Capacitor abnormal discharge detection method and device, storage medium and computer equipment

Abstract

The invention provides a capacitor abnormal discharge detection method and device, a storage medium and computer equipment, and relates to the technical field of capacitor state monitoring, and the method comprises the steps: collecting multiple paths of voltages and multiple paths of currents of a capacitor to be detected, and carrying out the pulse coding of each path of voltage and each path of current, and obtaining each path of pulse sequence; a spiking neural network is determined, the number of input channels of the spiking neural network is the same as the number of the pulse sequences, and the spiking neural network establishes a relation with the characteristic response of abnormal discharge by adopting a specific neuron group; and inputting each pulse sequence into the pulse neuron network to obtain the abnormal discharge type of the capacitor to be detected. Therefore, the adaptive learning capability of the complex pulse sequence can be realized, so that the health state of the capacitor can be accurately and comprehensively evaluated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of capacitor state monitoring, and particularly relates to a capacitor abnormal discharge detection method and device, a storage medium and computer equipment. BACKGROUND

[0002] Metalized film capacitors have been widely used in many electrical fields due to their excellent electrical characteristics. However, abnormal discharge may occur inside the capacitor during long-term operation. These abnormal discharges usually occur in the form of complex pulse sequences. If these pulse sequences are not detected and suppressed in time, the capacitor dielectric performance will degrade, the service life will be shortened, and even catastrophic failure may occur. Therefore, effective monitoring of the health status of metalized film capacitors, especially detection and analysis of abnormal discharge pulse sequences, is of great significance to ensure their safe and reliable operation.

[0003] In the aspect of monitoring the health status of metalized film capacitors, some technologies have been applied, but there are still obvious deficiencies. Most of the existing methods are only suitable for offline testing or require the capacitor to be shut down, and thus cannot monitor the running capacitor in real time. Therefore, it is difficult to capture the complex pulse sequences generated by internal abnormal discharge in time. At the same time, the monitoring process usually relies on manually set thresholds, such as degrading the capacitor capacity to 5% as a degradation criterion. This static empirical formula cannot adapt to different types of capacitors and variable operating conditions, and it is also difficult to reflect weak or irregular abnormal discharge characteristics. In addition, the existing technology mainly identifies abnormalities indirectly through leakage current, voltage fluctuation or harmonic signals, but the sensitivity to complex weak pulse sequences is limited and is easily disturbed by background noise.

[0004] Therefore, the existing method lacks adaptive learning ability for complex pulse sequences and is difficult to accurately and comprehensively evaluate the health status of the capacitor. SUMMARY

[0005] The present application aims to at least solve one of the above technical defects, in particular the technical defect that the existing method lacks adaptive learning ability for complex pulse sequences and is difficult to accurately and comprehensively evaluate the health status of the capacitor.

[0006] In a first aspect, the present application provides a capacitor abnormal discharge detection method, which comprises:

[0007] Collecting multiple voltage and multiple current of the capacitor to be detected, and pulse coding each voltage and each current to obtain each pulse sequence;

[0008] Determining a pulse neuron network, the number of input channels of the pulse neuron network is the same as the number of pulse sequences, and the pulse neuron network uses a specific neuron group to establish a connection with the characteristic response of abnormal discharge.

[0009] The pulse sequence of each channel is input into the pulse neuron network to obtain the abnormal discharge type of the capacitor to be detected.

[0010] In one embodiment, the step of pulse coding the voltage of each channel and the current of each channel to obtain the pulse sequence of each channel includes:

[0011] The voltage of each channel and the current of each channel are time-coded and frequency-coded, and the time stamp is kept synchronized to obtain the pulse sequence of each voltage and the pulse sequence of each current.

[0012] In one embodiment, the formula of time coding is:

[0013]

[0014] The formula of frequency coding is:

[0015]

[0016] Wherein, represents the pulse triggering time, represents the maximum triggering time, represents the original signal collected, represents the minimum amplitude of the signal, represents the maximum amplitude of the signal, represents the frequency of the pulse, represents the minimum firing frequency, represents the maximum firing frequency.

[0017] In one embodiment, the membrane potential dynamic equation of the pulse neuron network is:

[0018]

[0019] Wherein, represents the membrane potential time constant, represents the resting potential, represents the membrane potential of the i-th neuron at time t, represents the membrane resistance of the neuron, represents the input current of the i-th unit of the neuron; when is greater than or equal to the threshold voltage, the neuron immediately sends a pulse and resets the membrane potential to 0, entering the refractory period.

[0020] In one embodiment, the formula of the input current in the membrane potential dynamic equation of the pulse neuron network is:

[0021]

[0022]

[0023] wherein, denotes the total input current received by the jth neuron at time t, denotes the synaptic weight from the ith input neuron to the jth neuron, denotes the spike train of the ith input neuron, denotes the synaptic transmission delay, denotes the firing time difference, denotes the amplitude constant of LTP, denotes the amplitude constant of LTD, denotes the time constant of LTP, denotes the time constant of LTD.

[0024] In one embodiment, the spiking neural network comprises an input layer, a feature extraction layer, and an output discrimination layer;

[0025] The input layer has the same number of input channels as the number of spike trains, and is configured to receive each spike train and forward the signal;

[0026] The feature extraction layer comprises two layers of sparsely connected leaky integrate-and-fire neurons, and is configured to extract the characteristic response of abnormal discharge from the input signal and establish a corresponding relationship between a specific neuron group and the characteristic response of each type of abnormal discharge;

[0027] The output discrimination layer is configured to determine the type of abnormal discharge of the capacitor to be detected according to the characteristic response of the input signal and the activation of the neuron group.

