Power distribution network partial discharge on-line monitoring method and device
By combining Kalman filtering and convolutional twin neural networks, the signal interference and real-time issues in partial discharge monitoring of power distribution networks are solved, achieving high-precision fault diagnosis and real-time monitoring, and supporting intelligent early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG BOHONG ELECTRIC CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-05-29
AI Technical Summary
Traditional partial discharge monitoring methods suffer from significant signal interference and poor real-time performance in power distribution networks, making it difficult to accurately extract partial discharge signals and conduct timely fault assessments.
Kalman filtering is used to filter noise in the local current signal, and convolutional twin neural network is used to process the partial discharge characteristic signal. The results are integrated into an FPGA programmable logic array for real-time analysis, and environmental parameters are introduced for fault diagnosis and source tracing.
It improves the accuracy and real-time performance of partial discharge fault monitoring, enabling accurate fault type identification and intelligent early warning, thus meeting the needs of real-time online monitoring.
Smart Images
Figure CN122109742A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system condition monitoring, and in particular to a method and device for online monitoring of partial discharge in distribution networks. Background Technology
[0002] During long-term operation, the insulation materials of power distribution network equipment may experience partial discharge (PD) due to aging, external high-frequency induced by environmental factors, or manufacturing defects. Partial discharge is an early warning signal of power equipment degradation; failure to detect and address it in a timely manner can lead to equipment failure or even power grid accidents. Therefore, conducting online monitoring and source analysis of partial discharge in power distribution networks is of great significance for improving power supply reliability and equipment operation safety.
[0003] Currently, traditional partial discharge monitoring methods mainly rely on high-frequency current sensors and other methods for signal acquisition, followed by manual analysis or rule-based methods for source location. However, these traditional methods have the following problems:
[0004] Significant signal interference: Due to the complex operating environment of the power distribution network, there is a large amount of external noise such as electromagnetic interference and mechanical vibration. The frequency range of these noise signals overlaps with that of the partial discharge signal, affecting the accurate extraction of the partial discharge signal.
[0005] Poor real-time performance: Manual analysis methods rely on professional technicians to interpret and analyze the collected signals one by one. This method is not only inefficient, but also cannot dynamically assess the equipment's operating status in a timely manner, making it difficult to meet the needs of real-time online monitoring and unable to effectively prevent sudden failures. Summary of the Invention
[0006] To address the aforementioned technical problems, this invention provides a method and apparatus for online monitoring of partial discharge in power distribution networks, which improves the accuracy and real-time performance of fault monitoring through partial discharge.
[0007] The technical solution provided by this invention is as follows:
[0008] A method for online monitoring of partial discharge in a power distribution network, the method comprising:
[0009] S1: Real-time acquisition of local current signals and environmental parameters of the power distribution network;
[0010] The local current signal is collected by a high-frequency current sensor included in the monitoring module deployed on the power distribution network, and the environmental parameters are collected by an environmental sensor included in the monitoring module deployed on the power distribution network.
[0011] S2: The local current signal is subjected to noise filtering and denoising processing using the Kalman filter method to extract the partial discharge characteristic signal;
[0012] S3: Input the partial discharge characteristic signal and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type;
[0013] The partial discharge fault diagnosis model is a convolutional twin neural network constructed based on a convolutional neural network and a twin neural network. During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.
[0014] Furthermore, the monitoring module includes a main control unit, which includes an FPGA programmable logic array and an ARM processor. The trained partial discharge fault diagnosis model is deployed in the FPGA programmable logic array, and the high-frequency current sensor and environmental sensor are connected to the main control unit.
[0015] Furthermore, the environmental sensor includes a temperature sensor and a vibration sensor, and the environmental parameters include conductor temperature data of the power distribution network collected by the temperature sensor and conductor vibration data of the power distribution network collected by the vibration sensor.
[0016] Furthermore, the convolutional twin neural network includes parallel convolutional neural networks and twin neural networks, and the outputs of the parallel convolutional neural networks and twin neural networks are fused for decision-making.
