Power distribution network grounding fault detection method, device, system, medium and equipment

Abstract

The present application relates to the technical field of power distribution network fault detection, and discloses a power distribution network grounding fault detection method, device, system, medium and equipment, the method comprising collecting three-phase current signals on the power distribution network line and determining whether the fault detection starting condition is met; when it is met, the three-phase current waveform data within the preset time before and after the starting time is input into the convolutional neural network model to obtain the fault prediction probability of the power distribution network line; when the fault prediction probability is greater than the probability threshold, a trip command is generated to control the opening of the circuit breaker connected to the fault line in the power distribution network line; if no grounding fault is detected to be isolated within the preset time, the time level difference delay is used to trigger the upstream protection node to perform the backup trip operation. The above method combines pure current drive, intelligent identification, local speed operation and level difference backup to realize high-precision identification of power distribution network grounding faults with reduced dependence on conditions, and high-reliability small-range isolation of grounding faults.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault detection technology, and in particular to a method, device, system, medium and equipment for detecting grounding faults in power distribution networks. Background Technology

[0002] In 10kV distribution network systems, neutral point grounding methods typically include ungrounded, arc-suppression coil grounding, or low-resistance grounding. Among these, ungrounded and arc-suppression coil grounding systems are widely used in rural and some urban distribution networks due to their simple structure and good power supply continuity. Urban power grids, to improve their ability to quickly clear faults, increasingly adopt low-resistance neutral point grounding. However, regardless of the grounding method used, single-phase grounding faults are the most common type of fault in distribution networks. In ungrounded or arc-suppression coil grounding systems, single-phase grounding does not form a low-impedance short-circuit loop; the fault current is mainly the system-to-ground capacitance current, with a small amplitude and easily affected by line length, operating conditions, and the compensation effect of the arc-suppression coil. In low-resistance grounding systems, high-impedance grounding faults also result in weak fault current characteristics. Therefore, accurately and quickly identifying grounding faults and achieving precise isolation under complex operating conditions has become a key technical challenge to ensure the safe and reliable operation of the distribution network.

[0003] Currently, ground fault detection in distribution networks mainly relies on low-current fault location devices at substations, zero-sequence voltage / current power direction criteria, fault indicators, or manual testing. However, each of these methods has its own shortcomings: First, zero-sequence protection methods rely on zero-sequence voltage transformers, which are not configured in most lines. Furthermore, three-phase imbalance can easily generate false zero-sequence voltages, leading to false activation. The compensation effect of the arc suppression coil can also weaken the zero-sequence current characteristics, further reducing the reliability of the criteria. Second, existing fault location devices generally have low accuracy in arc suppression coil systems or high-resistance grounding scenarios, often requiring the rotation of non-faulty lines for troubleshooting, which can cause large-scale power outages and may also result in the accidental disconnection of normal loads. Finally, fault indicators are limited by the installation environment and are difficult to deploy in cable channels or enclosed lines. They are also susceptible to transient interference such as lightning strikes and switch operations, posing a risk of false activation or failure to activate. They can only roughly indicate the fault section and cannot pinpoint the exact location. In summary, existing technologies for detecting grounding faults in power distribution networks generally suffer from problems such as reliance on additional sensors, weak anti-interference capabilities, poor adaptability to high-resistance / compensation systems, low fault identification accuracy, and large power outage areas.

[0004] Therefore, a method for detecting grounding faults in distribution networks that does not rely on zero-sequence voltage, is based solely on conventional three-phase current, and has high identification accuracy is urgently needed. Summary of the Invention

[0005] In view of this, this application provides a method, device, system, medium and equipment for detecting grounding faults in power distribution networks. The main purpose is to solve the technical problems in the prior art, such as the reliance on additional sensors, weak anti-interference ability, poor adaptability to high-resistance / compensation systems and low fault identification accuracy in the detection of grounding faults in power distribution networks.

[0006] According to a first aspect of the present invention, a method for detecting grounding faults in a distribution network is provided, comprising:

[0007] Collect the three-phase current signals on the distribution network lines and determine whether the three-phase current signals meet the fault detection start conditions;

[0008] When the three-phase current signal meets the fault detection start-up condition, the three-phase current waveform data within a preset time before and after the start-up time are input into a preset convolutional neural network model to obtain the fault prediction probability that the distribution network line is in a ground fault state.

[0009] The fault prediction probability is compared with a preset probability threshold. When the fault prediction probability is greater than the probability threshold, a trip command is generated. The trip command is used to control the circuit breaker connected to the faulty line in the distribution network line to trip.

[0010] If the ground fault is not detected and isolated within the preset time, the upstream protection node will be triggered to perform a backup trip operation based on the preset time difference delay.

[0011] Optionally, the step of acquiring the three-phase current signal on the distribution network line and determining whether the three-phase current signal meets the fault detection start-up condition includes: sampling the three-phase current signal on the distribution network line in real time and calculating the current change at adjacent sampling points; calculating the steady-state average value and steady-state standard deviation of the current change within a preset sliding time window, and determining a dynamic start-up threshold based on the steady-state average value, the steady-state standard deviation, and a preset sensitivity coefficient; and determining that the three-phase current signal meets the fault detection start-up condition when the current change at the current sampling point is greater than the dynamic start-up threshold.

[0012] Optionally, determining the dynamic start-up threshold based on the steady-state average, the steady-state standard deviation, and a preset sensitivity coefficient includes: constructing the following formula based on the steady-state average, the steady-state standard deviation, and the preset sensitivity coefficient to determine the dynamic start-up threshold:

[0013]

[0014]

[0015]

[0016]

[0017] Where T is the dynamic startup threshold. 稳态 The steady-state average value is... 稳态 Let be the steady-state standard deviation. I k Let K be the change in current, K be the sensitivity coefficient, N be the number of sampling points in a sliding time window, and k be the number of sampling points.

