Arc fault detection method based on pulse neural network

Abstract

The invention discloses an arc fault detection method based on a pulse neural network, and belongs to the technical field of electrical detection, and the method comprises the following steps: obtaining a normal arc signal and a fault arc signal under multiple loads, and carrying out the preprocessing of all arc signals; the preprocessed arc signals are converted into first images through a Markov conversion field, and all the first images are divided into a training set and a test set; and coding each first image based on Poisson coding, and training a neuron model by using the coded training set and test set to obtain an arc fault detection model. According to the method, the arc fault detection model is trained in combination with pulse coding and the neuron model, and the stochastic gradient descent method is introduced for online learning of the model, so that the accuracy of a traditional neural network algorithm is improved, and the problems of misjudgment and missed judgment of a fault arc are solved.

Description

Technical Field

[0001] The invention relates to the technical field of electrical detection, and in particular to an arc fault detection method based on a pulse neural network. Background Art

[0002] With the continuous advancement and rapid development of science and technology, people's requirements for electrical safety have been further improved. Therefore, the research on AC arc fault detection methods has attracted the attention of scholars in more fields. How to further improve the effectiveness of arc fault detection and reduce the misjudgment rate based on the existing research results and combined with new scientific and technological applications, such as big data and artificial intelligence, is the current research hotspot and difficulty in the academic community.

[0003] Current arc fault detection technologies can be roughly divided into the following four categories:

[0004] 1. Arc fault detection technology based on simulation model: A large number of scholars have studied arc model simulation, and derived a variety of arc models based on the law of conservation of energy and arc column plasma characteristics, such as the Cassie model for simulating high-resistance arcs and the Mayr model for simulating low-resistance arcs. The arc harmonic content is obtained through the constructed arc model, and then the basis for arc fault judgment is extracted. However, the fault current of the power grid is small and easily affected by noise, and the arc model cannot fully consider the impact of noise;

[0005] 2. Arc fault detection technology based on physical characteristics: It mainly uses the characteristics of arc light, radiation, temperature and noise. When the above characteristics of the arc are detected, an arc fault is determined. In some scenarios, arc faults are identified by combining arc spectrum and current spectrum, or using temperature sensors, sound sensors and arc intensity sensors to extract arc fault characteristics, and then using neural network fusion algorithms to identify arc faults, or by detecting the electric field and magnetic field radiated by the arc, analyzing the arc electromagnetic radiation signals under different attenuation distances, considering shielding, nonlinear load interference and different sensor measurement orientations, and then identifying arc faults. However, since the sensor can only monitor the physical characteristics of the arc within a specific range, the arc fault detection technology based on physical characteristics is only suitable for arc fault detection in specific parts;

[0006] 3. Arc fault detection technology based on time-frequency characteristics: It mainly analyzes the distorted arc voltage waveform and arc current waveform, and then extracts parameters such as current average value, current change rate, current charge content change, and current effective value standard deviation as characteristic signals from the time domain. In the frequency domain, it extracts parameters such as broadband noise periodicity, harmonic content, total harmonic distortion rate, and interharmonic content as characteristic signals to identify arc faults. Arc fault detection technology based on time-frequency characteristics is currently the most commonly used arc fault detection method, but it only uses the time domain characteristics of arc faults for identification, resulting in a high misjudgment rate, and the collection of voltage waveforms is easily restricted by the installation location.

[0007] 4. Arc fault detection technology based on artificial intelligence algorithm: At present, the use of emerging technologies such as advanced intelligent computing, adaptive control algorithms, and logical strategy expression to intelligently process arc fault characteristics and design corresponding detection technologies has successfully broken through the limitations of traditional detection technologies and has become a new hotspot and key research direction. For example, the least squares support vector machine is used to classify the characteristic vectors of arc current signals to identify series arc faults, or characteristic signals are extracted from arc electromagnetic radiation signals and current signals, and a mapping relationship between each characteristic quantity and arc fault is established by a multi-layer feedforward neural network, namely, a BP neural network.