[0017] Furthermore, S2 includes:
[0018] S21: The acquired local current signal is converted into a one-dimensional time series signal by discrete sampling;
[0019] S22: After removing the DC component from the one-dimensional time series signal, normalize it to scale the signal amplitude to the range of [-1, 1].
[0020] S23: Based on the constructed Kalman filter model, the normalized one-dimensional time series signal is filtered and denoised to obtain a pure discharge signal;
[0021] S24: Based on the pure discharge signal, extract the signal peak value, discharge charge amount and pulse width, and combine them with the frequency domain characteristics of the pure discharge signal to obtain the partial discharge characteristic signal.
[0022] Furthermore, the method also includes:
[0023] S4: Based on the waveform characteristics of the partial discharge characteristic signal, trace the source of the discharge.
[0024] Furthermore, the method also includes:
[0025] S5: Based on the characteristic parameters of partial discharge characteristic signals, match them with historical data to calculate the health index, predict the fault development rate based on the health index, and provide early warning and maintenance suggestions.
[0026] An online monitoring device for partial discharge in a power distribution network, the device comprising:
[0027] The data acquisition module is used to acquire local current signals and environmental parameters of the power distribution network in real time.
[0028] The local current signal is collected by a high-frequency current sensor included in the monitoring module deployed on the power distribution network, and the environmental parameters are collected by an environmental sensor included in the monitoring module deployed on the power distribution network.
[0029] The preprocessing module is used to perform noise filtering and denoising on the local current signal using the Kalman filtering method to extract the partial discharge characteristic signal;
[0030] The fault reasoning module is used to input the partial discharge characteristic signals and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type.
[0031] The partial discharge fault diagnosis model is a convolutional twin neural network constructed based on a convolutional neural network and a twin neural network. During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.
[0032] Furthermore, the monitoring module includes a main control unit, which includes an FPGA programmable logic array and an ARM processor. The trained partial discharge fault diagnosis model is deployed in the FPGA programmable logic array, and the high-frequency current sensor and environmental sensor are connected to the main control unit.
[0033] Furthermore, the environmental sensor includes a temperature sensor and a vibration sensor, and the environmental parameters include conductor temperature data of the power distribution network collected by the temperature sensor and conductor vibration data of the power distribution network collected by the vibration sensor.
[0034] Furthermore, the convolutional twin neural network includes parallel convolutional neural networks and twin neural networks, and the outputs of the parallel convolutional neural networks and twin neural networks are fused for decision-making.
[0035] Furthermore, the preprocessing module includes:
[0036] Discretization unit, used to convert the acquired local current signal into a one-dimensional time series signal through discrete sampling;
[0037] The normalization unit is used to normalize a one-dimensional time series signal after removing the DC component, and to scale the signal amplitude to the range of [-1, 1].
[0038] The Kalman filter unit is used to filter and denoise the normalized one-dimensional time series signal based on the constructed Kalman filter model to obtain a pure discharge signal.
[0039] The partial discharge characteristic signal acquisition unit is used to extract the signal peak value, discharge charge and pulse width based on the pure discharge signal, and combine them with the frequency domain characteristics of the pure discharge signal to obtain the partial discharge characteristic signal.
[0040] Furthermore, the device also includes:
[0041] The fault tracing module is used to trace the location of the discharge source based on the waveform characteristics of the partial discharge characteristic signal.
[0042] Furthermore, the device also includes:
[0043] The health prediction module is used to match the characteristic parameters of partial discharge characteristic signals with historical data to calculate the health index, predict the fault development rate based on the health index, and provide early warnings and maintenance suggestions.
[0044] The present invention has the following beneficial effects:
[0045] This invention automatically analyzes the collected signals using a constructed partial discharge fault diagnosis model to determine the type of partial discharge fault. Compared with traditional manual analysis methods, it offers better real-time performance, meeting the needs of real-time online monitoring, effectively preventing sudden faults, and improving the accuracy of fault monitoring through partial discharge. Furthermore, this application incorporates environmental parameters, effectively reducing interference from the complex environment of the power distribution network and enabling accurate determination of the partial discharge fault type.