[0018] Optionally, the step of inputting the three-phase current waveform data within a preset time period before and after the start-up time into a preset convolutional neural network model to obtain the fault prediction probability of the distribution network line being in a ground fault state includes: obtaining the time-series representation corresponding to the three-phase current signal, obtaining the three-phase current waveform data, normalizing the three-phase current waveform data, and extracting data within a preset time period before and after the start-up time to form an input tensor; inputting the input tensor into the preset convolutional neural network model, wherein the convolutional neural network model includes a one-dimensional convolutional layer, a fully connected layer, and an output layer; extracting the coupling transient features of the three-phase current signal in the time dimension through the one-dimensional convolutional layer, and outputting a linear score through the fully connected layer; and converting the linear score into a normal prediction probability of the distribution network being in a normal operating state and a fault prediction probability of the distribution network being in a ground fault state through the output layer using the Softmax function, wherein the sum of the normal prediction probability and the fault prediction probability is 1.

[0019] Optionally, if no ground fault is detected and isolated within a preset time, the current protection node is triggered to perform a backup trip operation based on a preset time difference delay. This includes: sequentially arranging multiple protection nodes along the load side to the power supply side of the distribution network, wherein each protection node is configured with progressively increasing action delays, wherein the action delay of an adjacent protection node downstream of the current protection node is less than the action delay of the current protection node; continuously detecting whether a ground fault in the distribution network downstream of the current protection node is isolated during the delay timer of the current protection node; if a ground fault in the distribution network is detected and isolated before the delay timer of the current protection node ends, the current protection node automatically resets; if a ground fault in the distribution network is not detected and isolated before the delay timer of the current protection node ends, the current protection node performs a trip operation.

[0020] Optionally, whether the distribution network grounding fault has been isolated is determined by the following method: whether the fault current detected by the current protection node has disappeared.

[0021] According to a second aspect of the present invention, a power distribution network grounding fault detection device is provided. The device is used to implement the power distribution network grounding fault detection method as described above, comprising: [list of components connected in series].

[0022] The current sampling module is used to synchronously acquire three-phase current signals;

[0023] A large-capacity memory is used to store current waveform data before and after the time of the ground fault when a ground fault is detected.

[0024] The main control processing module is used to run a preset convolutional neural network model to identify grounding faults and output the fault judgment results.

[0025] The relay output module is used to output a trip command based on the fault judgment result.

[0026] The power supply module is used to convert externally supplied AC or DC power into digital power for digital circuits, analog power for analog circuits, and operating power for relay ports.

[0027] The wireless communication module is used to upload operating status data to the remote distribution network master station and receive setting adjustment instructions, remote control opening instructions and closing instructions issued by the remote distribution network master station;

[0028] The self-organizing network communication module is used to network multiple fault detection devices and exchange data with a remote computer.

[0029] According to a third aspect of the present invention, a power distribution network grounding fault detection system is provided, the system comprising:

[0030] The fault detection start-up unit is used to collect the three-phase current signals on the distribution network line and determine whether the three-phase current signals meet the fault detection start-up conditions.

[0031] The prediction probability output unit is used to input the three-phase current waveform data within a preset time before and after the start time into a preset convolutional neural network model when the three-phase current signal meets the fault detection start condition, so as to obtain the fault prediction probability that the distribution network line is in a ground fault state.

[0032] The fault line disconnection unit is used to compare the fault prediction probability with a preset probability threshold. When the fault prediction probability is greater than the probability threshold, a trip command is generated. The trip command is used to control the circuit breaker connected to the fault line in the distribution network line to trip.

[0033] The differential backup trip unit is used to trigger the current protection node to perform backup trip operation based on a preset time differential delay if no ground fault is detected and isolated within a preset time.

[0034] According to a fourth aspect of the present invention, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described method for detecting grounding faults in a power distribution network.

[0035] According to a fifth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for detecting grounding faults in power distribution networks.

[0036] This invention provides a method, device, system, storage medium, and computer equipment for detecting grounding faults in distribution networks. It detects and identifies faults solely based on conventional three-phase current signals, completely independent of zero-sequence voltage transformers or other additional sensors. This avoids maloperation due to the lack of zero-sequence equipment or interference from three-phase imbalance, improving applicability and deployment flexibility. By introducing a preset triggering mechanism and a convolutional neural network model, it can extract high-dimensional coupling features from transient current waveforms, effectively identifying weak fault characteristics such as high-resistance grounding and improving fault identification accuracy. The decision logic of generating tripping commands by comparing fault prediction probability with probability thresholds gives protection actions adjustable sensitivity and strong anti-interference capabilities. By setting a closed-loop mechanism for upstream time-differential backup tripping triggered by fault non-isolation, it can reliably isolate faults even when local protection fails to operate or circuit breakers malfunction, ensuring reliable fault isolation. In summary, the above method, combining pure current drive, intelligent identification, local rapid action, and time-differential backup, achieves high-precision identification of distribution network grounding faults with reduced reliance on specific conditions, and highly reliable small-range isolation of grounding faults.

[0037] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0038] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0039] Figure 1 A schematic flowchart of a power distribution network grounding fault detection method provided by an embodiment of the present invention is shown;

[0040] Figure 2 This diagram illustrates the structure of a power distribution network grounding fault detection device according to an embodiment of the present invention.

[0041] Figure 3 This diagram illustrates the structure of a power distribution network grounding fault detection system according to an embodiment of the present invention.