[0008] Therefore, these fault arc detection technologies have their own shortcomings, and they also face common challenges. First, under complex working conditions, the objects that the fault arc needs to deal with are becoming more and more complex, which undoubtedly increases the complexity of detection. Secondly, in actual situations, the fault arc is often interfered by noise such as sensors, resulting in a decrease in signal quality, which in turn causes misjudgment and missed detection of the fault arc. In addition, the existing detection technology cannot identify the fault arc in a very short time, and the effects in terms of real-time detection and response speed are not good. Summary of the invention

[0009] The purpose of the present invention is to provide an arc fault detection method based on a pulse neural network to solve the problem that the detection complexity is getting higher and higher and the existing technology cannot identify the fault arc in a very short time, and the detection result is also susceptible to noise interference from sensors.

[0010] To achieve the above object, the present invention adopts the following technical solutions:

[0011] An arc fault detection method based on a pulse neural network of the present invention comprises the following steps:

[0012] Obtain normal arc signals and fault arc signals under multiple loads, and pre-process all arc signals;

[0013] The preprocessed arc signals are converted into first images respectively by using a Markov transition field, and all the first images are divided into a training set and a test set;

[0014] Each first image is encoded based on Poisson coding, and the encoded training set and test set are used to train a neuron model to obtain an arc fault detection model.

[0015] Preferably, the preprocessing of all arc signals comprises:

[0016] The mean filter window size corresponding to each load is determined, and based on the window size, all the normal arc signal and the fault arc signal are input into the mean filter to perform mean calculation to obtain the preprocessed arc signal.

[0017] Preferably, the method of converting the pre-processed arc signals into first images respectively using a Markov transition field comprises:

[0018] Obtaining a time series of each preprocessed arc signal, and binning a value range of the time series;

[0019] Counting the frequency of bin transitions in the time series, and constructing a Markov transition matrix based on the frequency;

[0020] The Markov transformation matrix is ​​expanded into a Markov transformation field, and a first image is generated according to the Markov transformation field.

[0021] Preferably, encoding each first image based on Poisson coding comprises:

[0022] Normalize each first image to obtain a second image whose pixel values ​​are between [0, 1], and determine the encoding time step and total time according to the arc period;

[0023] Calculating the probability of pulse emission in each time step according to the pixel value of the second image, and generating a random number corresponding to each pixel in each time step;

[0024] The pulse value of each pixel in the corresponding time step is determined according to the random number and probability, and the above steps are repeated until the total time is reached to obtain the pulse sequence of each pixel.

[0025] Preferably, the calculating the probability of pulse emission in each time step according to the pixel value of the second image comprises:

[0026] When the random number is less than or equal to the pulse value of the corresponding pixel in the corresponding time step, a pulse is issued for the pixel in the time step.

[0027] Preferably, it is characterized in that the neuron model is a leakage-integration-ignition model, and the hidden layer of the leakage-integration-ignition model includes excitatory neurons and inhibitory neurons with an excitation-inhibition ratio of four to one.

[0028] Preferably, the method further comprises:

[0029] When a new fault arc signal is obtained, the parameters of the arc fault detection model are updated using the stochastic gradient descent method.

[0030] An electronic device comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement an arc fault detection method based on a pulse neural network as described in any one of the above.

[0031] A computer-readable storage medium storing a computer program, wherein the computer program enables a computer to implement an arc fault detection method based on a pulse neural network as described in any one of the above.

[0032] A computer program product comprises a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, an arc fault detection method based on a pulse neural network as described in any one of the above is implemented.