[0046] In this invention, the partial discharge fault diagnosis model is deployed in an FPGA programmable logic array, which integrates the acquisition and analysis functions into the monitoring module. This enables real-time monitoring, accurate tracing, and intelligent early warning of partial discharge in power distribution network equipment on-site, without the need to transmit a large amount of raw waveform data. This meets the real-time requirements of online equipment monitoring, completes end-side integrated processing of discharge signal acquisition, feature extraction, and fault diagnosis, effectively identifies valid partial discharge information, and eliminates interference signals. Attached Figure Description
[0047] Figure 1 This is a flowchart of the online monitoring method for partial discharge in power distribution networks according to the present invention;
[0048] Figure 2 This is a schematic diagram of the monitoring module;
[0049] Figure 3 This is a schematic diagram of the online monitoring device for partial discharge in power distribution networks according to the present invention. Detailed Implementation
[0050] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0051] This invention provides an online monitoring method for partial discharge in a power distribution network, such as... Figure 1 As shown, the method includes:
[0052] S1: Real-time acquisition of local current signals and environmental parameters of the power distribution network.
[0053] Local current signals and environmental parameters are acquired by a monitoring module, which is deployed at key nodes in the power distribution network. For example... Figure 2 As shown, it includes a main control unit (CPU), a high-frequency current sensor (HFCT), a power frequency voltage sensor (when needed), and an environmental sensor.
[0054] The high-frequency current sensor acquires high-frequency local current signals (acquiring both the magnitude and direction of the current). The high-frequency current signal acquisition circuit consists of an AMP signal processing circuit and four high-speed A / D synchronous conversion chips, which amplify the high-frequency current signal and acquire it at a frequency of 100MHz.
[0055] The power frequency voltage sensor consists of an AMP signal conditioning circuit and a four-channel A / D synchronous conversion chip to acquire the open delta voltage on the secondary side of the PT.
[0056] The aforementioned high-frequency current sensor and power frequency voltage sensor acquire current and voltage signals through mutual inductance, rather than being directly connected to the power distribution network.
[0057] Environmental sensors include temperature sensors (thermocouples, etc.) and vibration sensors. Environmental parameters include conductor temperature data of the power distribution network collected by temperature sensors and conductor vibration data of the power distribution network collected by vibration sensors.
[0058] The main control unit is connected to the aforementioned high-frequency current sensor, power frequency voltage sensor, and environmental sensor to acquire and process data from the high-frequency current, power frequency voltage, temperature, and vibration sensors. The main control unit includes an FPGA programmable logic array and an ARM processor (e.g., a 64-bit quad-core ARM-A53 processor), and is connected to 64GB of DDR4 memory.
[0059] The main control unit is also connected to 4G / 5G and wired network communication units, ensuring real-time transmission and remote access to monitoring data. The 4G / 5G communication module exchanges data with the main control unit via USB protocol; the wired network communication unit, consisting of a gigabit Ethernet port, is used for data exchange when connected via wire and interacts with the main control unit via PHY protocol.
[0060] S2: The local current signal is subjected to noise filtering and denoising processing using the Kalman filter method to extract the partial discharge characteristic signal.
[0061] This step is used to perform noise filtering and denoising on the pulse-type local current signal using the Kalman filter, to achieve recursive filtering and ensure that the core characteristics of the signal pulse, such as amplitude, width, and rising edge, are not distorted after filtering.
[0062] In one example, this step is implemented as follows:
[0063] S21: The original local current signal is converted into a one-dimensional time series signal by discrete sampling.
[0064] The original PD (partial current) signal is converted into a one-dimensional time series signal. k is the discrete sampling time. The total number of sampling points is used to store the one-dimensional time series signal in single-precision floating-point format; the duration of each sampling session is greater than 40ms to ensure coverage of the phase distribution characteristics of the partial discharge signal.
[0065] S22: After removing the DC component from the one-dimensional time series signal, normalize it to scale the signal amplitude to the range of [-1, 1].
