Switch cabinet partial discharge real-time monitoring method based on signal transmission

Abstract

The invention discloses a switch cabinet partial discharge real-time monitoring method based on signal transmission, and relates to the technical field of discharge monitoring, and the method comprises the steps: arranging a UHF antenna and an environment monitoring sensor at a key part in a switch cabinet, and collecting a partial discharge signal; a triggering condition is set to obtain a high-speed sampling waveform, and the signal-to-noise ratio is increased through wavelet packet decomposition and dynamic threshold denoising; a reinforcement learning algorithm is combined to adaptively adjust denoising parameters; performing classification reasoning on the de-noised signals by using a pulse neural network, and extracting confidence coefficients of multiple types of partial discharge modes; and determining a classification result or triggering a rollback and degradation mode based on a multi-dimensional index. The method has the capabilities of high-precision signal denoising, adaptive parameter optimization and multi-mode classification, can realize online identification and real-time response of partial discharge types, and is suitable for monitoring the state of the switch cabinet in a complex electromagnetic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of discharge monitoring, in particular to a real-time monitoring method for partial discharge of a switch cabinet based on signal transmission. Background Art

[0002] Switchgear is a crucial piece of equipment for power distribution, conversion, and protection in power systems. During long-term operation, it is prone to partial discharge (PD) defects. PD is an early sign of insulation defects in electrical equipment. If not detected and addressed promptly, the discharge defect can gradually develop along the insulation interface, ultimately leading to insulation damage or even equipment destruction, posing a serious threat to the safe and stable operation of the power system. Therefore, real-time monitoring of PD within switchgear has become a hot topic in research and application.

[0003] Existing partial discharge monitoring technologies primarily rely on ultra-high frequency (UHF) signals, ultrasound, and ultrasonic current transformers. UHF monitoring technology is widely used due to its high sensitivity to discharge signals and its resistance to electromagnetic interference. Traditional UHF monitoring often uses antenna arrays placed inside switchgear to identify discharge activity by capturing electromagnetic pulse signals in the 300MHz–3GHz range.

[0004] However, existing monitoring methods often have some shortcomings during implementation. First, conventional solutions often rely on fixed noise thresholds or empirical thresholds for sampling triggering, and are unable to adaptively adjust to the dynamic changes of transient interference inside the switchgear and environmental electromagnetic noise. This often leads to missed detections or false detections in high-noise environments or during switch operation. Second, most existing solutions directly perform simple filtering or fixed-threshold denoising after sampling, and fail to fully utilize multi-resolution analysis methods to separate signals from noise. Especially in high-noise scenarios, simple frequency band filtering or envelope detection cannot meet the requirements for extracting weak discharge signals, resulting in insufficient signal-to-noise ratio and affecting subsequent fault identification. Third, traditional discharge pattern recognition methods based on manual feature extraction or shallow neural networks require manual feature selection or long-term training and are highly sensitive to the quality of signal preprocessing. Generally, they are unable to quickly and accurately classify discharge types (such as surface discharge, creeping discharge, air gap discharge, and internal discharge). In particular, when the signal-to-noise ratio is low or the environment is highly variable, the recognition error increases significantly. Based on the above problems, there is an urgent need for a classification algorithm that can combine multi-channel acquisition, dynamic threshold adaptive denoising, and excellent real-time and anti-interference capabilities for timing pulse signals, so as to improve the accuracy and reliability of partial discharge monitoring of switchgear. Summary of the Invention

