Switch cabinet partial discharge monitoring and early warning method and system, electronic equipment and storage medium

Abstract

The invention provides a partial discharge monitoring and early warning method for a switch cabinet. The method comprises the following steps: acquiring ultrasonic sensing data and transient grounding voltage data through configuration acquisition software, and preprocessing the ultrasonic sensing data and the transient grounding voltage data; inputting the real-time signal peak value into a pre-constructed CNN and LSTM-based partial discharge prediction model, and outputting a combined sequence containing original data and a predicted value; and performing M-K detection on the joint sequence for trend analysis, and performing early warning analysis in combination with the predicted value and an alarm threshold. According to the technical scheme, the CNN-LSTM is utilized to predict the partial discharge data sequence, and then trend analysis is carried out through the M-K inspection method to comprehensively consider two indexes of the amplitude of the partial discharge sequence and the trend parameter so as to realize early warning of the partial discharge of the cable joint. The CNN-LSTM combined method is adopted to carry out partial discharge data sequence prediction, the feature information extraction capability of the CNN and the sensitivity of the LSTM to the time sequence can be effectively integrated, and the trend data prediction effect is improved.

Description

Technical Field

[0001] This invention relates to the field of monitoring and early warning technology, specifically to a method, system, electronic device, and storage medium for monitoring and early warning of partial discharge in switchgear. Technical Background

[0002] Switchgear is the most critical equipment in distribution networks or for direct power supply to large users. It is widely used in power supply and distribution systems, and its failure directly affects the quality of power supply services, leading to direct economic losses. During long-term operation, partial discharge (PD) may occur inside the switchgear due to manufacturing defects, mechanical vibration, environmental aging, and other factors. PD is an early sign of insulation degradation; if not detected in time, it may lead to insulation breakdown, resulting in equipment failure or even power outages. Therefore, accurately collecting PD signals, monitoring discharge activity in real time, and combining this with video surveillance for comprehensive diagnosis are of great significance for improving the operational reliability of switchgear.

[0003] Traditional switchgear status monitoring mainly relies on manual inspection and basic electrical measurements. Manual inspection is mainly achieved through visual inspection, auditory inspection, and tactile inspection. Basic electrical measurement technology is mainly achieved through insulation resistance testing (megohmmeter), loop resistance testing (microohmmeter), power frequency withstand voltage test, manual partial discharge detection, and mechanical characteristic testing. These methods have the following main problems: (1) passive detection, low efficiency, high missed detection rate, and inability to achieve fault alarm; (2) the collected data is discrete data at the time of collection, lacking the ability to analyze historical trends and unable to predict and analyze the status of the switchgear; (3) some tests require power outages, affecting the continuity of power supply.

[0004] Technical content

[0005] The present invention aims to solve at least one of the above-mentioned technical problems.

[0006] To address the aforementioned problems, this invention provides a method, system, electronic device, and storage medium for monitoring and early warning of partial discharge in switchgear.

[0007] In a first aspect, the present invention provides a method for monitoring and early warning of partial discharge in switchgear, comprising:

[0008] Ultrasonic sensor data and transient grounding voltage data are acquired and preprocessed using configuration acquisition software;

[0009] The real-time signal peak is input into a pre-built partial discharge prediction model based on CNN and LSTM, and the output is a joint sequence containing the original data and the predicted values.

[0010] The combined sequence is subjected to MK detection for trend analysis, and the predicted value is combined with the alarm threshold for early warning analysis.

[0011] Optionally, in the partial discharge prediction model, the CNN includes two convolutional layers and a pooling layer. The convolutional layer performs one-dimensional convolution on the input data according to the sliding window length, and the pooling layer outputs the convolution result to the next layer. An extension layer is used to flatten and expand the data processed by the CNN and feed it into the LSTM part.

[0012] Optionally, each convolutional layer of the CNN predefines multiple sets of convolutional kernels with the same shape but different feature sensitivities. The most suitable set of one or more convolutional kernels is automatically selected for computation based on changes in the input features. For example, if the input TEV signal peak sequence is x, multiple convolutional kernels W are set. i The output of dynamic convolution is: α i =Softmax(g(x)), where g(x) is a lightweight gating network.