[0066] The formula for eliminating the DC component is:
[0067]
[0068] in This is the signal after removing the DC component.
[0069] The normalization formula is as follows:
[0070]
[0071] S23: Based on the constructed Kalman filter model, the normalized one-dimensional time series signal is filtered and denoised to obtain a pure discharge signal.
[0072] The specific implementation method for this step is as follows:
[0073] 1. Construct a Kalman filter model adapted to the pulse signal, with the state equation as follows: The observation equation characterizes the dynamic evolution of the actual discharge signal as follows: This reflects the mapping relationship between noisy signals and real signals, where... , It is Gaussian white noise, and obeys ~N(0,Q), ~N(0,R) distribution.
[0074] 2. Initialize core parameters: The initial state estimate for filtering is taken from the first preprocessed observation. ,in This represents the optimal estimate of the initial state and the initial error covariance. Process noise covariance Q=10 −3 The observation noise covariance R = 0.05.
[0075] 3. Point-by-point iterative filtering: First, complete the state prediction. Predicting with prior error Then, the Kalman gain is calculated to weigh the reliability of the prior estimate against the observed value, with a value range of [0,1]. The closer the gain is to 1, the more reliable the observed value; the closer it is to 0, the more reliable the prior estimate.
[0076]
[0077] via status update With posterior error correction Output a pure discharge signal.
[0078] S24: Extract the signal peak value, discharge charge, and pulse width based on the pure discharge signal, and combine them with the frequency domain characteristics of the pure discharge signal to obtain the partial discharge characteristic signal.
[0079] signal peak The pulse intensity used to characterize partial discharge. .
[0080] Discharge charge ,in, This represents the equivalent charge of a single discharge pulse (unit: pc). This represents the signal sampling rate.
[0081] Pulse width ,in, Indicates the duration of a single discharge pulse. The sampling time when the rising edge of the pulse reaches the threshold. This is the sampling time when the falling edge of the pulse reaches the threshold.
[0082] Based on the above core time-domain features, combined with frequency-domain features, partial discharge characteristic signals are formed. When necessary, multi-sensor data fusion technology can be used to weighted fuse the characteristics of multi-channel signals at the same measurement point to complete the comprehensive analysis of the discharge signal and improve the effectiveness of the features.
[0083] S3: Input the partial discharge characteristic signals and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type.
[0084] Among them, the partial discharge fault diagnosis model is a convolutional twin neural network (CSN) built based on convolutional neural network (CNN) and Siamese neural network (SNN). During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.
[0085] Specifically, convolutional twin neural networks include parallel convolutional neural networks and twin neural networks, and the outputs of the parallel convolutional neural networks and twin neural networks are fused for decision-making.
[0086] Convolutional twin neural networks need to be trained before use, and the training process is as follows:
[0087] 1. Based on a historical fault database of 30 million records, the partial discharge characteristic signals, partial discharge fault types, equipment parameters, and environmental parameters in the historical database are standardized and divided into training set, validation set, and test set in a 7:2:1 ratio to construct a labeled fault feature dataset.
[0088] 2. Utilizing the convolutional layers of a CNN network With pooling layer Feature extraction is performed on the discharge time-frequency feature map, and a high-dimensional feature vector F is output, realizing in-depth mining of local and global features; the feature extraction formula is shown below:
[0089]
[0090] in, For convolution kernel, For bias, * indicates convolution operation. It is a non-linear activation function (such as ReLU).
[0091] The classification decision (SoftMax layer) formula is as follows:
[0092]
[0093] in, For classification probability, To output the weight matrix, This is the final feature mapping.
[0094] 3. Introduce an SNN network to perform a Siamese mapping f(x) on the fault features and calculate the similarity between samples:
[0095]
[0096] This addresses the challenge of classifying similar faults with small sample sizes. The CNN feature output is fed into an SNN classification layer, and the network weights are optimized through backpropagation. The fault tracing loss function (cross-entropy) is as follows:
[0097] in, For real category labels, To predict probabilities.
