elaborated upon.

[0119] Optionally, a fixed-length data segment can be extracted from the long-term source current signal, centered on each switching event or fault event. Each data segment is a sample source current signal, and a type label, such as normal, overload, short circuit, oscillation, etc., can be manually assigned to each sample source current signal.

[0120] Step 502: Determine the target penalty parameters and target kernel parameters of the initial SVM based on the training dataset and optimization algorithm to obtain the target SVM.

[0121] Optionally, considering that local deployment is limited by computing resources, the penalty parameters and kernel parameters of the SVM can be globally optimized in advance during the offline phase.

[0122] The search space for the penalty parameter, which is also the preset parameter range of the penalty parameter, is

[10] . -2 10 4 The search space for kernel parameters, which is also the preset parameter range of kernel parameters, is

[10] . -5 10 0 ].

[0123] Optionally, by optimizing the algorithm to perform global optimization in the search space and determining the target penalty parameter and target kernel parameter, the search efficiency can be greatly improved.

[0124] Optionally, the optimization algorithm can be an improved particle swarm optimization algorithm, or other types of optimization algorithms, such as ant colony optimization, genetic algorithms, etc. An improved particle swarm optimization algorithm is an optimization algorithm for finding the optimal solution. The flight process of a particle is its search process. The particle's flight speed can be dynamically adjusted based on the particle's historical best position and the population's historical best position; the particle's speed and position are continuously updated until a globally optimal solution satisfying the termination condition is found.

[0125] Taking the improved particle swarm optimization algorithm as an example, such as Figure 12As shown, the target penalty parameters and target kernel parameters of the initial SVM are determined based on the training dataset and optimization algorithm, including the following steps 601 to 602. Wherein:

[0126] Step 601: Randomly generate a particle swarm within a preset parameter range. The particle swarm includes the positions of multiple particles, and the positions of the particles represent a set of penalty parameters and kernel parameters.

[0127] Optionally, the parameters of the particle swarm optimization algorithm can be initialized first: the population size (N) represents the number of different failure modes or initial guesses; the maximum number of iterations (T) represents the maximum number of loops the algorithm can run; the inertia weight (ω) represents the influence of the particle's current velocity on the next velocity; the acceleration coefficients (c1) and (c2) represent the acceleration of the particle as it approaches its historical best position and global best position, respectively; the position boundary corresponds to the physical range of the voltage, current and other parameters of the converter; and the velocity boundary corresponds to the rate of change of the parameters.

[0128] Then, the positions of multiple particles can be randomly generated within a preset parameter range, and the velocity of each particle can be initialized.

[0129] For example, the initial position and initial velocity of the j-th particle can be expressed as:

[0130]

[0131] in, and These are the initial position and velocity, respectively. and These are the minimum and maximum values ​​of the position, respectively. and These are the maximum and minimum speed values, respectively.

[0132] Step 602: Iterate through the position update process multiple times until the iteration deadline is met, and determine the target penalty parameter and target kernel parameter based on the results of the iteration.

[0133] The i-th position update process includes:

[0134] An initial SVM is trained based on the position of each particle and the training dataset. The fitness of each particle is calculated according to the fitness function. The historical best position of each particle and the global best position of the particle swarm are determined based on the fitness of each particle. The fitness function includes classification accuracy or precision. The position of each particle is updated based on the historical best position of each particle and the global best position of the particle swarm.

[0135] Optionally, for each particle, the position [c, The initial SVM is then trained using the following parameters: penalty parameter, kernel parameter, and training dataset. The fitness value of each particle is calculated using classification accuracy, precision, or F1 score as the fitness function. The historical best position of each particle, as well as the global best position of the entire population, are also recorded.

[0136] Optionally, the fitness function may also include recall and F1 score; where,

[0137] Accuracy is a commonly used metric in model evaluation, used to measure the accuracy of model predictions. The calculation formula is as follows:

[0138]

[0139] Recall measures a model's ability to identify positive samples; it is also known as sensitivity or recall. The formula is as follows:

[0140]

[0141] Precision measures how accurately a model predicts positive samples. The formula is:

[0142]

[0143] The F1 score is a metric that considers both precision and recall, used to evaluate model performance. It is the harmonic mean of precision and recall, calculated as follows:

[0144]

[0145] In the above formula, TP is the number of true positive examples (model predicts positive samples and they are actually positive samples), TN is the number of true negative examples (model predicts negative samples and they are actually negative samples), FP is the number of false positive examples (model predicts positive samples but they are actually negative samples), and FN is the number of false negative examples (model predicts negative samples but they are actually positive samples).