[0042] Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0043] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0044] This application provides a method for detecting grounding faults in a power distribution network, such as... Figure 1 As shown, the method includes the following steps:

[0045] 101. Collect the three-phase current signals on the distribution network lines and determine whether the three-phase current signals meet the fault detection start conditions.

[0046] Among them, the three-phase current signal refers to the time-varying current waveform flowing through the A-phase, B-phase, and C-phase conductors in the distribution network line. It usually exists in the form of analog quantity or digital sequence after analog-to-digital conversion and is a basic electrical quantity reflecting the system's operating status. The fault detection start condition refers to the preset criterion threshold or logical rule used to trigger the identification algorithm, such as the current mutation exceeding the set value, the appearance of zero-sequence component, and the increase of three-phase current imbalance. Its function is to distinguish the initial characteristics of normal load fluctuations and potential grounding faults, avoid unnecessary in-depth analysis of non-fault disturbances, and thus improve the response efficiency and resource utilization of the protection system.

[0047] Specifically, the real-time acquisition of three-phase current signals on the distribution network lines usually requires filtering and high-speed analog-to-digital conversion before preprocessing, including noise reduction, synchronization alignment, and initial feature screening. Based on preset fault detection start conditions, a rapid judgment is made. Once the three-phase current signal meets the fault detection start conditions, it will enter the next stage of the fault identification process; if it does not meet the conditions, it will continue to monitor in a loop, realizing the synergistic effect of lightweight monitoring and intelligent triggering.

[0048] In this embodiment, by setting reasonable fault detection activation conditions, the indiscriminate deep learning inference on all current data is effectively avoided, reducing computational burden and power consumption. At the same time, normal transient disturbances caused by motor start-up and shutdown, capacitor switching, etc. are quickly filtered out, preventing the risk of misjudgment caused by false model activation and improving the reliability of protection actions. In addition, the subsequent identification process is activated only when a suspected fault occurs, taking into account both response speed and identification accuracy, and improving the sensitivity and intelligence level of distribution network grounding fault protection.

[0049] 102. When the three-phase current signal meets the fault detection start-up conditions, the three-phase current waveform data within a preset time before and after the start-up time are input into the preset convolutional neural network model to obtain the fault prediction probability that the distribution network line is in a ground fault state.

[0050] Among them, the start time refers to the time point when the fault detection start condition is first met, usually corresponding to the instant when the ground fault first occurs or the transient disturbance is significantly enhanced; the preset duration refers to a fixed time window intercepted around the start time, used to capture the complete transient waveform characteristics before and after the fault occurs; the three-phase current waveform data is a discrete sequence of the A, B, and C phase currents changing with time obtained by high-frequency sampling, forming a three-dimensional input tensor; the Convolutional Neural Network (CNN) model is a deep learning architecture, particularly suitable for processing one-dimensional signals with local correlation and temporal structure; the fault prediction probability refers to the model output value that represents the possibility that the current line is in a ground fault state, with a value range of 0 to 1, generated by normalization by the Softmax function.

[0051] Specifically, once the system determines that the three-phase current signal meets the fault detection start-up conditions, it uses this trigger moment as the start-up moment, extracts the original three-phase current sampling data within a preset time period before and after it from the buffer, and performs normalization processing to form a standardized input tensor. The obtained input tensor is then fed into a convolutional neural network model that has been pre-trained on a large number of ground fault and normal operation samples. The convolutional neural network model automatically extracts the coupling features of the current waveform in the time dimension through multiple one-dimensional convolutional layers, including complex patterns that are difficult to tune manually, such as amplitude abrupt changes, phase shifts, harmonic distortions, and three-phase asymmetry. These features are then mapped to linear scores through a fully connected layer, and finally output two probability values ​​through the Softmax function, corresponding to the normal operation state and the ground fault state, respectively. The probability of the ground fault state is the fault prediction probability, which serves as the basis for subsequent tripping decisions.

[0052] Specifically, the raw three-phase current sampling data is normalized using the Z-score normalization method based on a sliding time window. First, for each phase current signal, within a preset sliding window including the start-up time (the preset sliding window length is 200 ms), the mean μ and standard deviation σ of the phase current are calculated. Then, the extracted three-phase current waveform data x is normalized using the following formula. i Normalize:

[0053]

[0054] In the formula, x i This represents the original current value at the i-th sampling point. The normalized value. Let be a constant, where is a constant. Specifically, small positive constants are introduced to prevent the denominator from being zero or the value from being unstable. The value is determined based on the quantization accuracy and dynamic range of the current sampling module. For example, when the resolution of the current sampling module is 16 bits and the full scale range is ±5 A, the minimum resolvable current change is approximately:

[0055]

[0056] Considering that the standard deviation σ is typically on the order of magnitude of this resolution, therefore... Set to 10 6 Up to 10 4 Values ​​within a certain range can balance stability and accuracy. In a preferred embodiment, =10 6 It has been verified in extensive field data testing that it does not introduce observable normalization error and can effectively avoid division-to-zero anomalies.

[0057] Finally, the waveform sequences after three-phase normalization are stacked according to the channel dimension to form an input tensor of shape [3, N], where N is the time step, which serves as the input to the convolutional neural network model.

[0058] In a preferred embodiment, the statistical parameters of the sliding window can be dynamically updated to ensure that the normalized benchmark reflects the current system operating status; in another embodiment, the global mean and standard deviation obtained from a large amount of historical data during the offline training phase can also be used for normalization, which is suitable for distribution lines with relatively stable load characteristics.