[0028] In one embodiment, the output discrimination layer is specifically configured to determine the target neuron group activated by the characteristic response of the input signal, and determine the type of abnormal discharge corresponding to the target neuron group as the type of abnormal discharge of the capacitor to be detected.

[0029] In a second aspect, the present application provides a capacitor abnormal discharge detection device, comprising:

[0030] A spike train determination module is configured to collect multiple voltage and current of the capacitor to be detected, and perform pulse coding on each voltage and current to obtain each spike train;

[0031] A spiking neural network determination module is configured to determine a spiking neural network, the spiking neural network has the same number of input channels as the number of spike trains, and the spiking neural network is configured to establish a connection between a specific neuron group and the characteristic response of abnormal discharge;

[0032] An abnormal discharge type determination module is configured to input each spike train into the spiking neural network to obtain the type of abnormal discharge of the capacitor to be detected.

[0033] Thirdly, this application provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of any of the capacitor abnormal discharge detection methods described in the above embodiments.

[0034] Fourthly, this application provides a computer device, including: one or more processors, and a memory;

[0035] The memory stores computer-readable instructions, which, when executed by one or more processors, perform the steps of any of the capacitor abnormal discharge detection methods described in the above embodiments.

[0036] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0037] The capacitor abnormal discharge detection method, apparatus, storage medium, and computer equipment provided in this application acquire multiple voltage and current signals from the capacitor under test and perform pulse coding on these signals. This transforms the originally weak, complex, and easily masked abnormal discharge pulse sequences into standardized, processable pulse sequences. Based on this, a pulse neural network is constructed. The number of input channels in this network corresponds to the number of pulse sequences, and a response relationship is established between specific neuron groups and abnormal discharge characteristics, enabling the network to adaptively learn and recognize complex and diverse pulse patterns. After inputting the pulse sequences into the network, different types of abnormal discharges can be accurately distinguished, achieving comprehensive capture of abnormal behavior inside the capacitor. Therefore, this method has high sensitivity to weak and irregular abnormal signals, thereby accurately assessing the health status of the capacitor, timely detecting potential risks, effectively extending the service life of the capacitor, and improving operational safety. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A schematic flowchart of the capacitor abnormal discharge detection method provided in the embodiments of this application;

[0040] Figure 2 This is a schematic diagram of the capacitor abnormal discharge detection device provided in the embodiments of this application;

[0041] Figure 3This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] This application provides a method for detecting abnormal discharge in a capacitor. The following embodiments illustrate this method using a computer device as an example. It is understood that the computer device can be any device with data processing capabilities, including but not limited to a single server, server cluster, personal laptop, desktop computer, etc. Figure 1 As shown, the method may include the following steps:

[0044] S101: Collect multiple voltages and currents of the capacitor under test, and perform pulse encoding on each voltage and current to obtain pulse sequences for each channel.

[0045] Among these, multi-channel voltage refers to voltage signals simultaneously acquired from different output ports or different capacitor cells of the capacitor under test. These signals can reflect abnormal discharge activities that may occur during capacitor operation. Multi-channel current refers to the current signals corresponding to each port or cell of the capacitor, used to capture instantaneous current changes caused by dielectric discharge inside the capacitor. Pulse coding refers to converting continuous voltage and current signals into discrete pulse sequences through a specific encoding method, so as to effectively represent and process weak, fast, and complex abnormal discharge signals. The pulse sequence is a sequence of coded discrete signals, with each pulse reflecting the characteristic information of the original voltage or current signal in terms of time and amplitude, providing standardized input for intelligent analysis such as neural networks.

[0046] It should be noted that the responses of current and voltage to discharge behavior differ during capacitor discharge. The localized micro-arcs generated during internal self-healing typically induce short, sharp current pulses, while terminal voltage disturbances may be minimal; however, discharges along the dielectric surface or air gaps can also produce significant transient changes in the voltage path. Therefore, by arranging two or more current and voltage signal acquisition channels, the characteristics of discharge signals under different coupling paths can be captured. When multiple probes are placed in different locations, the propagation direction of the discharge source can be estimated using the time difference across channels, and it can be determined whether the discharge is localized or extends along the surface. Different discharge types not only differ in amplitude but also exhibit different patterns in spectral, temporal, and phase characteristics. Multi-channel acquisition can provide more dimensional data for spiking neural networks to learn from. Furthermore, by utilizing differences in energy distribution, relative amplitude and time delay, or phase relationships between current and voltage channels—for example, a certain type of discharge may have a strong current and weak voltage disturbance, or voltage may precede current—the ability to distinguish between different discharge types can be significantly improved.

[0047] During implementation, the first step is to synchronously acquire multiple voltage and current signals from the capacitor under test, ensuring that the signal from each channel fully reflects the capacitor's operating state. Subsequently, the acquired continuous signals are converted into pulse sequences according to predefined encoding rules. Features such as pulse trigger time, pulse width, and pulse interval preserve the transient information of the original signal. This process allows for refined processing of weak abnormal discharge pulses, enabling the effective representation and identification of complex pulse patterns. In implementation, multi-channel parallel processing improves acquisition and encoding efficiency, while noise suppression and signal enhancement algorithms ensure the accuracy and reliability of the pulse sequences.