[0013] Optionally, the LSTM is trained using a backpropagation algorithm over time, including:

[0014] Calculate the output value of each neuron forward;

[0015] The error value of each neuron is calculated in reverse, including backpropagation over time and propagation of the error to the next layer of the network.

[0016] Based on the corresponding error terms, the gradient of each weight is calculated, and the weights are updated using the gradient descent method.

[0017] Optionally, the LSTM uses HyperNetwork to dynamically adjust the gate weights, generating a set of scaling factors for adjusting the original weights: W gate =W base +ΔW hyper (x t ,h t-1 ).

[0018] Optionally, the modified dynamic gating is:

[0019] i t =σ((W) i +ΔW i (t))·[h t-1 ,x t ]+b i )

[0020] o t =σ((W) o +ΔW o (t))·[h t-1 ,x t ]+b o )

[0021] f t =σ((W f +ΔW f (t))·[h t-1 ,x t ]+b f )

[0022] Wherein, ΔW i (t), ΔW i (t) and ΔW i (t) represent the dynamic weights of the input gate, output gate, and forget gate, respectively; x t h is the current input. t-1 The hidden layer state at the previous time step; i t o t and f t These represent the state values ​​of the input gate, output gate, and forget gate, respectively; W i W o and W f b represents the weight matrices for the input gate, output gate, and forget gate, respectively; i b o and b f These represent the bias terms for the input gate, output gate, and forget gate, respectively.

[0023] Optionally, the step of performing trend analysis on the joint sequence using MK detection and combining the predicted value with the alarm threshold for early warning analysis includes: when the predicted value is ≤20dB, it is a normal state; when the predicted value is ≥25dB ​​or the predicted value is >20dB and the statistical value Z of MK detection is... k When the value is greater than 1.96, an alarm is issued; when 25dB > the predicted value > 20dB, a warning is issued.

[0024] Secondly, the present invention provides an early warning system for partial discharge monitoring of switchgear, comprising:

[0025] The acquisition module is used to acquire and preprocess ultrasonic sensor data and transient grounding voltage data through configuration acquisition software;

[0026] The prediction module is used to input the real-time signal peak into a pre-built partial discharge prediction model based on CNN and LSTM, and output a joint sequence containing the original data and the predicted values.

[0027] The verification module is used to perform trend analysis on the combined sequences using MK detection.

[0028] The early warning module is used to perform early warning analysis based on the results of the trend analysis, the predicted value, and the alarm threshold.

[0029] Thirdly, the present invention provides an electronic device, including a memory and a processor;

[0030] The memory is used to store computer programs;

[0031] The processor is configured to implement, when executing the computer program, a method for monitoring and early warning of partial discharge in a switchgear as described in the first aspect.

[0032] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for monitoring and early warning of partial discharge in a switchgear as described in the first aspect.

[0033] The technical solution of this invention, by constructing an improved partial discharge prediction model based on CNN and LSTM, avoids data loss and the introduction of unnecessary noise, and is less likely to get trapped in local optima, thereby improving the prediction effect of trend data. Furthermore, by combining MK detection and threshold judgment logic, the signal peak size and trend parameters are used together as early warning indicators, which effectively improves the accuracy of early warning. This solves the problems of poor real-time performance, high false alarm rate, and reliance on manual labor in traditional switchgear monitoring technology, providing a safer, more economical, and more reliable switchgear status management solution for power systems.

[0034] Instruction manual illustrations

[0035] Figure 1 This is a flowchart illustrating the partial discharge monitoring and early warning method for switchgear according to an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of a partial discharge prediction model according to an embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the LSTM model according to an embodiment of the present invention;

[0038] Figure 4 This is an architecture diagram of a switchgear partial discharge monitoring and early warning system according to an embodiment of the present invention;

[0039] Figure 5 This is a system architecture diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0041] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0042] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0043] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0044] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0045] like Figure 1 As shown in the figure, the present invention provides a method for monitoring and early warning of partial discharge in switchgear, and the entire process is as follows:

[0046] S1. Acquire ultrasonic sensor data and transient grounding voltage data through configuration acquisition software and preprocess them.