[0059] In this embodiment, by tightly integrating time window truncation with a deep learning model, high-precision and adaptive identification of distribution network grounding faults is achieved. Compared with traditional methods that rely on zero-sequence voltage or threshold setting, no additional voltage transformer is required. It can effectively identify weak fault characteristics such as high-resistance grounding and intermittent arcing using only three-phase current, significantly improving detection sensitivity. At the same time, the convolutional neural network model can automatically learn and fuse multi-dimensional transient features, avoiding the subjectivity and limitations of manually designed criteria and enhancing the generalization ability for different fault types and different system parameters. In addition, the convolutional neural network model is only called after the start-up conditions are met, balancing real-time performance and computational efficiency, and realizing localized and intelligent distribution network grounding protection.

[0060] 103. Compare the fault prediction probability with the preset probability threshold. When the fault prediction probability is greater than the probability threshold, generate a trip command. The trip command is used to control the circuit breaker connected to the faulty line in the distribution network to trip.

[0061] The preset probability threshold is a pre-defined decision boundary value used to distinguish between normal and fault states. Its value can be adjusted based on the statistical characteristics of historical fault samples or field operation experience. Specifically, after the convolutional neural network model is trained, ROC curves (receiver operating characteristic curves) are plotted on an independent validation dataset to analyze the TPR (true positive rate) and FPR (false positive rate) corresponding to different thresholds. Based on the distribution network's focus on protection performance, the probability value corresponding to the operating point that meets the engineering requirements is selected as the initial threshold. For example, the probability threshold can be set to a value between 0.5 and 0.7. When the system has high reliability requirements, the threshold can be set to 0. 7. This significantly reduces the risk of false tripping due to transient interference. When the system needs to detect high-resistance grounding faults, the threshold can be set to 0.5 to improve sensitivity. In addition, the probability threshold supports on-site online adjustment. Maintenance personnel can fine-tune the probability threshold based on actual operation feedback through the wireless communication module or local maintenance interface, and write the new value into the device configuration storage area so that the updated threshold can be automatically adopted in subsequent judgments. The trip command is a digital or relay contact signal issued by the protection device to drive the circuit breaker to perform the opening operation. The circuit breaker is a switching device installed in the distribution network line that can quickly cut off the current after receiving the trip command to achieve electrical isolation of the fault section.

[0062] Specifically, after obtaining the fault prediction probability output by the convolutional neural network, it is immediately compared with a preset probability threshold. If the fault prediction probability is greater than or equal to the preset probability threshold, it is determined that there is a valid grounding fault in the current line. The system then generates a trip command, which is transmitted to the relay output module through the internal logic circuit, driving the output relay to act and send a trip control signal to the circuit breaker directly connected to the faulty line. After receiving the signal, the circuit breaker completes mechanical tripping, thereby cutting off the faulty section from the distribution network and preventing the fault from further developing or affecting non-faulty areas. The entire judgment process does not require manual intervention and is completed autonomously by the device, ensuring the timeliness and consistency of the action.

[0063] In this embodiment, closed-loop control from intelligent identification to physical isolation is realized. Compared with traditional hard threshold criteria, the tripping logic based on probability thresholds is more flexible and robust, and a balance between sensitivity and reliability can be achieved by adjusting the threshold. At the same time, it avoids false tripping caused by a single feature mutation, significantly improving the selectivity and accuracy of protection. In addition, the tripping command acts directly on the local circuit breaker without relying on communication or upper-level scheduling, effectively shortening the fault clearing time and improving power supply reliability and system security.

[0064] 104. If the ground fault is not detected and isolated within the preset time, the current protection node will be triggered to perform a backup trip operation based on the preset time difference delay.

[0065] Among them, the isolation of a ground fault means that the circuit breaker in the section where the fault point is located (usually controlled by an adjacent protection device located downstream of the current protection node) has been successfully tripped, causing the fault current flowing through the current protection node to disappear or drop to the normal load level, indicating that the fault has been effectively cleared; the preset time refers to the time window used to wait for and verify the isolation result after the local protection action, which is usually 100 to 200 milliseconds, covering the mechanical action time of the circuit breaker and the signal feedback delay; the time difference delay refers to the additional action delay set by the current protection node relative to its downstream adjacent protection node in the multi-level protection configuration of the distribution network (for example, the current node delays by 0.2 seconds and the downstream node delays by 0 seconds) to achieve selective coordination; in this scenario, the current protection node undertakes the backup protection function for its downstream area, used to clear the fault when the downstream protection fails to operate or the circuit breaker fails.

[0066] Specifically, when an adjacent protection node downstream of the current protection node fails to operate as required, the current protection node will not trip immediately during its self-set operating delay. Instead, it will continuously monitor the fault isolation status of its downstream area. During this period, the current protection node will continuously monitor whether the three-phase current has returned to normal through its own current sampling module. If it confirms that the fault has been isolated before the end of its own operating delay, i.e., the downstream circuit breaker has tripped or the fault current has disappeared, the current protection node will automatically reset and will not perform a tripping operation. Conversely, if no signal of successful isolation is detected before the end of its own operating delay, it will determine that the downstream protection has failed to operate or the circuit breaker has malfunctioned. The current protection node will then perform a tripping operation to disconnect the line section between itself and the downstream, thereby ensuring that the fault is ultimately reliably isolated and avoiding the long-term existence of the fault due to protection failure, which could lead to serious consequences such as equipment damage, arc fires, or deterioration of system stability.