[0146] Optionally, such as Figure 13 As shown, the position of each particle is updated based on its historical best position and the global best position of the particle swarm, including steps 701 to 702. Wherein:

[0147] Step 701: For each particle, update the particle's flight speed based on the particle's historical best position and the global best position of the particle swarm.

[0148] Step 702: Update the particle's position based on its flight speed.

[0149] For example, the velocity and position of each particle can be updated according to the following formula:

[0150]

[0151] in, For the particle's historical optimal position, g best This is the global optimal position of the particle. and It is a random number between [0,1].

[0152] Optionally, the inertia weight can be adaptively adjusted during the iteration process to balance global and local search capabilities, thereby improving optimization accuracy and convergence speed. The inertia weight ω can be determined by the following formula:

[0153]

[0154] in, , These are the maximum and minimum values ​​of the inertia weight, respectively.

[0155] Understandably, after updating the particle position, the fitness calculation, velocity update, and position update steps can be repeated until the maximum number of iterations is reached or the fitness value converges; the globally optimal parameter combination [c,] is output. ].

[0156] For ease of understanding, the following describes the fault detection method provided in this application embodiment using a complete example.

[0157] Obtain a training dataset, which includes multiple training input data and corresponding fault labels for each training input data. Each training input data is a sample source current change rate signal. Randomly generate a particle swarm within a preset parameter range. The particle swarm includes the positions of multiple particles, and the particle positions represent a set of penalty parameters and kernel parameters. Iterate through multiple position update processes until the iteration cutoff condition is met. Determine the target penalty parameters and target kernel parameters based on the results of the iterations to obtain the target SVM. The i-th position update process includes: training an initial SVM based on the positions of each particle and the training dataset, calculating the fitness of each particle according to the fitness function, and determining the historical best position of each particle and the global best position of the particle swarm based on the fitness of each particle. The fitness function includes classification accuracy or precision. For each particle, update the particle's flight velocity based on the particle's historical best position and the global best position of the particle swarm. Update the particle's position based on the particle's flight velocity.

[0158] Acquire the source current signal of the solid-state circuit breaker; preprocess the source current signal, including normalization and smoothing filtering; perform first-order differentiation on the preprocessed source current signal according to the Savitzky-Golay algorithm to determine the source current rate of change signal corresponding to the source current signal.

[0159] The target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. The source current change rate signal is input into the short-circuit detection SVM to determine the short-circuit detection result. If the short-circuit detection result is no short circuit, the source current change rate signal is input into the overload detection SVM to determine the overload detection result. If the overload detection result is no overload, the source current change rate signal is input into the oscillation detection SVM to determine the oscillation detection result.

[0160] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0161] Based on the same inventive concept, this application also provides a fault detection device for implementing the fault detection method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more fault detection device embodiments provided below can be found in the limitations of the fault detection method described above, and will not be repeated here.

[0162] In one exemplary embodiment, such as Figure 14 As shown, a fault detection device 800 is provided, including: an acquisition module 801, a feature extraction module 802, and a detection module 803, wherein:

[0163] Acquisition module 801 is used to acquire the source current signal of the solid-state circuit breaker;

[0164] The feature extraction module 802 is used to extract features from the source current signal and determine the source current change rate signal corresponding to the source current signal.

[0165] The detection module 803 is used to input the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm.

[0166] In one embodiment, the target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. The detection module 803 is specifically used to input the source current change rate signal into the short-circuit detection SVM to determine the short-circuit detection result; if the short-circuit detection result is no short circuit, the source current change rate signal is input into the overload detection SVM to determine the overload detection result; if the overload detection result is no overload, the source current change rate signal is input into the oscillation detection SVM to determine the oscillation detection result.

[0167] In one embodiment, the feature extraction module 802 is specifically used to preprocess the source current signal, including normalization and smoothing filtering; and to perform first-order differentiation processing on the preprocessed source current signal according to the Savitzky-Golay algorithm to determine the source current rate of change signal corresponding to the source current signal.

[0168] In one embodiment, the fault detection device further includes a training module for acquiring a training dataset, wherein the training dataset includes multiple training input data and fault labels corresponding to each training input data, and each training input data is a sample source current change rate signal; the target penalty parameters and target kernel parameters of the initial SVM are determined according to the training dataset and the optimization algorithm to obtain the target SVM.