[0005] In view of the problems that the existing technologies in discharge monitoring mostly rely on fixed or empirical thresholds for sampling triggering, and are unable to adaptively adjust to the transient interference and environmental noise inside the switch cabinet, resulting in frequent missed detections or false detections; simple filtering or fixed threshold denoising is directly performed after sampling, and the multi-resolution analysis method is not fully utilized to separate the signal and noise; at the same time, the discharge pattern recognition method requires manual feature selection or long-term training, and is highly sensitive to the quality of signal preprocessing. Therefore, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to provide a real-time monitoring method for partial discharge of a switch cabinet based on signal transmission.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, an embodiment of the present invention provides a real-time monitoring method for partial discharge of a switch cabinet based on signal transmission, including: collecting original data, and when the collected signal exceeds a preset trigger threshold, recording the original time domain data at that moment and caching it locally; performing multi-layer wavelet packet decomposition on the cached original time domain data, denoising the trigger threshold, reconstructing the denoised time domain signal, and calculating the signal-to-noise ratio of the denoised signal; comparing the signal-to-noise ratio of the denoised signal with the preset threshold and making a strategy decision to obtain a denoised time domain signal that meets the conditions; generating a sparse pulse matrix for the denoised time domain signal through time-to-first pulse coding, inputting it into a pulse neural network and iteratively updating the neuron membrane potential and triggering pulses through the LIF model, and outputting the number of pulses corresponding to four types of partial discharge; counting the number of pulses of the output layer neurons in the entire time window, calculating the confidence of each type and comparing it with the preset threshold and multi-dimensional indicators and making a strategy decision, and finally outputting a usable classification result or temporary result.

[0009] As a preferred solution of the real-time monitoring method for partial discharge of a switchgear based on signal transmission described in the present invention, obtaining a denoised time domain signal includes: first applying a threshold function to each frequency band coefficient after wavelet packet decomposition to remove or weaken the noise-containing part so that the signal has a higher signal-to-noise ratio in the frequency domain; retaining the original structure of each frequency band coefficient after threshold processing and keeping the wavelet packet tree structure unchanged for reconstruction; and using the retained denoising coefficients to restore the processed frequency domain information back to the time domain through an inverse wavelet packet transform to obtain a denoised signal.

[0010] As a preferred solution of the real-time monitoring method for partial discharge of a switch cabinet based on signal transmission described in the present invention, calculating the signal-to-noise ratio of the denoised signal includes: extracting the original signal and the denoised signal data sequence respectively for subsequent comparative analysis; obtaining an error signal sequence by performing point-by-point difference between the original signal and the denoised signal, and calculating its overall energy or average power to represent the residual noise intensity; comparing the energy of the denoised signal with the energy of the error signal to measure the proportion of the effective signal to the residual noise, thereby obtaining a signal-to-noise ratio result.

[0011] As a preferred solution of the real-time monitoring method for partial discharge of switchgear based on signal transmission described in the present invention, the signal-to-noise ratio of the denoised signal is compared with a preset threshold and a strategy decision is made, including: if the requirements are met, proceed to the next step; otherwise, call the reinforcement learning strategy network according to the current state to adjust the denoising threshold and re-denoise, for a maximum of three iterations; if the requirements are still not met, switch to the suboptimal denoising mode and report the relevant data to the cloud.

[0012] As a preferred solution of the real-time monitoring method for partial discharge of switchgear based on signal transmission described in the present invention, generating a sparse pulse matrix includes: within the time dimension, collecting neuronal pulse responses according to a fixed time interval or event triggering mechanism to generate an initial pulse data matrix; applying a sparsification strategy to the initial pulse matrix, such as threshold screening or random discarding mechanism, to retain only pulse sites with high activation intensity or significant changes; mapping the screened sparse pulse signal to a predefined coding structure to form a sparse pulse matrix suitable for downstream recognition or control tasks.

[0013] As a preferred solution of the real-time monitoring method for partial discharge of switch cabinets based on signal transmission described in the present invention, updating the neuron membrane potential and triggering pulses includes: the neuron receives weighted inputs from the previous layer or input layer, and accumulates these signals into the current membrane potential; if the membrane potential reaches or exceeds the set threshold, the neuron triggers a pulse signal and records the release time; after the pulse is released, the membrane potential is cleared or reset to the initial state, and enters the next accumulation cycle.