[0067] This invention provides a method, device, system, storage medium, and computer equipment for detecting grounding faults in distribution networks. It detects and identifies faults solely based on conventional three-phase current signals, completely independent of zero-sequence voltage transformers or other additional sensors. This avoids maloperation due to the lack of zero-sequence equipment or interference from three-phase imbalance, improving applicability and deployment flexibility. By introducing a preset triggering mechanism and a convolutional neural network model, it can extract high-dimensional coupling features from transient current waveforms, effectively identifying weak fault characteristics such as high-resistance grounding and improving fault identification accuracy. The decision logic of generating tripping commands by comparing fault prediction probability with probability thresholds gives protection actions adjustable sensitivity and strong anti-interference capabilities. By setting a closed-loop mechanism for upstream time-differential backup tripping triggered by fault non-isolation, it can reliably isolate faults even when local protection fails to operate or circuit breakers malfunction, ensuring reliable fault isolation. In summary, the above method, combining pure current drive, intelligent identification, local rapid action, and time-differential backup, achieves high-precision identification of distribution network grounding faults with reduced reliance on specific conditions, and highly reliable small-range isolation of grounding faults.

[0068] Specifically, in the above embodiments, the process of acquiring three-phase current signals on the distribution network lines and determining whether the three-phase current signals meet the fault detection start-up conditions includes: real-time sampling of the three-phase current signals on the distribution network lines and calculating the current change at adjacent sampling points; statistically analyzing the steady-state average value and steady-state standard deviation of the current change within a preset sliding time window, and determining a dynamic start-up threshold based on the steady-state average value, steady-state standard deviation, and a preset sensitivity coefficient; and determining that the three-phase current signals meet the fault detection start-up conditions when the current change at the current sampling point is greater than the dynamic start-up threshold.

[0069] In this embodiment, to achieve efficient and reliable triggering of distribution network grounding faults and avoid false triggering due to transient interference such as load fluctuations, switching operations, or lightning strikes, while ensuring high sensitivity response to weak fault characteristics, a fault detection triggering criterion based on a dynamic threshold adaptive adjustment mechanism is designed. Specifically, firstly, the three-phase current signals of A, B, and C on the distribution network line are sampled at high frequency in real time to obtain raw current waveform data with sufficient resolution. Then, the current change between adjacent sampling points is calculated as a basic indicator reflecting the degree of current change. To further distinguish between normal disturbances and real faults, a preset sliding time window is introduced. Within this window, the steady-state average value and steady-state standard deviation of all current changes are continuously calculated. The steady-state average value represents the current system... The system measures the background noise or normal fluctuation level under normal operating conditions, while the standard deviation reflects the dispersion of the fluctuation. Based on this, combined with a configurable sensitivity coefficient, the dynamic start threshold at the current moment is calculated. The dynamic start threshold has adaptive characteristics: when the system is in a stable operating state, the steady-state average value and steady-state standard deviation are small, and the dynamic start threshold is also small, which is conducive to capturing weak high-resistance grounding faults. When the system experiences heavy load switching or external disturbances, the steady-state average value and steady-state standard deviation automatically increase, and the dynamic start threshold rises accordingly, effectively suppressing false triggering. Finally, if the current change at the current sampling point exceeds the dynamic start threshold, the three-phase current signal is determined to meet the fault detection start condition, triggering the subsequent deep learning recognition process; otherwise, the monitoring state continues.

[0070] Furthermore, based on the steady-state average, steady-state standard deviation, and a preset sensitivity coefficient, the following formula is constructed to determine the dynamic activation threshold:

[0071]

[0072]

[0073]

[0074]

[0075] Where T is the dynamic startup threshold. 稳态 This is the steady-state average value. 稳态 For steady-state standard deviation, I k Let K be the change in current, K be the sensitivity coefficient, N be the number of sampling points in a sliding time window, and k be the number of sampling points.

[0076] In this embodiment, 稳态It represents the average variation of current under normal fault-free conditions, reflecting the natural fluctuation level of system current (e.g., load fluctuation, measurement noise); 稳态 This indicates the range of fluctuations in current change, i.e., the degree of dispersion, which measures the instability of current changes. If the system current noise is very small and the fluctuations are very stable, 稳态 Similarly, the noise level is very low; however, if the system has high noise and frequent voltage fluctuations... 稳态 The value of K will increase; K is a coefficient that adjusts the threshold sensitivity, usually ranging from 2 to 5, depending on the noise level of the system and the sensitivity of the detection. The larger K is, the higher the dynamic trigger threshold, the stronger the anti-interference but the slower the response; the smaller K is, the higher the sensitivity, the more easily it is affected by noise and misjudged; T is the automatic tuning threshold for sudden change detection by the system. The threshold is not a fixed value, but an adaptive threshold that is automatically adjusted according to the current noise on site. Once a sudden change is detected, the subsequent feature extraction and convolution analysis process is started; Based on this, the technical solution of this application does not rely on zero-sequence current or voltage signals and can effectively identify transient disturbances such as arc grounding and high-resistance grounding.

[0077] Specifically, in the above embodiments, the three-phase current waveform data within a preset time period before and after the start-up time is input into a preset convolutional neural network model to obtain the fault prediction probability of the distribution network line being in a ground fault state. This includes: obtaining the time-series representation corresponding to the three-phase current signal, obtaining the three-phase current waveform data, normalizing the three-phase current waveform data, and extracting the data within a preset time period before and after the start-up time to form an input tensor; inputting the input tensor into the preset convolutional neural network model, wherein the convolutional neural network model includes a one-dimensional convolutional layer, a fully connected layer, and an output layer; extracting the coupling transient features of the three-phase current signal in the time dimension through the one-dimensional convolutional layer, and outputting a linear score through the fully connected layer; and converting the linear score into a normal prediction probability of the distribution network being in a normal operating state and a fault prediction probability of the distribution network being in a ground fault state through the output layer using the Softmax function, wherein the sum of the normal prediction probability and the fault prediction probability is 1.