[0047] Specifically, after connecting to the electrical partial discharge monitoring equipment through the configuration acquisition software, the configuration binding between the partial discharge gateway and the equipment room, and between the switch cabinet and the partial discharge sensor is first completed. Then, data acquisition is performed using the Modbus master station, which supports both Modbus RTU and TCP protocols.

[0048] According to the acquisition protocol, the parsed and identified data includes: sensor ID, system time, power level, ultrasonic (AE) characteristics such as alarm level, discharge type, peak signal value, average signal value, discharge frequency, 50Hz correlation, 100Hz correlation, etc., and transient ground voltage (TEV) characteristics such as alarm level, discharge type, peak signal value, average signal value, discharge frequency, 50Hz correlation, 100Hz correlation, etc. Finally, the data is integrated with the data push layer to complete the configuration and data entry into the lake.

[0049] After the data collection and storage steps are completed in the data lake, a real-time monitoring dashboard can be used to monitor and display new partial discharge data in real time. It should also provide a function to view historical data trends. The historical trend chart should support trend queries over a period of time (one day / one week / one month), and also support queries for specific time intervals, with a maximum support of one year of data. Multiple trend charts should be displayed, and trends for different units should be plotted on multiple vertical axes.

[0050] Preprocessing mainly includes using the IQR method to identify outliers and replace them with a 5-point moving average; and using linear interpolation to fill short-term missing values.

[0051] S2. Input the real-time signal peak into the pre-built partial discharge prediction model based on CNN and LSTM, and output a joint sequence containing the original data and the predicted values.

[0052] refer to Figure 2 The CNN in the partial discharge prediction model consists of two convolutional layers and a pooling layer. The convolutional layer performs one-dimensional convolution on the input data according to the sliding window length, and the pooling layer outputs the convolution result to the next layer. After the CNN part finishes processing, an extension layer flattens and expands the data before feeding it into the LSTM part. The input to the prediction model is a normalized sequence of partial discharge TEV monitoring values ​​(signal peaks), and its output is a joint sequence composed of the original monitoring data and the predicted data.

[0053] Using a CNN-LSTM joint method for partial discharge data sequence prediction can effectively combine the feature extraction ability of CNN and the sensitivity of LSTM to time series, thereby improving the prediction effect of trend data.

[0054] refer to Figure 3 LSTM employs a backpropagation algorithm over time, which means that the LSTM is unfolded into a deep feedforward neural network according to time sequence, and then trained using the backpropagation (BP) algorithm, as detailed below:

[0055] S21. Calculate the output value of each neuron in the forward pass.

[0056] Signal sequences are transmitted between neurons in the hidden layer through three controllable gates: the input gate, the forget gate, and the output gate. Among them, i i This indicates an input gate, used to control the input of information; o t f represents an output gate, used to output control information; t This represents the forget gate, which controls the retention of historical information. The gating structure uses the sigmoid activation function (σ function) to map variables to the interval [0,1].

[0057] The calculation formulas for the input gate, output gate, and forget gate are as follows:

[0058] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0059] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0060] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0061] Where, x t h is the current input. t-1 The hidden layer state at the previous time step; i t o t and f t These represent the state values ​​of the input gate, output gate, and forget gate, respectively; W i W o and W f b represents the weight matrices for the input gate, output gate, and forget gate, respectively; i b o and b f These represent the bias terms for the input gate, output gate, and forget gate, respectively.

[0062] The final output is determined by the output gate state value o. t and unit state c t Jointly determined:

[0063]

[0064] In the formula W c b c These are the weight matrix and bias term of the input unit state, respectively; c t , and h t , representing the candidate state, current state, and hidden state of a cell, respectively; the tanh layer is a hyperbolic tangent activation function that maps variables to the interval [-1, 1]. This indicates element-wise multiplication.

[0065] S22. Calculate the error value of each neuron in reverse, including backpropagation in time and propagation of the error to the next layer of the network in two directions;

[0066] S23. Based on the corresponding error terms, calculate the gradient of each weight and update the weights using the gradient descent method.