[0048] In one example, a high-precision current transformer and a high-bandwidth voltage probe are placed at the terminals of the metallized film capacitor to achieve four-channel sampling at a sampling frequency of at least 10 MHz, ensuring the capture of microsecond-level abnormal discharge pulse signals. To support real-time parallel processing of multi-channel signals, a high-performance GPU is used for data processing, ensuring signal processing capability and response speed at high sampling frequencies. The acquired signal is first pre-filtered using a wideband bandpass filter with an operating frequency range of 100 kHz to 5 MHz, effectively suppressing the influence of power frequency interference on weak pulse signals. Subsequently, the signal is processed by a preamplifier to provide an appropriate amplitude, keeping the analog-to-digital converter input signal within the range of 0 to 3.3 volts, preventing signal saturation or information loss, thereby ensuring the integrity and accuracy of the acquired signal.

[0049] The reason for using pulse coding to collect multiple voltage and current signals is understandable. The abnormal discharge pulses in the original signals often have low amplitude, short duration, and complex sequences, making it difficult to capture subtle features by directly processing continuous signals. Pulse coding transforms continuous signals into pulse sequences, preserving transient characteristics while removing redundant information. This allows for efficient processing and analysis of complex abnormal discharge patterns, leading to accurate identification of abnormal discharge behavior within the capacitor. This method improves the sensitivity and reliability of anomaly detection, provides high-quality input for subsequent neural network analysis, and enables a comprehensive assessment of the capacitor's health status.

[0050] S102: Determine the spiking neuron network. The number of input channels of the spiking neuron network is the same as the number of pulse sequences. The spiking neuron network establishes a connection between a specific group of neurons and the characteristic response of abnormal discharge.

[0051] Spiking neural networks (SNNs) are artificial neural network structures that take pulse signals as input and process information through neuron models. They can simulate the response characteristics of biological neurons to transient pulse signals, thereby enabling pattern recognition and classification of complex time-series signals. The number of input channels refers to the number of pathways in the network that can receive independent pulse sequence inputs. Each channel corresponds to a single acquired pulse sequence, ensuring that the features of each signal are completely transmitted to the network. A specific neuron group refers to a set of neurons trained and regulated within the network based on abnormal discharge characteristics. These neurons can respond to specific types of pulse patterns, thus forming a mapping relationship between abnormal discharge characteristics and the activation state of the neuron group.

[0052] In the implementation process, the input layer of the spiking neuron network is first constructed based on the number of acquired multi-channel pulse sequences, ensuring that each pulse sequence corresponds to an independent input channel. Each input channel can fully receive the amplitude, time, and spectral characteristics of its corresponding pulse sequence, guaranteeing that the signal is not compressed or lost when entering the network, thereby achieving parallel processing and high-fidelity transmission of multiple signals. Following the input layer, a multi-layered spiking neuron structure is established, with each layer consisting of multiple neuron groups connected internally according to a specific topology, used for comprehensive analysis of the timing, amplitude, and phase characteristics of the input pulse sequences.

[0053] Subsequently, the neuron groups within the network are defined and initialized, enabling each neuron group to develop a sensitive response to specific abnormal firing patterns. To achieve this characteristic response, the network needs to be trained. The training process uses pulse sequences of known abnormal firing types as training samples and employs supervised learning methods. By minimizing classification error, the connection weights and activation thresholds between neurons are adjusted, ensuring that specific neuron groups produce the expected activation patterns when receiving corresponding pulse sequences. Alternatively, reinforcement learning strategies can be used to optimize the response strategies of neuron groups by providing feedback on the correctness of abnormal firing identification results, thereby enhancing the network's adaptability to unknown or weak pulse patterns.

[0054] During training and use, the network structure can be dynamically adjusted according to the temporal resolution and amplitude range of the pulse sequence. For example, for pulse signals with extremely short duration or low amplitude, the network can adjust the time constant and threshold of neurons to improve sensitivity to weak transient signals; for pulses with large amplitude or long duration, the inhibitory mechanism of neurons and the interlayer connection weights can be adjusted to avoid signal saturation or response confusion. Through this refined adjustment of structure and parameters, the network can adaptively capture the characteristics of different types of abnormal discharge pulses, achieving high-precision identification and classification of weak, complex, and irregular pulse patterns, providing reliable data support for capacitor health status assessment.

[0055] It is understandable that defining a spiking neuron network with the same number of input channels as the number of pulse sequences ensures that each acquired pulse sequence can be input into the network completely and independently, thus achieving full representation and processing within the neural network. By establishing response relationships between specific neuron groups and abnormal discharge characteristics, the network can adaptively identify different types of pulse patterns and extract weak and complex discharge signal features. This significantly improves the ability to distinguish abnormal discharge types, ensuring that the feature information of complex pulse sequences is accurately captured and utilized. This enables comprehensive and accurate identification of abnormal behavior within the capacitor, providing a reliable foundation for subsequent health status assessment and improving the accuracy and safety of capacitor operation monitoring.

[0056] S103: Input the pulse sequences of each path into the spiking neuron network to obtain the abnormal discharge type of the capacitor to be detected.

[0057] Among them, abnormal discharge type refers to different categories of internal discharge behavior obtained by classifying pulse characteristics, such as local self-healing discharge, surface discharge or air gap discharge, etc., and its type information can be used to assess the health status of capacitors.