[0033] The present invention has the following beneficial effects:

[0034] The present invention combines pulse coding and neuron models to train arc fault detection models, and introduces stochastic gradient descent method to perform online learning of the model, which not only improves the accuracy of traditional neural network algorithms, but also improves the problems of misjudgment and missed judgment of fault arcs. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of an arc fault detection method based on a pulse neural network provided by an embodiment of the present invention;

[0036] Figure 2 is a schematic diagram of a fault arc experimental platform in an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of an equivalent circuit diagram of LIF in an embodiment of the present invention;

[0038] Figure 4 is a flow chart of the stochastic gradient descent method in an embodiment of the present invention;

[0039] Figure 5 is a flow chart of a Markov transition field in an embodiment of the present invention;

[0040] Figure 6is a flow chart of Poisson coding in an embodiment of the present invention;

[0041] Figure 7 is a comparison diagram of the loss rate of training in an embodiment of the present invention;

[0042] Figure 8 is a comparison chart of the accuracy of training in an embodiment of the present invention;

[0043] Fig. 9 It is a schematic diagram of an electronic device for implementing an arc fault detection method based on a pulse neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the technical solution of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The terms "first", "second", etc. in the claims and specification of the present invention are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances. This is only to describe the distinction method used in the description of objects with the same attributes in the embodiments of the present invention. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that a process, method, system, product or device containing a series of units is not necessarily limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or devices.

[0045] In some embodiments, Figure 1 As shown, an arc fault detection method based on a pulse neural network comprises the following steps:

[0046] S110, obtaining normal arc signals and fault arc signals under multiple loads, and preprocessing all arc signals;

[0047] S120, using a Markov transition field to convert the preprocessed arc signals into first images respectively, and dividing all the first images into a training set and a test set;

[0048] S130, encoding each first image based on Poisson coding, and using the encoded training set and test set to train a neuron model to obtain an arc fault detection model.

[0049] In this embodiment, the normal arc signals and fault arc signals collected under multiple loads are from a point-contact series fault arc experimental platform built according to the national standard GB / T 31143-2014, such as Figure 2As shown in the figure, the experimental platform consists of an arc generator, a 220V AC power supply, a current transformer, an oscilloscope, a switch and a load, wherein the load includes a vacuum cleaner, a switching power supply, a capacitor starter motor, a fluorescent lamp, a halogen lamp and an electric hand tool. After the signal acquisition is completed, all the collected arc signals are imported into MATLAB and preprocessed. MATLAB is an advanced technical computing language and interactive environment for algorithm development, data visualization, data analysis and numerical calculation. It provides a rich function library and toolbox, enabling users to solve technical computing problems more quickly and efficiently.

[0050] In this embodiment, the preprocessing is to create a mean filter in MATLAB, and then select a suitable mean filter window size for each load according to the complexity of the load. Since the signal does not require pixel details like a picture, a smaller window can be selected to remove fine noise, and then all the collected arc signals are input into the mean filter to perform mean calculation according to the selected window size. Preferably, mean filtering is implemented using a filter function, thereby removing uniform noise and Gaussian noise in the arc signal, making the data smoother. Here, the window size used is windowSize=5, which will not cause signal information loss due to a window that is too large, nor will the denoising be unclear due to a window that is too small.

[0051] Each preprocessed arc signal is converted into a first image using Markov Transition Fields (MTF) to obtain a first image set so that the pulse neural network can better identify relevant information. MTF is an image encoding method that uses the transition probability matrix of the Markov chain to convert time series data into a two-dimensional image, which not only contains the transition probability information between states, but may also contain other features or attributes related to state transition.

[0052] Then, to ensure the validity of the data set, the first image set is divided into a training set and a test set. In this embodiment, 60 groups of normal data and fault data are collected under each load, totaling 720 groups of data. Under each load, MATLAB's FOR statement is used to randomly select 50 groups of first images for model training and verification to form a training set, and the remaining 10 groups of first images are used for model testing and evaluation to form a test set, thereby determining the actual application effect of the model.

[0053] After dividing the test set and the training set, Poisson coding is used to encode each first image to obtain the second image. Poisson coding is a coding method that converts a signal into a pulse sequence. The encoded training set and test set are then input into the neuron model. In essence, the second image in the training set is first input into the neuron model for training, and then the model trained by the training set is verified by the test set. The final result is the arc fault detection model, in which the neuron model is the leakage-integration-ignition model, namely the LIF model. The LIF model is one of the first models that can truly simulate neurons to a certain extent. It intuitively describes the relationship between the neuron membrane potential V and its input current I. The electrical signal transmitted by the presynaptic neuron will generate a PSP on the postsynaptic neuron and accumulate on the membrane voltage. When the membrane voltage exceeds the I value, the neuron ignites. "Integration" refers to the accumulation of membrane voltage, and "leakage" refers to the action that the membrane voltage gradually decays to the resting potential when no new PSP is received (if it does not decay, it is the IF model).