[0078] In this embodiment, to achieve high-precision intelligent identification of grounding faults in the distribution network, once the fault detection initiation condition is met, the system will extract a representative three-phase current waveform data around the initiation moment and input it into a pre-trained convolutional neural network model to output a quantified fault confidence score. Specifically, the system first performs time-series alignment and synchronization processing on the real-time acquired A, B, and C three-phase current signals to form a three-dimensional time-series representation, which includes three channels (corresponding to the three phases) and a sequence of original waveforms with several time steps. Then, the waveform data is normalized to eliminate dimensional differences and improve model convergence stability. Based on this, the system extracts a preset time period before and after the initiation condition (e.g., from 20 ms before the fault to 40 ms after the fault, for a total length of 60 ms) centered on the moment the initiation condition is triggered. The input tensor (ms) forms a fixed-size input tensor that serves as the sole input to the CNN model. The convolutional neural network model consists of sequentially connected one-dimensional convolutional layers, fully connected layers, and an output layer. The one-dimensional convolutional layers employ multiple learnable one-dimensional convolutional kernels to perform a sliding scan of the three-phase current waveform in the time dimension, automatically extracting coupled transient features related to grounding faults. These features include asymmetry in the amplitude changes of the three-phase current, phase shift patterns, harmonic distortion morphology, and implicit features of the zero-sequence component. These features are difficult to manually tune using traditional methods, but the CNN model, through end-to-end training, can adaptively focus on the most discriminative time-frequency combination. The high-dimensional feature map output from the convolutional layers, after pooling or direct flattening, is input to the fully connected layer, mapping local features to a global semantic representation and outputting a linear score vector corresponding to two hypotheses: normal operation and grounding fault state. Finally, the output layer uses the Softmax function to normalize the linear score, converting it into two probability values, where P... 正常 P represents the probability that the model determines the current line is in normal operating condition. 接地 This represents the probability that the model determines the current line is in a ground fault state. The sum of the probabilities of the two is 1, forming a standard probability distribution. The probability output not only provides a clear classification result, but also reflects the confidence level of the model, providing a quantitative basis for subsequent tripping decisions. The entire CNN model has been trained and verified in the offline stage with a large number of measured or simulated samples covering different grounding methods, different fault impedances, different load conditions and interference scenarios to ensure that it has strong generalization ability and robustness in complex field environments. The preset probability threshold is usually 0.5, but it can also be adjusted statistically through training samples to improve reliability.

[0079] Specifically, in the above embodiments, if no ground fault is detected to be isolated within a preset time, the current protection node is triggered based on a preset time difference delay to perform a backup tripping operation, including: sequentially arranging multiple protection nodes along the direction from the load side to the power supply side of the distribution network, wherein the action delay of the adjacent protection node located downstream of the current protection node is less than the action delay of the current protection node; continuously detecting whether the ground fault of the distribution network in the downstream area of ​​the current protection node is isolated during the delay timer of the current protection node; if the ground fault of the distribution network is detected to be isolated before the delay timer of the current protection node ends, the current protection node automatically resets; if the ground fault of the distribution network is not detected to be isolated when the delay timer of the current protection node ends, the current protection node performs a tripping operation.

[0080] Furthermore, whether a ground fault in the distribution network has been isolated is determined by the following method: whether the fault current detected by the current protection node has disappeared.

[0081] In this embodiment, to ensure that the grounding fault of the distribution network can still be reliably isolated in abnormal situations such as downstream protection failure or circuit breaker failure, and to avoid unnecessary power outages caused by cascading trips in non-faulty areas, this application establishes a multi-level backup protection mechanism based on time-level coordination. It relies on multiple intelligent protection nodes arranged sequentially from the load side to the power supply side along the distribution network. Each node has independent fault detection, trip control and communication capabilities, and the action delay is configured to increase step by step according to the electrical topology hierarchy, forming a selective protection timing ladder.

[0082] Specifically, each protection node not only undertakes the main protection function for the line segment it directly manages, but also assumes the backup protection responsibility for the corresponding downstream adjacent area. Among them, the action delay of the adjacent protection node located downstream of the current protection node is less than the action delay of the current protection node (for example, the downstream node delay is 0 seconds, and the current node delay is 0.2 seconds). When a ground fault occurs in the distribution network, the downstream protection node closest to the fault point should take the lead in operating and clearing the fault. If it fails to complete the isolation due to failure to operate or circuit breaker failure, the current protection node continuously monitors the three-phase current flowing through it during its own set action delay period to determine whether the ground fault in its downstream area has been effectively isolated.

[0083] The criteria for determining whether a ground fault is isolated are as follows: whether the fault current detected by the current protection node has disappeared or returned to the normal load level. If the fault current has disappeared during the delay time of the current protection node (e.g., 0.2 seconds), it is determined that the downstream fault has been successfully cleared, and the current protection node automatically resets without performing a trip operation, thereby avoiding maloperation of non-faulty lines. Conversely, if the fault current has not disappeared by the end of the delay time, i.e., the fault characteristic current continues to exist, it is determined that the downstream protection or its circuit breaker has failed to operate, and the current protection node immediately performs a trip operation to disconnect the line section between itself and the downstream as a reliable backup isolation method.

[0084] This application achieves a balance between high selectivity and high reliability by establishing a closed-loop logic that integrates downstream main protection instantaneous action, current node delay backup, and isolation verification based on local current. Under normal operating conditions, only the faulty section is quickly isolated locally, maximizing the continuity of power supply. In the event of equipment failure, the system still has tiered backup capabilities to prevent long-term faults from causing equipment burnout, arcing risks, or deterioration of system stability. Furthermore, the entire mechanism is based entirely on local current sampling and preset time differences, without relying on downstream trip signal feedback or high-speed communication links. This makes it particularly suitable for application scenarios with limited communication conditions but high requirements for power supply reliability, significantly improving the overall robustness and practicality of distribution network grounding fault handling.