[0058] During implementation, each pulse sequence is first input into the input layer of the spiking neuron network according to the channel correspondence. Each pulse sequence enters the network through an independent channel, ensuring that the amplitude, time, and spectral characteristics of each pulse sequence are completely transmitted to the network, avoiding information loss or signal aliasing. During input, to ensure that the signal's numerical range is suitable for network processing, the pulse sequences can be normalized to standardize the amplitude to an acceptable range for the network. Simultaneously, high-frequency noise or power frequency interference is filtered and suppressed to enhance the network's sensitivity to weak abnormal signals.

[0059] After entering the network, the internal neuron group performs multidimensional feature analysis on the input pulse sequence, including amplitude changes, pulse intervals, temporal relationships, and phase characteristics. Through weighted connections between neurons and a pulse triggering mechanism, it compares and responds to the mapping relationship between pulse patterns and abnormal discharge types established during the training phase. When the input pulse sequence matches the characteristics of a certain abnormal discharge pattern, the corresponding neuron group is activated. The network classifies the signals based on the activation intensity and distribution, and finally outputs the abnormal discharge type of the capacitor to be detected.

[0060] The entire process leverages high-performance parallel computing to achieve synchronous processing and rapid response of multi-channel pulse sequences. The characteristics of each pulse sequence can be analyzed and classified in real time within the network, enabling high-precision identification of weak, complex, and irregular abnormal discharge pulses. This processing method significantly improves the accuracy and reliability of abnormal discharge identification, providing a comprehensive and accurate reference for capacitor health status assessment and effectively supporting real-time monitoring of capacitor operational safety.

[0061] It is understandable that inputting the pulse sequences into the spiking neural network ensures that the amplitude, timing, and spectral characteristics of each signal are fully transmitted to the network. Through the mapping relationship established between the neuron group and the abnormal discharge characteristics, the network adaptively learns and responds, thereby achieving automatic identification and classification of different types of abnormal discharges. This method significantly improves the sensitivity and recognition accuracy of weak and complex pulse patterns, ensuring that abnormal behavior inside the capacitor is fully captured, providing a reliable basis for subsequent health status assessment, and enhancing the accuracy and safety of capacitor operation monitoring.

[0062] In the above embodiments, by acquiring multiple voltage and current signals from the capacitor under test and performing pulse coding on these signals, the originally weak, complex, and easily masked abnormal discharge pulse sequences can be transformed into standardized and processable pulse sequences. Based on this, a pulse neural network is constructed, where the number of input channels corresponds to the number of pulse sequences, and a response relationship is established between specific neuron groups and abnormal discharge characteristics, enabling the network to adaptively learn and recognize complex and diverse pulse patterns. After inputting the pulse sequences into the network, different types of abnormal discharges can be accurately distinguished, achieving comprehensive capture of abnormal behavior inside the capacitor. Therefore, this method has high sensitivity to weak and irregular abnormal signals, thereby accurately assessing the health status of the capacitor, timely detecting potential risks, effectively extending the capacitor's service life, and improving operational safety.

[0063] In one embodiment, the step of pulse encoding each voltage and current to obtain a pulse sequence includes:

[0064] Each voltage and current channel is time-coded and frequency-coded, and the timestamps are kept synchronized to obtain the pulse sequence of each voltage channel and the pulse sequence of each current channel.

[0065] Time encoding refers to the process of converting continuous voltage or current signals into discrete pulse sequences in chronological order. The position of each pulse on the time axis corresponds to the occurrence time of characteristic events in the original signal, thus preserving the signal's timing information. Frequency encoding converts the frequency characteristics of a signal into variations in pulse firing patterns or pulse intervals, allowing the signal's spectral information to be expressed within the pulse sequence. Timestamp synchronization ensures that each voltage and current signal maintains the same time reference during the encoding process, enabling precise temporal alignment of pulse sequences across different channels and preserving the relative timing relationships across channels.

[0066] During implementation, each acquired voltage and current signal is first processed by time encoding and frequency encoding. In time encoding, pulse events are generated based on transient changes and amplitude abrupt changes in the signal, and the corresponding timestamps are recorded. In frequency encoding, the frequency components of the signal are analyzed, mapping different frequency characteristics to pulse firing patterns or interval variations, thus reflecting the spectral information of the original signal in the pulse sequence. During encoding, the timestamps of each signal are aligned using a global clock or synchronization mechanism to ensure that the pulse sequences of different channels remain synchronized in time, facilitating the network's simultaneous processing of multiple signals and analysis of cross-channel timing relationships.

[0067] In this embodiment, time-coding and frequency-coding of each voltage and current channel while maintaining timestamp synchronization fully preserves the transient, temporal, and frequency characteristics of the signal, providing high-quality, standardized input data for the spiking neuron network. This ensures that weak and complex abnormal discharge pulses are effectively captured and characterized across multiple channels, enabling the network to fully utilize cross-channel temporal and spectral information for learning and classification. This significantly improves the accuracy and sensitivity of abnormal discharge identification, provides a reliable basis for comprehensive assessment of capacitor health status, and enhances the accuracy and safety of operational monitoring.

[0068] In one embodiment, the time encoding formula is:

[0069]

[0070] The formula for frequency coding is:

[0071]

[0072] in, Indicates the pulse trigger time. Indicates the maximum trigger time. This represents the original signal that was collected. Indicates the minimum amplitude of the signal. Indicates the maximum amplitude of the signal. Indicates the frequency of the emitted pulse. Indicates the minimum distribution frequency. This indicates the highest frequency of distribution.