[0054] The LIF model is equivalent to a capacitor and a resistor in parallel. The equivalent circuit diagram is as follows: Figure 3 As shown, according to the KCL law, the equivalent current of the cell membrane is as follows:

[0055] I(t)=I R +I C (1)

[0056] Among them, the current IR flowing through the resistor can be expressed as:

[0057]

[0058] The current IC passing through the capacitor can be expressed as:

[0059]

[0060] Substituting equations (2) and (3) into equation (1), we can obtain:

[0061]

[0062] Multiply both ends of equation (4) by the membrane resistance R m , and the standard formula of LIF model is obtained:

[0063]

[0064] Then introduce the time constant τ m =R m C m , the LIF standard form can be simplified to the following first-order differential equation form:

[0065]

[0066] In the above expression, I(t) represents the sum of the synaptic currents generated by the discharge behavior of each synaptic neuron, V rest Represents the resting potential.

[0067] In this embodiment, the LIF model includes an input layer, a hidden layer and an output layer, wherein the image receiving size set in the input layer is 28*28, and the hidden layer mainly includes excitatory neurons and inhibitory neurons, and the ratio is 4:1. According to the experiment of this application, when the excitation-inhibition ratio is 4:1, the inhibitory neurons help prevent the network from being overly excited and experiencing "explosive activity" and maintain overall stability. At the same time, the inhibitory neurons can inhibit some specific neuron groups, help distinguish different input modes, and improve the computing power and information processing accuracy of the network. The output layer has two types of neurons, and the goal is to distinguish and identify accurate signals from incorrect signals.

[0068] When a new fault arc signal is obtained, the stochastic gradient descent algorithm, SGD algorithm, is used to update the parameters of the arc fault detection model. SGD algorithm is a popular algorithm for optimizing machine learning models. It is suitable for large-scale data sets, such as Figure 4 As shown in the figure, the calculation process is as follows: randomly select a sample of a training set and a test set in each iteration. When a new fault arc signal is input, initialize the parameters of the arc fault detection model, preprocess the new fault arc signal, and then input it into the arc fault detection model for training. Extract features from the new input data, perform Poisson encoding based on the features and pass the features into the first layer of neurons, calculate the output, and then pass the data output from the first layer into the second layer of neurons, calculate the output, complete forward propagation, and obtain the predicted value of the model training. Then compare the predicted value obtained from the training with the label in the test set, calculate the error, and start from the output layer and calculate the gradient of the error layer by layer to complete the back propagation until the gradient of the input layer is calculated and the parameters of the model are updated.

[0069] Use the SDG algorithm to update the weights and biases of the model. The update rules are as follows:

[0070] ω t+1 =ω t -η▽f i (ω t )

[0071] In this formula, ω represents the model parameters, t represents the current number of iterations, η is the learning rate, which controls the size of the step, and fi is the loss function for the i-th data point or data batch.

[0072] In the continuous iteration, the test set is used to evaluate whether the accuracy of the model update meets the expectations, so as to decide whether to continue to update the model parameters. If the expected effect is achieved, the online learning is stopped, the results are output and the updated model parameters are saved, and the new fault arc signal is passed into the fault arc database. Otherwise, the data is transmitted to the first layer of neurons to continue iteration to improve the accuracy of the simulation.

[0073] This embodiment can identify faults more quickly and cut off fault arcs through pulse coding, thereby improving the stability of fault detection. At the same time, new fault points are continuously learned through the training and online learning of the pulse neuron model, and the database of current arcs is expanded, thereby improving the accuracy of fault arcs.