[0014] As a preferred solution of the real-time monitoring method for partial discharge of switch cabinets based on signal transmission described in the present invention, the calculation of confidence includes: traversing the pulse counter of each category in the output layer of the neural network, recording the number of pulses generated by each category within a set time window; comparing the number of pulses in each category with the sum of pulses of all categories to obtain the confidence value corresponding to each category as the probability basis for it being judged as the correct category; sorting according to the confidence value, selecting the category with the highest confidence as the recognition output result of the current neural network, and using it for subsequent decision-making or feedback adjustment.

[0015] In the second aspect, in order to further solve the problems existing in discharge monitoring, the present invention provides a real-time monitoring system for partial discharge of switch cabinets based on signal transmission, which includes: a signal acquisition module for arranging UHF antennas at key positions of the switch cabinet to receive electromagnetic pulse signals generated by partial discharge; an adaptive preprocessing module for setting a trigger threshold, triggering sampling of the original waveform, performing multi-layer decomposition of the signal using wavelet packet decomposition, adopting a dynamic threshold soft denoising strategy and calculating the signal-to-noise ratio; a reinforcement learning threshold optimization module for, when the signal-to-noise ratio is lower than the set threshold, sending a signal to the system based on the current state through a deep DQN strategy network. The denoising parameters are adjusted by quantitative selection actions, and denoising is iteratively performed until the preset signal-to-noise ratio requirements are met or the suboptimal denoising strategy is triggered; the pulse neural network inference module is used to pulse encode the denoised signal and input it into the multi-layer pulse convolutional neural network, and output the confidence of four types of discharge modes: surface discharge, surface discharge, air gap discharge and internal discharge; the decision and output module is used to count the number of pulses in the SNN output layer and calculate the classification confidence, and perform multi-dimensional evaluation based on indicators such as signal-to-noise ratio and processing delay. If all are met, the final classification result is output; otherwise, the rollback mechanism or degradation mode is triggered, and the original data is uploaded to the cloud for analysis.

[0016] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the real-time monitoring method for partial discharge of a switch cabinet based on signal transmission as described in the first aspect of the present invention is implemented.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for real-time monitoring of partial discharge of a switch cabinet based on signal transmission as described in the first aspect of the present invention is implemented.

[0018] The beneficial effects of the present invention are:

[0019] 1. The present invention adopts adaptive wavelet packet decomposition combined with dynamic threshold denoising algorithm, which can automatically calculate the threshold and perform soft threshold processing according to the on-site noise level, effectively suppressing high-frequency environmental interference, significantly improving the signal-to-noise ratio, and thus ensuring high-quality input of subsequent fault signals;

[0020] 2. This invention introduces a reinforcement learning decision-making mechanism based on the DQN policy network, which can adjust the denoising threshold in real time according to multiple dimensions such as the current environmental noise estimation, the system's remaining energy budget, and the denoising delay, ensuring stable and efficient denoising effects under different noise conditions, reducing false detection and missed detection rates;

[0021] 3. The present invention combines pulse neural networks for time series coding and classification reasoning. While retaining the signal timing characteristics, it uses multi-layer pulse convolution and a fully connected structure to achieve rapid identification of various types of partial discharges, such as surface discharge, creeping discharge, air gap discharge, and internal discharge. The classification speed is fast and the robustness is strong, meeting the low-latency requirements for online real-time monitoring.

[0022] 4. The present invention uses output layer pulse statistics and a multi-dimensional indicator decision-making mechanism, comprehensively considers indicators such as signal-to-noise ratio, classification confidence, and overall delay, and rolls back the classification results multiple times or enters a degradation mode to ensure the accuracy of the judgment results and the reliability of the system, effectively avoiding misjudgment caused by the failure of a single indicator. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0024] Figure 1 This is a flow chart for implementing the present invention in Example 1.

[0025] Figure 2 This is a flowchart of the confidence decision in the present invention in Example 1.