[0085] This application also provides a distribution network grounding fault detection device for implementing any of the above-mentioned distribution network grounding fault detection methods, such as... Figure 2 As shown, the system includes the following components connected in series: a current sampling module for synchronously acquiring three-phase current signals; a large-capacity memory for storing current waveform data before and after a ground fault is detected; a main control processing module for running a preset convolutional neural network model to identify ground faults and output fault judgment results; a relay output module for outputting trip commands based on the fault judgment results; a power supply module for converting externally supplied AC or DC power into digital power for digital circuits, analog power for analog circuits, and operating power for relay ports; a wireless communication module for uploading operating status data to a remote distribution network master station and receiving setting adjustment commands, remote tripping commands, and closing commands from the remote distribution network master station; and a self-organizing network communication module for networking multiple fault detection devices and exchanging data with a remote computer.

[0086] The distribution network grounding fault detection device provided in this application adopts a modular architecture. Each functional unit is connected in series via an internal bus or signal path to collaboratively complete the entire process from current acquisition and intelligent identification to tripping control and remote communication. Specifically, the device includes a current sampling module, a large-capacity memory, a main control processing module, a relay output module, a power supply module, a wireless communication module, and a self-organizing network communication module. The current sampling module, as the front-end sensing unit of the device, is equipped with a high-precision current transformer and a high-speed analog-to-digital converter circuit. It is used to synchronously and at high frequency sample the three-phase current signals (IA, IA', IB, IB', IC, IC') of the A, B, and C phases in the distribution network line (typical sampling frequency not less than 10 Hz). The system uses a current sampling module (kHz) to acquire raw electrical quantity data reflecting the system's operating status and transmits the digitized current waveform to subsequent modules in real time. A large-capacity memory connected to the current sampling module automatically saves complete three-phase current waveform data within a preset time period before and after the ground fault occurs, after the main control processing module determines that a ground fault has occurred. The stored data can be used not only for local fault backtracking analysis but also uploaded to a remote master station in a standard format (e.g., Comtrade) to provide maintenance personnel with accurate fault recording data. The main control processing module is the core control and intelligent decision-making unit of the entire device. It has a pre-trained convolutional neural network embedded in it, which receives real-time waveform data from the current sampling module. When the fault detection start condition is met, it calls the CNN model to perform deep identification of the ground fault, outputs a predicted probability representing the fault confidence level, and generates the final fault judgment result based on a preset threshold. The relay output module is electrically connected to the main control processing module to receive the fault judgment result. Upon confirmation of a ground fault, the internal output relay is activated, outputting a trip command in the form of a hard contact, directly controlling the circuit breaker connected to the faulty line to trip, achieving rapid local isolation. The power supply module provides stable and reliable power support for the entire device, compatible with AC or DC input, suitable for different types of distribution network ring network cabinets or switch station environments, ensuring normal operation of the device under various power supply conditions. The wireless communication module is connected to the main control processing module to upload the device's operating status data (including switch quantities, telemetry, SOE events, etc.) in real time, and to receive remote tripping or closing commands from remote computers or dispatch master stations, supporting remote monitoring and emergency operation. The self-organizing network communication module further enhances the device's networking capability, enabling the establishment of low-power, highly reliable data links with other protection nodes or communication management units, aggregating local information and forwarding it to remote computers through a self-organizing network, suitable for power distribution scenarios without public network coverage or requiring the construction of dedicated communication channels.In summary, through the organic integration of the above modules, this device achieves ground fault protection functions such as local sensing, intelligent identification, rapid action, and remote interaction. It does not rely on zero-sequence voltage transformers and can complete high-precision fault detection and reliable isolation based solely on conventional three-phase current, significantly improving the safe operation level of the distribution network under complex grounding conditions.

[0087] In addition, the power distribution network grounding fault detection device can also be equipped with an Ethernet communication interface, specifically a standard RJ45 network socket. The interface connects to the main control processing module and supports wired Ethernet communication based on the TCP / IP protocol. This enables high-speed data interaction between the device and the substation monitoring system, communication management unit, or local maintenance terminal. Through the RJ45 interface, fault waveform data, protection action events, device self-test status, and other information can be uploaded in real time. It also supports remote parameter configuration, model updates, and debugging operations. In power distribution automation scenarios with fiber optic or industrial Ethernet cabling capabilities, the interface can serve as a supplement or alternative to the wireless communication module, further improving communication stability and bandwidth capabilities, and meeting the access requirements of smart substation communication standards such as IEC61850.

[0088] Furthermore, as Figure 1 To specifically implement the method, this application provides a grounding fault detection system, such as... Figure 3 As shown, the device includes: a fault detection start-up unit 201, a prediction probability output unit 202, a fault line clearing unit 203, and a differential backup tripping unit 204.

[0089] The fault detection start-up unit 201 is used to collect the three-phase current signal on the distribution network line and determine whether the three-phase current signal meets the fault detection start-up conditions.

[0090] The prediction probability output unit 202 is used to input the three-phase current waveform data within a preset time before and after the start time into a preset convolutional neural network model when the three-phase current signal meets the fault detection start conditions, so as to obtain the fault prediction probability that the distribution network line is in a ground fault state.

[0091] The fault line disconnection unit 203 is used to compare the fault prediction probability with a preset probability threshold. When the fault prediction probability is greater than the probability threshold, a trip command is generated. The trip command is used to control the circuit breaker connected to the fault line in the distribution network line to open.

[0092] The differential backup trip unit 204 is used to trigger the current protection node to perform backup trip operation based on a preset time differential delay if no ground fault is detected and isolated within a preset time.