[0074] In some embodiments, Figure 5 As shown, the pre-processed arc signals are converted into first images respectively using a Markov transition field, including:

[0075] S122, obtaining the time series of each preprocessed arc signal, and binning the value range of the time series;

[0076] S124, counting the frequency of bin transfers in the time series, and constructing a Markov transition matrix based on the frequency;

[0077] S126: Expand the Markov transformation matrix into a Markov transformation field, and generate a first image according to the Markov transformation field.

[0078] In this embodiment, after the data preprocessing is completed, the time series X of the preprocessed arc signal is first obtained. n}, divide their value range into Q bins, then each x i , can be mapped to a corresponding q i (1≤i≤Q), that is:

[0079] X={x1,x2,x3,...,x n}={q1,q2,......,q i ,......,q Q}.

[0080] Since the state space of the time series after binning is limited, the Markov transition matrix corresponding to the time series can be obtained:

[0081]

[0082] Among them, P ij =p(x t ∈q i |x t-1 ∈qj )=c i，j / ∑ i c i，j Indicated by the state q i Transfer to state q j The probability of i，j (1≤i, j≤Q) represents state q i Transition to state q j The count of all possible data points, and each row i satisfies Conditions required.

[0083] Further, the matrix P QQ Expanded into the MTF matrix, called M nn :

[0084]

[0085] Among them, m ij Represents bucket q j The specific element x in j Migrate to q i The specific element x i The probability that for each row element i satisfies

[0086] The resulting MTF matrix is ​​then visualized as an image, where different colors or brightness represent different transition probabilities.

[0087] This embodiment uses MTF to convert normal arc signals and fault arc signals into images, which can enable the pulse neural network to better recognize the information therein.

[0088] In some embodiments, Figure 6 As shown, encoding each first image based on Poisson coding includes:

[0089] S132, normalize each first image to obtain a second image whose pixel values ​​are all between [0, 1], and determine the encoding time step and total time according to the arc period;

[0090] S134, calculating the probability of pulse emission in each time step according to the pixel value of the second image, and generating a random number corresponding to each pixel in each time step;

[0091] S136. Determine the pulse value of each pixel in the corresponding time step according to the random number and probability, and repeat the above steps until the total time is reached to obtain the pulse sequence of each pixel.

[0092] Each first image is normalized to obtain several second images, and the pixel value of each pixel in the second image is between [0,1]. Then, the encoding time step and total time T are determined according to the arc cycle. This total time T determines how long each pixel will be encoded into a pulse sequence. In each time step, the probability of generating a pulse in the time step is calculated according to the pixel value of the second image. This probability is proportional to the pixel value. At the same time, for each pixel and each time step, a random number between [0,1] is generated. If the generated random number is less than or equal to the probability of the pixel being emitted in the current time step, a pulse is emitted for the pixel in the time step, usually recorded as 1, otherwise no pulse is emitted and recorded as 0. The above process is repeated until the total time T is reached to obtain a pulse sequence for each pixel. The Poisson encoding calculation process is simple and suitable for real-time detection of sudden faults of current arcs.

[0093] This example selects three models for experiment:

[0094] Model 1: The arc fault detection model proposed in the present invention uses the above-mentioned arc fault detection model construction method based on pulse neural network to train the model;

[0095] Model 2: Traditional CNN model, trained using existing machine learning;

[0096] Model 3: Traditional RNN model, trained using existing machine learning.

[0097] In some embodiments, three models are selected for experiments, namely, a CNN model, an RNN model and the above-mentioned arc fault detection model. 120 sets of data are collected for each load, totaling 720 sets of data. In each load, 50 sets of data are randomly selected for model training and verification, and the remaining 10 sets of data are used for model testing and evaluation. All three models are trained 50 times, the training time is recorded, and the accuracy, loss and training time of each training of the three models are recorded and calculated.