[0098] Through the above training process, the model is trained and iteratively optimized to learn the relationship between partial discharge characteristics and faults, achieve accurate classification of partial discharge fault types (corona discharge, surface discharge, internal air gap discharge) and fault severity, and build a complete AI model for partial discharge fault diagnosis.
[0099] After training, the partial discharge fault diagnosis model can be deployed in an FPGA programmable logic array. The model can be used to analyze and process the input partial discharge characteristic signals and environmental parameters. Based on the learned relationship between the partial discharge characteristic signals and environmental parameters and the partial discharge fault type, the model can output the partial discharge fault type (corona discharge, surface discharge, internal air gap discharge, etc.). When necessary, the severity of the fault can also be classified.
[0100] This invention automatically analyzes the collected signals using a constructed partial discharge fault diagnosis model to determine the type of partial discharge fault. Compared with traditional manual analysis methods, it offers better real-time performance, meeting the needs of real-time online monitoring, effectively preventing sudden faults, and improving the accuracy of fault monitoring through partial discharge. Furthermore, this application incorporates environmental parameters, effectively reducing interference from the complex environment of the power distribution network and enabling accurate determination of the partial discharge fault type.
[0101] In traditional monitoring systems, the acquisition and analysis devices are separate. When a fault occurs, a large amount of raw waveform data is collected, resulting in high data transmission bandwidth requirements, extremely high demands on the communication network, and a significant increase in data transmission costs. This is not conducive to the widespread application and promotion of monitoring systems.
[0102] In this invention, the trained partial discharge fault diagnosis model is deployed in an FPGA programmable logic array, integrating both data acquisition and analysis functions within the monitoring module. This enables real-time monitoring, precise source tracing, and intelligent early warning of partial discharge in power distribution network equipment on-site. Specifically, it offers the following advantages:
[0103] Abandoning the traditional trigger-based acquisition mode, this system utilizes an FPGA hardware platform to achieve real-time acquisition of full-frequency signals. Equipped with a high-speed acquisition module, it continuously acquires partial discharge signals across the entire frequency band without omission, matching the sampling rate to the frequency domain characteristics of the discharge signal to ensure signal integrity. Furthermore, the trained CNN+SNN model is quantized and compressed before being ported to the FPGA hardware architecture, enabling model inference within the monitoring module without the need to transmit large amounts of raw waveform data. Parallel computing logic optimizes the model inference process, achieving an AI model inference time of <1ms, meeting the real-time requirements of online equipment monitoring. This enables integrated end-side processing of discharge signal acquisition, feature extraction, and fault diagnosis, effectively identifying valid partial discharge information and eliminating interference signals.
[0104] Partial discharge signals are affected by equipment structure and environmental factors during propagation, which causes signal attenuation, distortion and reflection. The difficulty of accurately locating the discharge source using traditional methods makes source tracing and localization very difficult.
[0105] To solve the above problems, the method of the present invention may further include:
[0106] S4: Based on the waveform characteristics of the partial discharge characteristic signal, trace the source of the discharge source and automatically locate the location of the fault.
[0107] Specifically, the traveling wave propagation equation must first be satisfied. ,in For signal propagation speed, For transmission distance, The signal angular frequency, For time. Discharge traveling wave signals are acquired using sensors, and the arrival times t1, t2, and t3 of the signals at each measuring point are recorded. This is combined with the known distance L between the measuring points. 12 L 23 Establish a set of time difference positioning equations:
[0108]
[0109] Based on the waveform characteristics of the partial discharge characteristic signal, combined with the time difference positioning equations and the distribution network topology, the location of the discharge source can be obtained.
[0110] Furthermore, it can integrate signal amplitude attenuation characteristics and propagation dispersion characteristics extracted by Kalman filtering to correct and optimize the results obtained from the time difference positioning equations, offset the impact of line loss and environmental interference on propagation speed, significantly improve the accuracy of power source location calculation, and achieve precise fault point tracing.