[0073] In this embodiment, the time encoding formula maps the amplitude of the original signal to the pulse trigger time, allowing the transient changes of the signal to be reflected in the pulse sequence in chronological order. Signals with larger or smaller amplitudes correspond to different pulse trigger times, thus preserving the temporal information of the signal. The frequency encoding formula maps the signal amplitude to the pulse firing frequency, allowing signals with different amplitudes to express spectral characteristics at different rates in the pulse sequence, reflecting the signal's variation in the frequency domain. Through these two encoding methods, the amplitude, time, and frequency characteristics of voltage and current signals are transformed into pulse trigger times and firing rates, achieving a comprehensive expression of the signal in a discrete pulse sequence. This ensures that weak and complex abnormal discharge pulse characteristics are preserved with high fidelity in the pulse sequence, enabling the spiking neural network to simultaneously acquire the temporal and frequency information of the signal. This significantly improves the accuracy and sensitivity of abnormal discharge type identification, provides a reliable basis for comprehensive assessment of capacitor health status, and enhances the accuracy and safety of operational monitoring.

[0074] In one embodiment, the dynamic equation of the membrane potential of a spiking neuron network is:

[0075]

[0076] in, Represents the membrane potential time constant. Indicates the resting potential. This represents the membrane potential of the i-th neuron at time t. Indicates neuronal membrane resistance. This represents the input current of the i-th unit of the neuron; when When the voltage is greater than or equal to the threshold voltage, the neuron immediately sends a pulse and resets the membrane potential to 0, entering the refractory period.

[0077] In this embodiment, the membrane potential dynamic equation describes the change of neuronal membrane potential over time. Under the influence of input current, the neuronal membrane potential gradually accumulates and deviates from the resting potential. When the accumulation reaches a certain threshold, the neuron generates a pulse output and rapidly resets the membrane potential, entering the refractory period. This mechanism enables the neuron to respond quickly to transient changes in the input signal. Through this dynamic model, the spiking neuron network can perform time-sensitive processing of the input pulse sequence, capturing the weak amplitude and rapidly changing characteristics of the signal, while avoiding false triggering caused by continuous high-frequency pulses. This enhances the network's ability to recognize complex and irregular pulse patterns, enabling accurate extraction and classification of abnormal discharge pulse characteristics. This improves the monitoring accuracy and reliability of abnormal behavior within the capacitor, providing a high-quality data foundation for health status assessment and ensuring operational safety.

[0078] In terms of specific parameter settings, the membrane potential time constant is set between 5 and 20 milliseconds, the membrane resistance ranges from 1 to 10 megohms, the resting potential is set to 0 millivolts, the threshold voltage is set to 20 millivolts, the membrane potential is reset to 0 after firing, and the neuron refractory period is 2 milliseconds. These parameters ensure the stability and response characteristics of the neuron when capturing weak and fast pulse signals.

[0079] Furthermore, for more complex discharge patterns, such as sudden or adaptive discharges, the LIF model can be extended to an Izhikevich model to capture the diverse characteristics of different types of anomalous discharges. The Izhikevich model uses four core parameters, a, b, c, and d, to regulate the firing behavior of neurons. Different parameter combinations can fit the anomalous discharge characteristics of different capacitors. Under normal operating conditions, the parameters are set as a=0.02, b=0.2, ... d=8, while under abnormal discharge conditions the parameters are set to a=0.1, b=0.2, d=4. Through this parameterization, the network can adaptively simulate various abnormal discharge modes, improve the ability to identify complex pulse sequences, and provide a reliable neurodynamic basis for the accurate assessment of capacitor health status.

[0080] In one embodiment, the formula for the input current in the membrane potential dynamic equation of a spiking neuron network is:

[0081]

[0082]

[0083] in, This represents the total input current received by the j-th neuron at time t. This represents the synaptic weights from the i-th input neuron to the j-th neuron. This represents the pulse sequence of the i-th input neuron. Indicates synaptic transmission delay, Indicates the time difference in distribution. The amplitude constant representing long-term enhancement. The amplitude constant representing long-term suppression. The time constant representing long-term enhancement. The time constant represents long-term suppression.

[0084] In this embodiment, the synaptic input current received by the neuron is calculated by weighted summation of the pulse sequences of each input neuron. Each input pulse, when transmitted to the target neuron, takes into account synaptic delay, which is randomly distributed within the range of 1 to 5 milliseconds. The initial synaptic weights are set between 0.05 and 0.5 to ensure that each input signal has an appropriate impact on the neuronal membrane potential in the initial state of the network. The input pulse sequence can fully reflect the temporal characteristics of the original signal and is transmitted to the target neuron through synaptic weighting, achieving comprehensive perception of multi-channel pulse information.

[0085] To enable the network to adaptively learn the temporal patterns of pulses, a time-based synaptic plasticity mechanism (STDP) is employed to adjust synaptic weights. When the time difference between consecutive pulse firings is positive, the synaptic weights increase according to the long-term enhancement rule, with an amplitude constant of 0.01 and a time constant of 20 milliseconds. When the time difference is negative, the synaptic weights decrease according to the long-term inhibition rule, with an amplitude constant of 0.012 and a time constant of 20 milliseconds. Through this mechanism, neurons can automatically capture and learn the temporal characteristics of partial discharge pulses, thereby distinguishing between normal pulses and abnormal discharge signals, improving the ability to identify complex and weak pulse patterns, and providing reliable data support for capacitor health status assessment.