[0098] In this embodiment, the three models are compared in terms of average accuracy, training loss and training time. The comparison results are shown in Table 1. Figure 7 and Figure 8 As shown, among which, Figure 7 As shown in , the loss rate of the above arc fault detection model training is better than that of RNN and CNN models, as shown in Figure 8 As shown, the overall training accuracy of the arc fault detection model is also higher than that of the RNN model and the CNN model, and the accuracy of the model can basically reach 100% after 10 training times.

[0099] Table 1 Comparison of the effects of three models

[0100]

[0101] In this embodiment, after data preprocessing, a Markov transition field is introduced to convert the arc signal into an image form, and the image is encoded using Poisson coding, which is then input into the LIF model for training. The SGD algorithm is added to learn new fault data points online, expand the fault arc data set, and improve the accuracy and rapid response capability of current arc fault detection.

[0102] In some embodiments, Fig. 9 As shown, a computer device is provided, including a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the above-mentioned arc fault detection method based on pulse neural network.

[0103] The computer device can be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a pulse neural network-based arc fault detection method is implemented.

[0104] This embodiment also provides a computer-readable storage medium on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the arc fault detection method based on a pulse neural network as described above are implemented.

[0105] This embodiment also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the steps of the above-mentioned arc fault detection method based on a pulse neural network.

[0106] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0107] The computer program instructions for performing the operation of the present invention may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present invention.

[0108] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0109] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0110] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An arc fault detection method based on pulse neural network, characterized in that: The following steps are involved: Obtain normal arc signals and fault arc signals under multiple loads, and pre-process all arc signals; The preprocessed arc signals are converted into first images respectively by using a Markov transition field, and all the first images are divided into a training set and a test set; Each first image is encoded based on Poisson coding, and the encoded training set and test set are used to train a neuron model to obtain an arc fault detection model.

2. The arc fault detection method based on pulse neural network according to claim 1 is characterized in that: The preprocessing of all arc signals comprises: The mean filter window size corresponding to each load is determined, and based on the window size, all the normal arc signal and the fault arc signal are input into the mean filter to perform mean calculation to obtain the preprocessed arc signal.

3. The arc fault detection method based on pulse neural network according to claim 1, characterized in that: The method of using the Markov transition field to convert the pre-processed arc signals into first images respectively includes: Obtaining a time series of each preprocessed arc signal, and binning a value range of the time series; Counting the frequency of bin transitions in the time series, and constructing a Markov transition matrix based on the frequency; The Markov transformation matrix is ​​expanded into a Markov transformation field, and a first image is generated according to the Markov transformation field.

4. The arc fault detection method based on pulse neural network according to claim 1, characterized in that: The encoding of each first image based on Poisson coding includes: Normalize each first image to obtain a second image whose pixel values ​​are between [0, 1], and determine the encoding time step and total time according to the arc period; Calculating the probability of pulse emission in each time step according to the pixel value of the second image, and generating a random number corresponding to each pixel in each time step; The pulse value of each pixel in the corresponding time step is determined according to the random number and probability, and the above steps are repeated until the total time is reached to obtain the pulse sequence of each pixel.

5. The arc fault detection method based on pulse neural network according to claim 4 is characterized in that: The step of calculating the probability of pulse emission in each time step according to the pixel value of the second image comprises: When the random number is less than or equal to the pulse value of the corresponding pixel in the corresponding time step, a pulse is issued for the pixel in the time step.

6. The arc fault detection method based on pulse neural network according to claim 1, characterized in that: The neuron model is a leakage-integration-ignition model, and the hidden layer of the leakage-integration-ignition model includes excitatory neurons and inhibitory neurons with an excitation-inhibition ratio of four to one.

7. The arc fault detection method based on pulse neural network according to claim 1, characterized in that: The method further comprises: When a new fault arc signal is obtained, the parameters of the arc fault detection model are updated using the stochastic gradient descent method.

8. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement an arc fault detection method based on a pulse neural network as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to implement an arc fault detection method based on a pulse neural network as claimed in any one of claims 1 to 7 when executed.

10. A computer program product, comprising a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, the arc fault detection method based on a pulse neural network as claimed in any one of claims 1 to 7 is implemented.

Images