[0026] Figure 3 This is a flowchart of the multi-dimensional decision-making of the present invention in Example 1. DETAILED DESCRIPTION

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0030] Example 1

[0031] Reference Figure 1 - Figure 3 , which is the first embodiment of the present invention, provides a real-time monitoring method for partial discharge of a switch cabinet based on signal transmission, comprising the following steps:

[0032] Step S1: Hardware environment and raw data acquisition, which includes the following sub-steps:

[0033] S1-1: Hardware environment, which includes sensor layout and front-end construction. Specifically, at least one pair of UHF antennas (frequency band 300MHz-3GHz) is arranged inside the target switch cabinet and key locations (such as busbar connections, insulator surfaces, etc.) to receive electromagnetic pulse signals generated by partial discharge; at the same time, environmental monitoring sensors are installed in the cabinet: temperature sensor (measuring range -40℃~85℃; accuracy ±0.5℃), humidity sensor (detection range 0~100%RH; accuracy ±3%RH) and switch action acquisition module (high-voltage switch contact status digital input); each UHF antenna is directly connected to the RF front-end of the edge microprocessor via a 50Ω coaxial cable, and the RF amplifier gain G amp , bandwidth 300MHz-3GHz, which can ensure the capture of high-frequency signals above 300MHz and send them to the back-end ADC (sampling rate f s =200MS / s, resolution 14bit).

[0034] S1-2: Set the sampling trigger condition. Specifically, the noise waveform {x n [m]}, calculate the baseline mean and the noise standard deviation σ n ; Then set the trigger threshold, when the signal amplitude captured at any time A high-speed sampling is triggered when the sampling length is T win =1ms, total acquisition N=f s ×T win =200MS / s×1ms=2×10 5 Point time domain waveform {x n [n]},n=0,...,N-1(where the trigger coefficient k t It can be 3 to 5, set manually);

[0035] The above parameters can be expressed by the following formula:

[0036]

[0037] S1-3: When sampling is triggered, the current corresponding temperature T is read synchronously amb 、Humidity H relAnd the switch action signal state (digital signal: 0 or 1) within 0.1s before this sampling moment is recorded as the environment vector e = [T amb ,H rel , switch status]; the original time domain data collected this time Together with the environment vector e and the trigger timestamp t0, they are cached in the local temporary memory for subsequent preprocessing.

[0038] Step S2: Adaptive wavelet packet decomposition and dynamic threshold denoising, which includes the following sub-steps:

[0039] S2-1: Select the wavelet packet decomposition parameters. This includes determining the wavelet basis as Daubechies4 (db4), which has good time-frequency resolution for high-frequency mutation signals; setting the decomposition level L = 4, and converting the 1ms original waveform {x n [n]} is decomposed layer by layer, and finally 2 L = 16 sub-band coefficients {W i,j [k]}, where i = 1, 2, 3, 4, j = 0, ... 2 i -1, k is the length of the subband; the unit of the wavelet packet decomposition process is kept consistent with the input signal (V), and the noise standard deviation σ corresponding to each subband is obtained n,i Used for threshold calculation.

[0040] S2-2: Threshold initialization and soft threshold denoising, which includes estimating σ according to each subband noise n,i , initialize the threshold of layer i Then the j-th subband coefficient W of the i-th layer is i,j [k] Apply soft thresholding and output the thresholded coefficient W` i,j [k], and the denoised signal is obtained by wavelet packet reconstruction

[0041] The above parameters can be expressed by the following formula:

[0042]

[0043] Where, α i ∈[1.5,3.0], set manually;

[0044]

[0045]

[0046] Where, is the wavelet packet reconstruction basis function, and the unit is dimensionless.

[0047] S2-3: Calculate the signal-to-noise ratio (SNR), which can be expressed by the following formula:

[0048]

[0049] Where, P sig is the signal power after denoising; P noise is the residual noise power after denoising;

[0050] The above parameters can be expressed by the following formula:

[0051]

[0052] Where x pd [n] is a "pure PD signal" sample annotated by experts or extracted by a high-precision cloud-based model. If annotated samples are not available on site, the noise subband energy can be estimated using wavelet decomposition of "continuous non-trigger segments" as an alternative.