[0086] The combined effect of the input current formula and the STDP mechanism enables the spiking neuron network to enhance its sensitivity to the temporal dependence of the input pulse, improve its ability to capture weak and complex abnormal discharge pulses, and dynamically optimize the network's recognition accuracy for diverse discharge patterns through synaptic plasticity. This allows for comprehensive monitoring of abnormal capacitor behavior and improves the accuracy and reliability of health status assessment.

[0087] In one embodiment, the spiking neural network includes an input layer, a feature extraction layer, and an output discriminant layer;

[0088] The number of input channels in the input layer is the same as the number of pulse sequences. The input layer is used to receive each pulse sequence and forward the signal.

[0089] The feature extraction layer contains two layers of sparsely connected leaky integral firing neurons. The feature extraction layer is used to extract the feature responses of abnormal discharges from the input signal and establish a correspondence between specific neuron groups and the feature responses of various abnormal discharges.

[0090] The output discriminant layer is used to determine the type of abnormal discharge of the capacitor to be detected based on the characteristic response of the input signal and the activation status of the neuron group.

[0091] In this embodiment, the input layer is the first layer of the spiking neuron network, used to receive each pulse sequence and forward the signal. Each input channel corresponds to one pulse sequence, which can completely preserve the amplitude, timing, and frequency characteristics of the signal, providing standardized input for subsequent network processing. The feature extraction layer is the intermediate processing layer of the network, composed of two layers of sparsely connected leaky integral firing neurons. Its main function is to perform multidimensional feature analysis on the input pulse sequence, including amplitude changes, pulse intervals, timing relationships, and cross-channel phase characteristics. It also establishes a mapping relationship between the neuron group and the feature responses of different abnormal discharge types to extract abnormal features from the input signal. The output discrimination layer is the final processing layer of the network, used to classify and determine the abnormal discharge type of the capacitor to be detected based on the activation status of the neuron group in the feature extraction layer and the feature responses of the input signal.

[0092] During implementation, each pulse sequence is input to the input layer according to the channel correspondence. The input layer transmits the signal completely to the feature extraction layer, ensuring that the temporal and amplitude characteristics of the signal are not lost. The leaky integral firing neurons in the feature extraction layer respond to the pulse sequence through membrane potential accumulation and threshold firing mechanisms. The specific connections and weight adjustments of the neuron group enable it to generate characteristic responses to different types of abnormal discharge patterns, thus forming a representation of abnormal pulse patterns within the network. The output discriminant layer analyzes and judges based on the activation distribution of the neuron group in the feature extraction layer, mapping different characteristic responses to specific abnormal discharge types, achieving accurate classification of the capacitor state.

[0093] In one example, the input layer of the spiking neuron network consists of an array of spiking neurons corresponding to each input channel. The acquisition end has four monitoring channels, including two current channels and two voltage channels. Therefore, the input layer contains a total of four multiplied by N neurons, where N is the number of discrete time points after pulse encoding for each channel. The main function of the input layer neurons is to completely forward the pulse signal to subsequent network layers without performing complex calculations, ensuring the complete preservation of the temporal and amplitude characteristics of the input signal. The feature extraction layer consists of two layers of sparsely connected leaky integral firing neurons. Each neuron randomly selects approximately one-tenth to one-fifth of its synapses from the input layer as input. The feature extraction layer neurons learn the temporal pattern of the input pulses through a temporal-dependent plasticity mechanism, enabling the neuron group to form characteristic responses to different types of firing patterns, including isolated pulses, continuous pulse clusters, and complex anomalous pulse patterns such as amplitude steps. The first feature layer contains 200 neurons, and the second feature layer contains 100 neurons. These two layers of feature extraction achieve multi-dimensional analysis and representation of the input signal. The output discrimination layer contains nine neurons, each corresponding to a specific type of abnormal discharge, including self-healing discharge, air gap discharge, and surface discharge. The output layer employs a threshold determination mechanism: when the number of pulses from a specific neuron group exceeds a threshold within a set time window, it is determined to be an abnormal event. This enables accurate identification and classification of abnormal capacitor discharge types, providing a reliable basis for health status assessment.

[0094] It is understandable that the execution method of using the input layer for signal reception and forwarding, the feature extraction layer for extracting abnormal discharge feature responses, and the output discrimination layer for classification and judgment can ensure the network's sensitivity and recognition accuracy to weak, complex, and multi-channel abnormal discharge pulses. This allows the timing and amplitude features of each pulse sequence to be fully utilized, enhancing the network's adaptive learning ability. As a result, comprehensive monitoring of abnormal capacitor behavior can be achieved, providing a reliable basis for health status assessment and improving the accuracy and safety of operation monitoring.

[0095] In one embodiment, the output discrimination layer is specifically used to determine the target neuron group activated by the feature response of the input signal, and to take the abnormal discharge type corresponding to the target neuron group as the abnormal discharge type of the capacitor to be detected.

[0096] In this embodiment, the output discriminant layer is the final processing layer of the spiking neuron network, used to classify and determine the pulse signals output by the feature extraction layer. The output discriminant layer contains several target neuron groups, each corresponding to a specific type of abnormal discharge, such as self-healing discharge, air gap discharge, and surface discharge. Feature responses refer to the activation patterns of neurons in the feature extraction layer after receiving an input signal, including the number, timing, and population distribution of pulses. These responses reflect the characteristic information of abnormal discharges in the input pulse sequence. The target neuron group is a set of neurons in the output discriminant layer corresponding to a specific type of abnormal discharge; its activation status is used to determine the category of abnormal discharge represented by the input signal.