[0067] Specifically, due to the low peak value of TEV signals under certain operating conditions and the sudden surge under others, the noise components differ at different switchgear / bus locations, and the peak value is easily affected by ambient temperature and humidity, load, and partial discharge status. Therefore, for the CNN network of this invention, instead of using a single fixed convolutional kernel, multiple sets of convolutional kernels with the same shape but different feature sensitivities are predefined for each convolutional layer, and the network automatically selects the most suitable set of one or more sets of convolutional kernels to participate in the calculation based on changes in input features. That is, the convolutional kernels are no longer fixed parameters, but are dynamically combined in an input-driven manner, thereby achieving adjustability and adaptability in the convolutional feature extraction process.

[0068] Let the peak sequence of the input TEV signal be x, and set multiple convolutional kernels W. i The output of dynamic convolution is:

[0069]

[0070] Where, α i =Softmax(g(x)), the lightweight gated network g(x) can be a fully connected layer (FC), a 1×1 convolutional network, or a more complex but less computationally expensive structure. Its output needs to be normalized using softmax.

[0071] Meanwhile, because traditional LSTM models have fixed hidden layer weights, they cannot dynamically adjust the memory strategy according to changes in signal strength and periodic changes in discharge conditions. Therefore, this invention employs a HyperLSTM mechanism, which dynamically modifies the LSTM's gating weights through an auxiliary network (HyperNetwork) to achieve dynamic parameter adaptation. This model can dynamically adjust memory and forgetting based on different input features, rather than using fixed weights.

[0072] Unlike directly predicting the complete weight matrix, HyperLSTM employs a more efficient low-rank dynamic adjustment strategy: the auxiliary LSTM does not directly generate new weights, but instead generates a set of scaling coefficients to adjust the original weights. These coefficients are used to modulate the input weights, hidden state weights, and bias terms of the main LSTM, giving the gating different sensitivities at different time steps. In other words, HyperLSTM transforms a "fixed-weight LSTM" into a dynamic LSTM that is input-driven and time-varying, thus significantly improving the model's ability to handle non-stationary sequences. Compared to static LSTM, it can effectively capture short-term mutations; its robustness to noise is greatly improved, it has a stronger ability to model long-term trends, and it can better adapt to different operating conditions. The specific principle is as follows:

[0073] W gate =W base +ΔW hyper (x t ,h t-1 )

[0074] Wherein, ΔW hyper The gating weights are dynamically generated by HyperNetwork based on the current input or context, and change dynamically at each time step. The gating parameters are still calculated using the LSTM formula, but their weights are no longer fixed. The modified dynamic gating formula is shown below:

[0075] i t =σ((W) i +ΔW i (t))·[h t-1 ,x t ]+b i )

[0076] o t =σ((W) o +ΔW o (t))·[h t-1 ,x t ]+b o )

[0077] f t =σ((W) f +ΔW f (t))·[h t-1 ,x t ]+b f )

[0078] Wherein, ΔW i (t), ΔW i (t) and ΔW i (t) represents the dynamic weights of the input gate, output gate, and forget gate, respectively.

[0079] S3. Perform trend analysis on the combined sequence using MK detection, and combine the predicted value with the alarm threshold for early warning analysis.

[0080] Specifically, the basic principle of the MK test is as follows:

[0081] For a TEV prediction sequence x with a sample size of n (preferably 500 groups in this embodiment), construct a new sequence:

[0082]

[0083] in:

[0084]

[0085] Among them, S k This represents the cumulative number of sample data points at time i that are greater than the number of data points at time j.

[0086] The statistic Z is defined below under the assumption of random independence in time series. k :

[0087]

[0088] In the sequence x1, x i ,…,x n When they are independent, S k The mean E(S) k ) and variance var(S k They have the same continuous distribution. When k=1, Z k =0.

[0089] Generally, a significance level of p = 0.05 is chosen for the MK test on the time series T(t). The statistical statistic Z and the significance level p value are then calculated through statistical analysis, where Z... k It follows a standard normal distribution and is a statistical sequence calculated in chronological order of time. It is used to reveal the changing trends and abrupt changes in partial discharge data.