[0111] This invention combines traveling waves, high-frequency partial discharge current signals, and a convolutional twin neural network model to achieve high-precision monitoring, intelligent source tracing, and health prediction of partial discharge, thereby improving the operational safety of power grids and promoting the development of smart grids.
[0112] This invention can also predict the health status of power lines and provide maintenance suggestions by comparing historical data, achieving AI-powered intelligent early warning. It includes the following steps:
[0113] S5: Based on the characteristic parameters of partial discharge characteristic signals, match them with historical data to calculate the health index, predict the fault development rate based on the health index, and provide early warning and maintenance suggestions.
[0114] The specific implementation process is as follows:
[0115] 1. Health Status Quantification: Extract current discharge characteristic parameters (discharge quantity, repetition rate, location result), perform similarity matching with fault development data of similar equipment under the same operating conditions in the historical database, and introduce a health index:
[0116]
[0117] in , These are the weighting coefficients. , The fault threshold is used to quantitatively assess the current health status of the line.
[0118] 2. Failure Trend Prediction: Based on the trend of health index changes, combined with the equipment's operating years and environmental factors, the failure development rate is predicted to reduce failure risk.
[0119] 3. Intelligent early warning and maintenance suggestions: When the health index H is lower than the early warning threshold, or the fault trend shows that there is a risk of deterioration in the short term, the AI intelligent early warning is triggered in time. Combined with historical fault handling cases, targeted maintenance suggestions (defect elimination, equipment maintenance, line replacement) are output to realize early warning and closed-loop management of partial discharge faults and ensure the safe and stable operation of equipment.
[0120] This invention also provides an online monitoring device for partial discharge in power distribution networks, such as... Figure 3 As shown, the device includes:
[0121] Data acquisition module 1 is used to acquire local current signals and environmental parameters of the power distribution network in real time.
[0122] The local current signal is collected by a high-frequency current sensor included in the monitoring module deployed on the power distribution network, and the environmental parameters are collected by an environmental sensor included in the monitoring module deployed on the power distribution network.
[0123] Preprocessing module 2 is used to perform noise filtering and denoising on the local current signal using the Kalman filtering method to extract the partial discharge characteristic signal.
[0124] The fault reasoning module 3 is used to input partial discharge characteristic signals and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type.
[0125] Among them, the partial discharge fault diagnosis model is a convolutional twin neural network based on convolutional neural networks and twin neural networks. During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.
[0126] The aforementioned monitoring module includes a main control unit, which includes an FPGA programmable logic array and an ARM processor. The trained partial discharge fault diagnosis model is deployed in the FPGA programmable logic array, and high-frequency current sensors and environmental sensors are connected to the main control unit.
[0127] Specifically, environmental sensors include temperature sensors and vibration sensors. Environmental parameters include conductor temperature data of the power distribution network collected by temperature sensors and conductor vibration data of the power distribution network collected by vibration sensors.
[0128] The convolutional twin neural network of the present invention includes a parallel convolutional neural network and a twin neural network, and the outputs of the parallel convolutional neural network and the twin neural network are fused for decision-making.
[0129] As one implementation, the preprocessing module includes:
[0130] Discretization unit is used to convert the acquired local current signal into a one-dimensional time series signal through discrete sampling.
[0131] The normalization unit is used to normalize a one-dimensional time series signal after removing the DC component, scaling the signal amplitude to the range of [-1, 1].
[0132] The Kalman filter unit is used to filter and denoise the normalized one-dimensional time series signal based on the constructed Kalman filter model to obtain a clean discharge signal.
[0133] The partial discharge characteristic signal acquisition unit is used to extract the signal peak value, discharge charge and pulse width based on the pure discharge signal, and combine them with the frequency domain characteristics of the pure discharge signal to obtain the partial discharge characteristic signal.
[0134] As an improvement to an embodiment of the present invention, the apparatus further includes:
[0135] The fault tracing module is used to trace the location of the discharge source based on the waveform characteristics of the partial discharge characteristic signal.
[0136] The health prediction module is used to match the characteristic parameters of partial discharge characteristic signals with historical data to calculate the health index, predict the fault development rate based on the health index, and provide early warnings and maintenance suggestions.