[0053] Step S3: Determine and make a decision on the signal-to-noise ratio (SNR) after denoising, including the following sub-steps:

[0054] S3-1: Set threshold SNR min =15dB. If the calculated value SNR in the above step S2 is ≥15dB, it is determined as "denoising is successful" and enters the SNN reasoning stage; if SNR<15dB, it is determined as "denoising is insufficient" and enters the reinforcement learning decision stage.

[0055] S3-2: Reinforcement learning decision, which includes constructing the current state vector s; calculating each action a={ΔT i}Q value, select the best action ΔT i Update threshold T new ; With the new threshold T new Return to steps S2-2 and S2-3 and re-calculate the new signal-to-noise ratio SNR new If the new SNR is ≥ 15dB, it will enter the SNN reasoning phase; otherwise, the iteration counter will be accumulated. If the number of iterations is less than or equal to 3, DQN will be called to update the threshold and re-denoise. If the number of iterations exceeds 3 and still does not reach 15dB, it will be judged as a "high noise event", and the local denoising loop will be exited and the "suboptimal denoising mode" will be entered. The current denoising failed will be Environmental parameters and other data are reported to the cloud for subsequent model training or manual analysis. After local preprocessing is completed, the process ends.

[0056] For example, the suboptimal noise reduction mode may be to use only the empirical threshold (e.g., α i =2.5,σ n,i ) for fast peak detection;

[0057] The above parameters can be expressed by the following formula:

[0058]

[0059] Where, SNR prev is the signal-to-noise ratio after the last denoising; T prev =[T1, T2, T3, T4] is the threshold set used last time; E rem The remaining energy budget of the system can be estimated by the power and power consumption model of the edge microprocessor; L prev is the last denoising delay; Estimate the standard deviation of the current ambient noise;

[0060] For example, the value of each action is calculated by the DQN policy network, which includes the action being defined as an increase or decrease ΔT of a certain layer threshold i = ±0.1σ n,i ,i=1,2,3,4, select the best action ΔT i Updated threshold T new =T prev +ΔT, where ΔT=[ΔT1, ΔT2, ΔT3, ΔT4].

[0061] Step S4: Inferring confidence through a spiking neural network (SNN), including the following sub-steps:

[0062] S4-1: The denoised signal Perform "time to first pulse" encoding and get t s (n), and t s (n) Discretize to time slot Δt = 5ns accuracy to generate sparse pulse matrix S[m,n];

[0063] Specifically, the above parameters can be expressed by the following formula:

[0064]

[0065] Where, is the minimum value of the denoised signal; is the maximum value of the denoised signal, τ0 = 0.1ms;

[0066]

[0067] Where m = 0, 1, ..., M max -1,

[0068] S4-2: SNN network structure and pulse processing, which includes a network input layer, which consists of N pulse input neurons and is responsible for receiving the sparse matrix S[m,n]; the first and second pulse convolution layers: (convolution kernel 5×5, stride 1, padding 0) 16 channels each; followed by three layers of pulse fully connected layers (128→64→4 neurons); the output of the fourth layer is used for classification (4 types of PD include: surface discharge, surface discharge, air gap discharge, internal discharge); the LIF model is used to describe neural dynamics; neuron execution and timing update, where the timing discrete step size Δt SNN = 0.1ms, the entire 1ms (or 0.8ms after rollback) time window requires K max = 10 iterations;

[0069] Specifically, the above-mentioned LIF model can be used to describe neural dynamics through the following formula:

[0070]

[0071] Where V i (t) is the membrane potential of neuron No. i; V rest , is the resting potential, take -0.065V; R m , is the membrane resistance, take 100MΩ; C m is the membrane capacitance, take 100pF; I i (t) is the input current, which is generated by the pulse of the previous layer after being weighted by the synaptic weight;