[0090] If Z k A value >1.96 and p<0.05 indicate a significantly upward trend;

[0091] If Z k If the value is less than -1.96 and p < 0.05, the trend is considered to be significantly decreasing.

[0092] Otherwise, the trend is not significant. That is, in Z... k >1.96 or Z k A value <-1.96 indicates a significant upward or downward trend, and values ​​exceeding the critical threshold are identified as the time range where a sudden change occurred.

[0093] Therefore, this embodiment uses the MK test to analyze partial discharge data in order to clearly reveal the trend changes of partial discharge.

[0094] After obtaining the predicted values ​​and trend changes, the approach to early warning analysis is as follows:

[0095] In particular, to improve the accuracy of the prediction, the real-time signal peak value can be repeatedly input into the partial discharge prediction model and the average value can be taken to obtain the predicted value. In this embodiment, the predicted value is the absolute value of TEV.

[0096] When the predicted value does not exceed 20dB, it is defined as a normal state; when the predicted value is greater than 20dB and less than 25dB, it is defined as a warning state; when the predicted value is 25dB or above, an alarm is issued; or when the predicted value is greater than 20dB and Zk is greater than 1.96, an alarm is issued.

[0097] This embodiment only uses the predicted value of TEV for early warning analysis. Of course, the value of ultrasonic AE can also be predicted to set a corresponding early warning threshold for partial discharge monitoring and analysis. The principle and approach are the same: using CNN-LSTM to predict the partial discharge data sequence, and then using the MK test to perform trend analysis, comprehensively considering both the amplitude of the partial discharge sequence and trend parameters to achieve early warning of partial discharge.

[0098] like Figure 4 As shown, this embodiment of the invention provides a switchgear partial discharge monitoring and early warning system 300, comprising:

[0099] The acquisition module 310 is used to acquire and preprocess ultrasonic sensing data and transient grounding voltage data through configuration acquisition software;

[0100] The prediction module 320 is used to input the real-time signal peak into a pre-built partial discharge prediction model based on CNN and LSTM, and output a joint sequence containing the original data and the predicted values.

[0101] The verification module 330 is used to perform trend analysis on the combined sequences using MK detection.

[0102] The early warning module 340 is used to perform early warning analysis based on the results of trend analysis, predicted values ​​and alarm thresholds.

[0103] like Figure 5 As shown, an electronic device 400 provided in this embodiment of the invention includes a memory 410 and a processor 420; the memory 410 is used to store a computer program; the processor 420 is used to implement a switch cabinet partial discharge monitoring and early warning method as described above when the computer program is executed.

[0104] Alternatively, an electronic device 400 includes a memory 410 and a processor 420 coupled to the memory 410; the memory 410 is configured to store a computer program; and the processor 420 is configured to perform the following operations when the computer program is executed:

[0105] Ultrasonic sensor data and transient grounding voltage data are acquired and preprocessed using configuration acquisition software;

[0106] The real-time signal peak is input into a pre-built partial discharge prediction model based on CNN and LSTM, and the output is a joint sequence containing the original data and the predicted values.

[0107] The combined sequences are subjected to MK detection for trend analysis, and the predicted values ​​are combined with alarm thresholds for early warning analysis.

[0108] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements a switchgear partial discharge monitoring and early warning method as described above.

[0109] Alternatively, a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the following operations:

[0110] Ultrasonic sensor data and transient grounding voltage data are acquired and preprocessed using configuration acquisition software;

[0111] The real-time signal peak is input into a pre-built partial discharge prediction model based on CNN and LSTM, and the output is a joint sequence containing the original data and the predicted values.

[0112] The combined sequences are subjected to MK detection for trend analysis, and the predicted values ​​are combined with alarm thresholds for early warning analysis.

[0113] The present invention will now be described an electronic device 400 that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic device 400 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 400 can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0114] Electronic device 400 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or a computer program loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0115] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. In this application, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention according to actual needs. Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units can be implemented in hardware or as software functional units.