[0169] In one embodiment, the optimization algorithm is an improved particle swarm optimization algorithm. The training module is specifically used to randomly generate a particle swarm within a preset parameter range. The particle swarm includes the positions of multiple particles, and each particle's position represents a set of penalty parameters and kernel parameters. The module iteratively executes multiple position update processes until an iteration cutoff condition is met. Based on the results of the iterations, the target penalty parameters and target kernel parameters are determined. The i-th position update process includes:

[0170] An initial SVM is trained based on the position of each particle and the training dataset. The fitness of each particle is calculated according to the fitness function. The historical best position of each particle and the global best position of the particle swarm are determined based on the fitness of each particle. The fitness function includes classification accuracy or precision.

[0171] The positions of each particle are updated based on its historical best position and the global best position of the particle swarm.

[0172] In one embodiment, the training module is specifically used to update the flight velocity of each particle based on the particle's historical best position and the global best position of the particle swarm; and to update the particle's position based on the particle's flight velocity.

[0173] Each module in the aforementioned fault detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the solid-state circuit breaker in hardware form or independent of it, or stored in the memory of the solid-state circuit breaker in software form, so that the processor can call and execute the corresponding operations of each module.

[0174] In one exemplary embodiment, a solid-state circuit breaker is provided, the internal structure of which can be shown in the figure below. Figure 15 As shown, the solid-state circuit breaker includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data related to the fault detection method. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a fault detection method.

[0175] Those skilled in the art will understand that Figure 15 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the solid-state circuit breaker to which the present application is applied. A specific solid-state circuit breaker may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0176] In one exemplary embodiment, a solid-state circuit breaker is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps described in any of the above methods.

[0177] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs the steps described in any of the methods above.

[0178] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps described in any of the methods above.

[0179] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0180] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0181] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A fault detection method, characterized in that, The method includes: Obtain the source current signal of the solid-state circuit breaker; Feature extraction is performed on the source current signal to determine the source current change rate signal corresponding to the source current signal; The source current change rate signal is input into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm.

2. The method according to claim 1, characterized in that, The target SVM includes a short-circuit detection SVM, an overload detection SVM, and an oscillation detection SVM. The source current change rate signal is input into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM, including: The source current change rate signal is input into the short-circuit detection SVM to determine the short-circuit detection result; If the short circuit detection result is no short circuit, the source current change rate signal is input into the overload detection SVM to determine the overload detection result; If the overload detection result is no overload, the source current change rate signal is input into the oscillation detection SVM to determine the oscillation detection result.

3. The method according to claim 1, characterized in that, The step of extracting features from the source current signal to determine the source current change rate signal corresponding to the source current signal includes: The source current signal is preprocessed, including normalization and smoothing filtering. The source current rate of change signal corresponding to the preprocessed source current signal is determined by performing first-order differential processing on the preprocessed source current signal according to the Savitzky-Golay algorithm.

4. The method according to claim 1, characterized in that, The training process of the target support vector machine (SVM) includes: Obtain a training dataset, wherein the training dataset includes multiple training input data and fault labels corresponding to each training input data, and each training input data is a sample source current change rate signal; Based on the training dataset and optimization algorithm, the target penalty parameters and target kernel parameters of the initial SVM are determined, and the target SVM is obtained.

5. The method according to claim 4, characterized in that, The optimization algorithm is an improved particle swarm optimization algorithm. The step of determining the target penalty parameters and target kernel parameters of the initial SVM based on the training dataset and the optimization algorithm includes: A particle swarm is randomly generated within a preset parameter range. The particle swarm includes the positions of multiple particles, and the positions of the particles represent a set of penalty parameters and kernel parameters. The position update process is executed iteratively multiple times until the iteration cutoff condition is met, and the target penalty parameter and target kernel parameter are determined based on the results of the iterative execution. The location update process for the i-th time includes: An initial SVM is trained based on the position of each particle and the training dataset. The fitness of each particle is calculated according to the fitness function. The historical optimal position of each particle and the global optimal position of the particle swarm are determined based on the fitness of each particle. The fitness function includes classification accuracy or precision. The positions of each particle are updated based on its historical best position and the global best position of the particle swarm.

6. The method according to claim 5, characterized in that, The step of updating the position of each particle based on its historical best position and the global best position of the particle swarm includes: For each particle, the particle's flight velocity is updated based on the particle's historical best position and the particle swarm's global best position. The position of the particle is updated based on its flight speed.

7. A fault detection device, characterized in that, The device includes: The acquisition module is used to acquire the source current signal of the solid-state circuit breaker; The feature extraction module is used to extract features from the source current signal and determine the source current change rate signal corresponding to the source current signal. The detection module is used to input the source current change rate signal into a pre-trained target support vector machine (SVM) to determine the fault detection result based on the output of the target SVM. The target penalty parameter and target kernel parameter in the target SVM are determined based on an optimization algorithm.

8. A solid-state circuit breaker, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images