[0072] It should be noted that when V i (t) reaches the threshold V th =-0.050V, the neuron emits a pulse at the current moment and V i (t) Reset to V reset =-0.065V, enters the absolute refractory period τ ref =2ms;

[0073] Among them, each iteration time k (corresponding to the real time t = kΔt SNN ) Perform the following sub-steps A1-A4:

[0074] A1: Check whether there is a pulse S[m,n] arriving at the input layer;

[0075] A2: Synaptic weighting:

[0076]

[0077] Where, is the weight between the i-th neuron in layer l and the j-th neuron in layer l-1, obtained through line training; Issue a flag for the pulse;

[0078] A3: Membrane potential update (Euler discretization):

[0079]

[0080] A4: Pulse trigger: If but and At the same time, it enters the refractory period τ ref ;otherwise

[0081] Among them, after the 10th iteration, the accumulated pulse numbers of the four neurons in the output layer {N1, N2, N3, N4} are used to calculate the confidence of each category.

[0082] Step S5: Output layer pulse statistics and confidence calculation, including the following sub-steps:

[0083] S5-1: Count the total number of pulses N of the four output neurons in the entire time window i , total pulse

[0084] S5-2: Define the confidence level P for each category i for;

[0085]

[0086] Get the maximum confidence value:

[0087]

[0088] Corresponding categories:

[0089] S5-3: Setting the confidence threshold θ conf =0.80, refer to Figure 2 , make confidence judgment;

[0090] For example, if conf≥θ conf , then the local confirmed classification result is "class i * ” (e.g. 1 = surface discharge, 2 = creeping discharge, 3 = air gap discharge, 4 = internal discharge); if conf < θ conf , then enter multi-dimensional decision-making;

[0091] S5-4: Multi-dimensional decision making, refer to Figure 3 , which includes the following sub-steps D1-D2;

[0092] D1: Construct indicator vector:

[0093] M=[SNR,conf,L tl];

[0094] Where, SNR is the signal-to-noise ratio after denoising; conf is the confidence of this classification; L tl is the total delay from sampling to denoising to the end of inference, obtained by the edge microprocessor timer;

[0095] D2: Set the indicator's allowable range:

[0096]

[0097] For example, the specific decision-making situations are as follows:

[0098] Case 1: If all indicators are within the allowable range, the classification result is confirmed and the process ends;

[0099] Case 2: If only a few indicators (such as conf < θ conf ) exceeds the standard, but the remaining indicators (SNR, L tl ) are satisfied, the rollback strategy is adopted, as follows:

[0100] Rollback strategy 1: Modify the SNN input preprocessing parameters without changing the denoising threshold, such as changing the STFT window length from 1ms to 0.8ms;

[0101] Rollback strategy 2: If conf < θ conf , then it is allowed to re-execute the denoising-inference closed loop and return to step S2 to resubmit the denoising;

[0102] It should be noted that the maximum number of rollback iterations is 2 (including rollback strategy 1 and rollback strategy 2). If the limit is exceeded and the conditions are still not met, the "manual confirmation mode" will be triggered;

[0103] Case 3: If SNR < 15dB and conf < θ conf , then first roll back to step S2 and resubmit denoising;

[0104] Case 4: If at any time L tl >L max , it immediately enters "degradation mode" (using a simple peak detection algorithm to quickly output temporary results) and uploads the original data for manual judgment;

[0105] S5-5: When a pulse event completes the denoising and classification iterations, and all indicators are confirmed to be met through "multi-dimensional decision-making", the classification result PD is output. 类型 =i (i=1, 2, 3, 4, representing 1=surface discharge, 2=creeping discharge, 3=air gap discharge, 4=internal discharge).