[0116] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for monitoring and early warning of partial discharge in switchgear, characterized in that... include: Ultrasonic sensor data and transient grounding voltage data are acquired and preprocessed using configuration acquisition software; The real-time signal peak is input into a pre-built partial discharge prediction model based on CNN and LSTM, and the output is a joint sequence containing the original data and the predicted values. The combined sequence is subjected to MK detection for trend analysis, and the predicted value is combined with the alarm threshold for early warning analysis.

2. The method for monitoring and early warning of partial discharge in switchgear as described in claim 1, characterized in that... In the partial discharge prediction model, the CNN includes two convolutional layers and a pooling layer. The convolutional layer performs one-dimensional convolution on the input data according to the sliding window length, and the pooling layer outputs the convolution result to the next layer. An extension layer is used to flatten and expand the data processed by the CNN and feed it into the LSTM part.

3. A method for monitoring and early warning of partial discharge in a switchgear as described in claim 1 or 2, characterized in that... Each convolutional layer of the CNN predefines multiple sets of convolutional kernels with the same shape but different feature sensitivities. The most suitable set of one or more sets of convolutional kernels is automatically selected for computation based on the changes in the input features. Let the peak sequence of the input TEV signal be x, and set multiple convolutional kernels W. i The output of dynamic convolution is: α i =Softmax(g(x)), where g(x) is a lightweight gating network.

4. The method for monitoring and early warning of partial discharge in switchgear as described in claim 1, characterized in that... The LSTM is trained using a backpropagation algorithm over time, including: Calculate the output value of each neuron forward; The error value of each neuron is calculated in reverse, including backpropagation over time and propagation of the error to the next layer of the network. Based on the corresponding error terms, the gradient of each weight is calculated, and the weights are updated using the gradient descent method.

5. The method for monitoring and early warning of partial discharge in switchgear as described in claim 4, characterized in that... The LSTM uses HyperNetwork to dynamically adjust the gated weights, generating a set of scaling factors for adjusting the original weights: W gate =W base +ΔW hyper (x t ,h t-1 ).

6. The method for monitoring and early warning of partial discharge in switchgear as described in claim 5, characterized in that... The revised dynamic gating is as follows: to t =σ((W i +ΔW i (t))·[h t-1 ,x t ]+b i ) o t =σ((W o +ΔW o (t))·[h t-1 ,x t ]+b o ) f t =σ((W f +ΔW f (t))·[h t-1 ,x t ]+b f ) Wherein, ΔW i (t), ΔW i (t) and ΔW i (t) represent the dynamic weights of the input gate, output gate, and forget gate, respectively; x t h is the current input. t-1 The hidden layer state at the previous time step; i t o t and f t These represent the state values ​​of the input gate, output gate, and forget gate, respectively; W i W o and W f b represents the weight matrices for the input gate, output gate, and forget gate, respectively; i b o and b f These represent the bias terms for the input gate, output gate, and forget gate, respectively.

7. The method for monitoring and early warning of partial discharge in switchgear as described in claim 1, characterized in that... The step of performing trend analysis on the combined sequence using MK detection and combining the predicted value with the alarm threshold for early warning analysis includes: When the predicted value is ≤20dB, it is considered normal; when the predicted value is ≥25dB ​​or >20dB and the statistical value Z of MK detection is... k When the value is greater than 1.96, an alarm is issued; when 25dB > the predicted value > 20dB, a warning is issued.

8. The early warning system for partial discharge monitoring of switchgear as described in claim 1, characterized in that... include: The acquisition module is used to acquire and preprocess ultrasonic sensor data and transient grounding voltage data through configuration acquisition software; The prediction module is used to input the real-time signal peak into a pre-built partial discharge prediction model based on CNN and LSTM, and output a joint sequence containing the original data and the predicted values. The verification module is used to perform trend analysis on the combined sequences using MK detection. The early warning module is used to perform early warning analysis based on the results of the trend analysis, the predicted value, and the alarm threshold.

9. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement, when executing the computer program, a method for monitoring and early warning of partial discharge in a switchgear as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements a method for monitoring and warning of partial discharge in a switchgear as described in any one of claims 1 to 7.

Images