[0097] During implementation, the input signal is processed by the feature extraction layer, and the feature response is then passed to the output discriminant layer. The output discriminant layer determines which neuron groups are effectively activated by detecting the number of pulses and activation patterns of each target neuron group within a set time window. The network maps the activated target neuron groups to their corresponding abnormal discharge types, outputting the type as the abnormal discharge type of the capacitor to be detected, thereby completing the classification and determination of the abnormal behavior of the capacitor.

[0098] In one example, output layer pulses quantify anomalous features through pulse frequency encoding. The anomalous amplitude is represented by the output layer pulse frequency; the greater the discharge intensity, the higher the frequency of pulses emitted by the output neurons, thus reflecting the amplitude characteristics of the anomalous signal. The duration of the anomalous event is characterized by the pulse cluster length; longer discharges correspond to longer pulse clusters, enabling the network to reflect the sustained nature of the anomalous event. Anomalous categories are distinguished by the activation of different output neuron groups; different types of anomalous discharges activate different target neuron groups. Self-healing discharge corresponds to the activation of output layer neurons 1 to 3, manifesting as high-frequency short pulse clusters; the neuron responds quickly but with weak amplitude. Air gap discharge corresponds to the activation of output layer neurons 4 to 6, presenting stable medium-frequency pulses; the neuron output rhythm is uniform, reflecting continuous and regular discharge characteristics. Surface discharge corresponds to the activation of output layer neurons 7 to 9; the pulses are sparse but with large amplitudes; the neuron output is sparse and strong, effectively characterizing the strong transient characteristics of surface discharge. Through this pulse frequency coding method, the network can effectively quantify the features of abnormal amplitude, duration and category, enabling refined identification and classification of abnormal capacitor discharge, and improving the accuracy and reliability of health status monitoring.

[0099] It can be understood that the output discriminant layer is used to determine the target neuron group activated by the feature response of the input signal, and to take the abnormal discharge type corresponding to the neuron group as the abnormal discharge type of the capacitor to be detected. This allows the multidimensional pulse patterns captured by the feature extraction layer to be directly mapped to specific abnormal event categories, achieving an effective conversion from input signal to abnormal type. Through this processing method, the network can accurately distinguish different types of abnormal discharges, maintain sensitivity to weak and complex pulse features, and quantify the activation patterns of neuron groups into abnormal type outputs, improving the accuracy and reliability of capacitor health status assessment and achieving comprehensive monitoring of abnormal behavior.

[0100] The capacitor abnormal discharge detection device provided in the embodiments of this application is described below. The capacitor abnormal discharge detection device described below can be referred to in correspondence with the capacitor abnormal discharge detection method described above. Figure 2 As shown, this application provides a capacitor abnormal discharge detection device, the device comprising:

[0101] The pulse sequence determination module 201 is used to acquire multiple voltages and multiple currents of the capacitor under test, and to perform pulse encoding on each voltage and current to obtain each pulse sequence.

[0102] The spiking neuron network determination module 202 is used to determine the spiking neuron network. The number of input channels of the spiking neuron network is the same as the number of pulse sequences. The spiking neuron network establishes a connection between a specific group of neurons and the characteristic response of abnormal discharge.

[0103] The abnormal discharge type determination module 203 is used to input the pulse sequences of each channel into the spiking neuron network to obtain the abnormal discharge type of the capacitor to be detected.

[0104] In one embodiment, the pulse sequence determination module 201 includes:

[0105] The pulse sequence determination unit is used to perform time encoding and frequency encoding on each voltage and current channel, and keep the timestamp synchronized to obtain the pulse sequence of each voltage channel and the pulse sequence of each current channel.

[0106] In one embodiment, the time encoding formula is:

[0107]

[0108] The formula for frequency coding is:

[0109]

[0110] in, Indicates the pulse trigger time. Indicates the maximum trigger time. This represents the original signal that was collected. Indicates the minimum amplitude of the signal. Indicates the maximum amplitude of the signal. Indicates the frequency of the emitted pulse. Indicates the minimum distribution frequency. This indicates the highest frequency of distribution.

[0111] In one embodiment, the dynamic equation of the membrane potential of a spiking neuron network is:

[0112]

[0113] in, Represents the membrane potential time constant. Indicates the resting potential. This represents the membrane potential of the i-th neuron at time t. Indicates neuronal membrane resistance. This represents the input current of the i-th unit of the neuron; when When the voltage is greater than or equal to the threshold voltage, the neuron immediately sends a pulse and resets the membrane potential to 0, entering the refractory period.

[0114] In one embodiment, the formula for the input current in the membrane potential dynamic equation of a spiking neuron network is:

[0115]

[0116]

[0117] in, This represents the total input current received by the j-th neuron at time t. This represents the synaptic weights from the i-th input neuron to the j-th neuron. This represents the pulse sequence of the i-th input neuron. Indicates synaptic transmission delay, Indicates the time difference in distribution. The amplitude constant representing long-term enhancement. The amplitude constant representing long-term suppression. The time constant representing long-term enhancement. The time constant represents long-term suppression.

[0118] In one embodiment, the spiking neural network includes an input layer, a feature extraction layer, and an output discriminant layer;

[0119] The number of input channels in the input layer is the same as the number of pulse sequences. The input layer is used to receive each pulse sequence and forward the signal.

[0120] The feature extraction layer contains two layers of sparsely connected leaky integral firing neurons. The feature extraction layer is used to extract the feature responses of abnormal discharges from the input signal and establish a correspondence between specific neuron groups and the feature responses of various abnormal discharges.