[0106] In summary, the present invention realizes the synchronous acquisition of partial discharge signals and environmental parameters by arranging multiple sensors and UHF antennas inside the switch cabinet, and effectively suppresses high-frequency interference and improves the signal-to-noise ratio by combining adaptive wavelet packet decomposition and dynamic threshold denoising algorithm. On this basis, the reinforcement learning decision mechanism based on the DQN strategy network can adjust the denoising threshold in real time according to the environmental noise and system status to ensure the denoising effect under different working conditions. Subsequently, the pulse neural network is used to perform multi-layer convolution and coding inference on the time series signal to realize rapid identification of various partial discharge types. Finally, the classification results are comprehensively judged through the multi-dimensional indicator decision and rollback or degradation mechanism of the output layer, which further improves the detection accuracy and system reliability.

[0107] Example 2

[0108] Embodiment 2 is the second embodiment of the present invention. This embodiment differs from the first embodiment in that it further provides a switch cabinet partial discharge real-time monitoring system based on signal transmission, including:

[0109] The signal acquisition module is used to place UHF antennas at key locations in the switchgear to receive electromagnetic pulse signals generated by partial discharges. The RF front-end amplifies and bandpass filters the signals and transmits them to the ADC for high-frequency sampling. It also collects temperature, humidity, and switch status signals to form an environmental vector.

[0110] Adaptive preprocessing module, used to set the trigger threshold, trigger sampling of the original waveform, perform multi-layer decomposition of the signal using wavelet packet decomposition, adopt dynamic threshold soft denoising strategy and calculate the signal-to-noise ratio;

[0111] The reinforcement learning threshold optimization module is used to adjust the denoising parameters based on the current state vector through the deep DQN policy network when the signal-to-noise ratio falls below the set threshold, and iteratively perform denoising until the preset signal-to-noise ratio requirement is met or the suboptimal denoising strategy is triggered;

[0112] The pulse neural network inference module is used to pulse encode the denoised signal and input it into a multi-layer pulse convolutional neural network to output the confidence level of four types of discharge modes: surface discharge, creeping discharge, air gap discharge, and internal discharge;

[0113] The decision and output module is used to count the number of pulses in the SNN output layer and calculate the classification confidence. It combines indicators such as signal-to-noise ratio and processing delay for multi-dimensional evaluation. If all the indicators are met, the final classification result is output; otherwise, the rollback mechanism or degradation mode is triggered, and the original data is uploaded to the cloud for analysis.

[0114] This embodiment also provides a computer device, which is applicable to the case of a real-time monitoring method for partial discharge of a switch cabinet based on signal transmission, and includes: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the real-time monitoring method for partial discharge of a switch cabinet based on signal transmission proposed in the above embodiment.

[0115] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0116] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for real-time monitoring of partial discharge of a switch cabinet based on signal transmission as proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0117] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A real-time monitoring method for partial discharge of a switch cabinet based on signal transmission, characterized by: include: S1, collects raw data, and when the collected signal exceeds the preset trigger threshold, records the raw time domain data at that moment and caches it locally; S2, performing multi-layer wavelet packet decomposition on the cached original time domain data, denoising the trigger threshold, reconstructing the denoised time domain signal, and calculating the signal-to-noise ratio of the denoised signal; S3, comparing the signal-to-noise ratio of the denoised signal with a preset threshold and making a strategic decision to obtain a denoised time domain signal that meets the conditions; S4, generates a sparse pulse matrix from the denoised time domain signal through time-to-first pulse coding, inputs it into the spiking neural network, and iteratively updates the neuron membrane potential and triggers pulses through the LIF model, outputting the number of pulses corresponding to the four types of partial discharges; S5 counts the number of pulses of the output layer neurons in the entire time window, calculates the confidence of each type, compares it with the preset threshold and multi-dimensional indicators, makes strategic decisions, and finally outputs a usable classification result or temporary result.

2. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, characterized in that: Obtaining the denoised time domain signal includes: first applying a threshold function to the frequency band coefficients after wavelet packet decomposition to remove or weaken the noise-containing parts so that the signal has a higher signal-to-noise ratio in the frequency domain; retaining the original structure of the frequency band coefficients after threshold processing and keeping the wavelet packet tree structure unchanged for reconstruction; using the retained denoising coefficients, the processed frequency domain information is restored back to the time domain through the inverse wavelet packet transform to obtain the denoised signal.

3. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, characterized in that: Calculating the signal-to-noise ratio of the denoised signal includes: extracting the original signal and the denoised signal data series respectively for subsequent comparative analysis; obtaining an error signal series by performing point-by-point subtraction between the original signal and the denoised signal, and calculating its overall energy or average power to represent the residual noise intensity; comparing the energy of the denoised signal with the energy of the error signal to measure the proportion of the effective signal to the residual noise, thereby obtaining the signal-to-noise ratio result.

4. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, wherein: The signal-to-noise ratio of the denoised signal is compared with the preset threshold and a strategy decision is made: if the requirements are met, proceed to the next step; otherwise, call the reinforcement learning strategy network based on the current state to adjust the denoising threshold and re-denoise, for up to three iterations; if the requirements are still not met, switch to the suboptimal denoising mode and report the relevant data to the cloud.

5. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, wherein: Generating a sparse spike matrix includes: within the time dimension, collecting neuronal spike responses according to fixed time intervals or event triggering mechanisms to generate an initial spike data matrix; applying a sparsification strategy to the initial spike matrix, such as threshold screening or random discarding mechanism, to retain only spike sites with high activation intensity or significant changes; mapping the screened sparse spike signal to a predefined coding structure to form a sparse spike matrix suitable for downstream recognition or control tasks.

6. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, characterized in that: Updating the neuron membrane potential and triggering pulses includes: the neuron receives weighted inputs from the previous layer or input layer, and accumulates these signals into the current membrane potential; if the membrane potential reaches or exceeds the set threshold, the neuron triggers a pulse signal and records the time of the pulse; after the pulse is emitted, the membrane potential is cleared or reset to the initial state, entering the next accumulation cycle.

7. The method for real-time monitoring of partial discharge in a switch cabinet based on signal transmission according to claim 1, wherein: The confidence calculation includes: traversing the pulse counter of each category in the output layer of the neural network, recording the number of pulses generated by each category within the set time window; comparing the number of pulses in each category with the sum of pulses of all categories to obtain the confidence value corresponding to each category as the probability basis for it to be judged as the correct category; sorting according to the confidence value, selecting the category with the highest confidence as the recognition output result of the current neural network, and using it for subsequent decision-making or feedback adjustment.

8. A switchgear partial discharge real-time monitoring system based on signal transmission, based on the switchgear partial discharge real-time monitoring method based on signal transmission according to any one of claims 1 to 7, characterized in that: include, Signal acquisition module, used to place UHF antennas at key locations in the switchgear to receive electromagnetic pulse signals generated by partial discharge; Adaptive preprocessing module, used to set the trigger threshold, trigger sampling of the original waveform, perform multi-layer decomposition of the signal using wavelet packet decomposition, adopt dynamic threshold soft denoising strategy and calculate the signal-to-noise ratio; The reinforcement learning threshold optimization module is used to adjust the denoising parameters based on the current state vector through the deep DQN policy network when the signal-to-noise ratio falls below the set threshold, and iteratively perform denoising until the preset signal-to-noise ratio requirement is met or the suboptimal denoising strategy is triggered; The pulse neural network inference module is used to pulse encode the denoised signal and input it into a multi-layer pulse convolutional neural network to output the confidence level of four types of discharge modes: surface discharge, creeping discharge, air gap discharge, and internal discharge; The decision and output module is used to count the number of pulses in the SNN output layer and calculate the classification confidence. It combines indicators such as signal-to-noise ratio and processing delay to perform multi-dimensional evaluation. If all indicators are met, the final classification result is output; Otherwise, the rollback mechanism or degradation mode is triggered, and the original data is uploaded to the cloud for analysis.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the real-time monitoring method for partial discharge of a switch cabinet based on signal transmission according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the real-time monitoring method for partial discharge of a switch cabinet based on signal transmission according to any one of claims 1 to 7 are implemented.

Images