[0137] The apparatus provided in this embodiment of the invention operates on the same principle and produces the same technical effects as the aforementioned method embodiments. For the sake of brevity, any parts not mentioned in the apparatus embodiments can be referred to the corresponding content in the aforementioned method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the apparatus and units described above can all be referred to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.
[0138] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.
Claims
1. A method for online monitoring of partial discharge in a power distribution network, characterized in that, The method includes: S1: Real-time acquisition of local current signals and environmental parameters of the power distribution network; The local current signal is collected by a high-frequency current sensor included in the monitoring module deployed on the power distribution network, and the environmental parameters are collected by an environmental sensor included in the monitoring module deployed on the power distribution network. S2: The local current signal is subjected to noise filtering and denoising processing using the Kalman filter method to extract the partial discharge characteristic signal; S3: Input the partial discharge characteristic signal and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type; The partial discharge fault diagnosis model is a convolutional twin neural network constructed based on a convolutional neural network and a twin neural network. During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.
2. The method for online monitoring of partial discharge in a power distribution network according to claim 1, characterized in that, The monitoring module includes a main control unit, which includes an FPGA programmable logic array and an ARM processor. The trained partial discharge fault diagnosis model is deployed in the FPGA programmable logic array, and the high-frequency current sensor and environmental sensor are connected to the main control unit.
3. The method for online monitoring of partial discharge in a power distribution network according to claim 2, characterized in that, The environmental sensors include a temperature sensor and a vibration sensor, and the environmental parameters include conductor temperature data of the power distribution network collected by the temperature sensor and conductor vibration data of the power distribution network collected by the vibration sensor.
4. The method for online monitoring of partial discharge in a power distribution network according to claim 1, characterized in that, The convolutional twin neural network includes parallel convolutional neural networks and twin neural networks, and the outputs of the parallel convolutional neural networks and twin neural networks are fused for decision-making.
5. The method for online monitoring of partial discharge in a power distribution network according to any one of claims 1-4, characterized in that, S2 includes: S21: The acquired local current signal is converted into a one-dimensional time series signal by discrete sampling; S22: After removing the DC component from the one-dimensional time series signal, normalize it to scale the signal amplitude to the range of [-1, 1]. S23: Based on the constructed Kalman filter model, the normalized one-dimensional time series signal is filtered and denoised to obtain a pure discharge signal; S24: Based on the pure discharge signal, extract the signal peak value, discharge charge amount and pulse width, and combine them with the frequency domain characteristics of the pure discharge signal to obtain the partial discharge characteristic signal.
6. The method for online monitoring of partial discharge in a power distribution network according to claim 5, characterized in that, The method further includes: S4: Based on the waveform characteristics of the partial discharge characteristic signal, trace the source of the discharge.
7. The method for online monitoring of partial discharge in a power distribution network according to claim 5, characterized in that, The method further includes: S5: Based on the characteristic parameters of partial discharge characteristic signals, match them with historical data to calculate the health index, predict the fault development rate based on the health index, and provide early warning and maintenance suggestions.
8. An online monitoring device for partial discharge in a power distribution network, characterized in that, The device includes: The data acquisition module is used to acquire local current signals and environmental parameters of the power distribution network in real time. The local current signal is collected by a high-frequency current sensor included in the monitoring module deployed on the power distribution network, and the environmental parameters are collected by an environmental sensor included in the monitoring module deployed on the power distribution network. The preprocessing module is used to perform noise filtering and denoising on the local current signal using the Kalman filtering method to extract the partial discharge characteristic signal; The fault reasoning module is used to input the partial discharge characteristic signals and environmental parameters into the trained partial discharge fault diagnosis model to obtain the partial discharge fault type. The partial discharge fault diagnosis model is a convolutional twin neural network constructed based on a convolutional neural network and a twin neural network. During training, the convolutional twin neural network learns the relationship between partial discharge characteristic signals and environmental parameters and partial discharge fault types.