[0121] The output discriminant layer is used to determine the type of abnormal discharge of the capacitor to be detected based on the characteristic response of the input signal and the activation status of the neuron group.

[0122] In one embodiment, the output discrimination layer is specifically used to determine the target neuron group activated by the feature response of the input signal, and to take the abnormal discharge type corresponding to the target neuron group as the abnormal discharge type of the capacitor to be detected.

[0123] In one embodiment, this application also provides a storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the capacitor abnormal discharge detection method as described in any of the above embodiments.

[0124] In one embodiment, this application also provides a computer device storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the capacitor abnormal discharge detection method as described in any of the above embodiments.

[0125] Indicatively, such as Figure 3 As shown, Figure 3 This is a schematic diagram of the internal structure of a computer device 300 provided in an embodiment of this application. The computer device 300 can be provided as a server. (Refer to...) Figure 3 The computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions, such as application programs, that can be executed by the processing component 302. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 302 is configured to execute instructions to perform the capacitor abnormal discharge detection method of any of the above embodiments.

[0126] The computer device 300 may also include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate on an operating system stored in memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0127] Those skilled in the art will understand that Figure 3 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 computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0128] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the related listed items.

[0129] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0130] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method of detecting abnormal discharge of a capacitor, characterized by, The method comprises: Collecting multiple voltage and multiple current of the capacitor to be detected, and pulse coding each voltage and each current to obtain each pulse sequence; Determining a pulse neuron network, the number of input channels of the pulse neuron network being the same as the number of the pulse sequences, the pulse neuron network establishing a connection between a specific neuron group and a characteristic response of abnormal discharge; Inputting each pulse sequence into the pulse neuron network to obtain the abnormal discharge type of the capacitor to be detected.

2. The capacitor abnormal discharge detection method according to claim 1, characterized by, The step of pulse coding each voltage and each current to obtain each pulse sequence comprises: Time coding and frequency coding each voltage and each current, and keeping the time stamp synchronized to obtain the pulse sequence of each voltage and the pulse sequence of each current.

3. The capacitor abnormal discharge detection method according to claim 2, characterized by, The formula of time coding is: The formula of frequency coding is: wherein denotes the pulse trigger time, denotes the maximum trigger time, denotes the raw signal acquired, denotes the minimum amplitude of the signal, denotes the maximum amplitude of the signal, denotes the frequency of the pulses emitted, denotes the minimum emission frequency, denotes the maximum emission frequency.

4. The capacitor abnormal discharge detection method according to claim 1, characterized by, The membrane potential dynamic equation of the pulse neuron network is: wherein, represents the membrane potential time constant, represents the resting potential, represents the membrane potential of the i-th neuron at time t, represents the membrane resistance of the neuron, represents the input current of the i-th unit of the neuron; when greater than or equal to the threshold voltage, the neuron immediately sends a pulse and resets the membrane potential to 0, entering the refractory period.

5. The capacitor abnormal discharge detection method according to claim 4, characterized by, The formula of input current in the membrane potential dynamic equation of the pulse neuron network is: wherein, denotes the total input current received by the jth neuron at time t, denotes the synaptic weight from the ith input neuron to the jth neuron, denotes the spike train of the ith input neuron, denotes the synaptic transmission delay, denotes the difference in firing times, denotes the amplitude constant of long-term potentiation, denotes the amplitude constant of long-term depression, denotes the time constant of long-term potentiation, denotes the time constant of long-term depression.

6. The capacitor abnormal discharge detection method according to claim 1, characterized by, The pulse neuron network comprises an input layer, a feature extraction layer and an output discrimination layer; The number of input channels of the input layer is the same as the number of the pulse sequences, and the input layer is used for receiving each pulse sequence and forwarding signals; The feature extraction layer comprises two layers of sparsely connected leaky integrate-and-fire neurons, and is used for extracting the characteristic response of abnormal discharge from the input signal and establishing a corresponding relationship between a specific neuron group and each type of characteristic response of abnormal discharge; The output discrimination layer is used for determining the abnormal discharge type of the capacitor to be detected according to the characteristic response of the input signal and the neuron group activation.

7. The capacitor abnormal discharge detection method according to claim 6, wherein The output discrimination layer is specifically used for determining the target neuron group activated by the characteristic response of the input signal, and taking the abnormal discharge type corresponding to the target neuron group as the abnormal discharge type of the capacitor to be detected.

8. A capacitor abnormal discharge detection device characterized by comprising: The device comprises: A pulse sequence determination module, which is used for collecting multiple voltage and multiple current of the capacitor to be detected, and pulse coding each voltage and each current to obtain each pulse sequence; A pulse neuron network determination module, which is used for determining a pulse neuron network, the number of input channels of the pulse neuron network being the same as the number of the pulse sequences, the pulse neuron network establishing a connection between a specific neuron group and a characteristic response of abnormal discharge; An abnormal discharge type determination module, which is used for inputting each pulse sequence into the pulse neuron network to obtain the abnormal discharge type of the capacitor to be detected.

9. A storage medium characterized by: The storage medium stores computer readable instructions, and the computer readable instructions are executed by one or more processors to make the one or more processors execute the steps of the capacitor abnormal discharge detection method in any one of claims 1 to 7.

10. A computer device, comprising: Comprise: One or more processors, and a memory; The memory stores computer readable instructions, and the computer readable instructions are executed by the one or more processors to execute the steps of the capacitor abnormal discharge detection method in any one of claims 1 to 7.

Images