[0093] In specific application scenarios, the fault detection start-up unit 201 is specifically used to sample the three-phase current signal on the distribution network line in real time, calculate the current change at adjacent sampling points, statistically analyze the steady-state average value and steady-state standard deviation of the current change within a preset sliding time window, and determine the dynamic start-up threshold based on the steady-state average value, steady-state standard deviation and preset sensitivity coefficient; when the current change at the current sampling point is greater than the dynamic start-up threshold, it is determined that the three-phase current signal meets the fault detection start-up conditions.

[0094] In specific application scenarios, the fault detection activation unit 201 can also be used to construct the following formula based on the steady-state average value, steady-state standard deviation, and preset sensitivity coefficient to determine the dynamic activation threshold:

[0095]

[0096]

[0097]

[0098]

[0099] Where T is the dynamic startup threshold. 稳态 This is the steady-state average value. 稳态 For steady-state standard deviation, I k Let K be the change in current, K be the sensitivity coefficient, N be the number of sampling points in a sliding time window, and k be the number of sampling points.

[0100] In specific application scenarios, the prediction probability output unit 202 can be used to obtain the time-series representation of the three-phase current signal, obtain the three-phase current waveform data, normalize the three-phase current waveform data, and extract data within a preset time period before and after the start-up time to form an input tensor; input the input tensor into a preset convolutional neural network model, wherein the convolutional neural network model includes a one-dimensional convolutional layer, a fully connected layer, and an output layer; the one-dimensional convolutional layer extracts the coupled transient features of the three-phase current signal in the time dimension, and the fully connected layer outputs a linear score; the output layer uses the Softmax function to convert the linear score into the normal prediction probability of the distribution network being in normal operation and the fault prediction probability of the distribution network being in a ground fault state, wherein the sum of the normal prediction probability and the fault prediction probability is 1.

[0101] In specific application scenarios, the tiered backup tripping unit 204 can be used to sequentially arrange multiple protection nodes along the load side to the power supply side of the distribution network. Each protection node is configured with progressively increasing action delays, wherein the action delay of the adjacent protection node downstream of the current protection node is less than the action delay of the current protection node. During the delay timer of the current protection node, it continuously detects whether the distribution network ground fault in the downstream area of ​​the current protection node has been isolated. If the ground fault in the distribution network is detected to be isolated before the delay timer of the current protection node ends, the current protection node automatically resets. If the ground fault in the distribution network is not detected to be isolated when the delay timer of the current protection node ends, the current protection node performs a tripping operation.

[0102] In specific application scenarios, the differential backup trip unit 204 can also be used to determine whether the ground fault in the distribution network has been isolated by the following method: whether the fault current detected by the current protection node has disappeared.

[0103] It should be noted that other corresponding descriptions of the functional units involved in the power distribution network grounding fault detection system provided in this embodiment can be found in [reference]. Figure 1 The corresponding descriptions in [the document] will not be repeated here.

[0104] Based on the above, Figure 1 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method for detecting grounding faults in power distribution networks.

[0105] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), including several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the ground fault detection method of various implementation scenarios of this application.

[0106] Based on the above, Figure 1 The method shown, and Figure 3 The illustrated embodiment of the power distribution network grounding fault detection device, in order to achieve the above objectives, such as... Figure 4 As shown, this embodiment also provides a physical device for detecting grounding faults in a power distribution network. This device includes a communication bus, a processor, a memory, and a communication interface. It may also include input / output interfaces and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the power distribution network grounding fault detection method described in the above embodiment.

[0107] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0108] Those skilled in the art will understand that the physical device structure for detecting grounding faults in a power distribution network provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0109] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0110] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the technical solution of this application, fault detection and identification are performed based solely on conventional three-phase current signals, without relying on zero-sequence voltage transformers or other additional sensors, thus avoiding the problem of false tripping due to the lack of zero-sequence equipment or interference from three-phase imbalance, improving applicability and deployment flexibility; by introducing a preset start-up condition triggering mechanism combined with a convolutional neural network model, high-dimensional coupling features can be extracted from transient current waveforms, effectively identifying weak fault features such as high-resistance grounding, and improving fault identification accuracy; the decision logic of generating tripping commands by comparing fault prediction probability and probability threshold enables the protection action to have adjustable sensitivity and strong anti-interference capability; by setting a closed-loop mechanism for upstream time-level backup tripping triggered by fault non-isolation, the fault can still be reliably cleared when the local protection fails to operate or the circuit breaker fails, ensuring the reliability of fault clearing. In summary, the above method combines pure current drive, intelligent identification, local rapid action and differential backup to achieve high-precision identification of grounding faults in distribution networks with reduced dependence on conditions, as well as high-reliability small-scale isolation of grounding faults.

[0111] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0112] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for detecting grounding faults in a power distribution network, characterized in that, include: Collect the three-phase current signals on the distribution network lines and determine whether the three-phase current signals meet the fault detection start conditions; When the three-phase current signal meets the fault detection start-up condition, the three-phase current waveform data within a preset time before and after the start-up time is input into a preset convolutional neural network model. The convolutional neural network model directly outputs the fault prediction probability that the distribution network line is in a ground fault state based on the input tensor formed by the three-phase current waveform data. The fault prediction probability is compared with a preset probability threshold. When the fault prediction probability is greater than the probability threshold, a trip command is generated. The trip command is used to control the circuit breaker connected to the faulty line in the distribution network line to trip. If the ground fault is not detected and isolated within the preset time, the current protection node is triggered to perform a backup trip operation based on the preset time difference delay. If no ground fault is detected and isolated within a preset time, the current protection node is triggered to perform a backup trip operation based on a preset time difference delay, including: Multiple protection nodes are arranged sequentially along the direction from the load side to the power supply side of the distribution network. Each protection node is configured with an incrementally increasing action delay. The action delay of the adjacent protection node located downstream of the current protection node is less than the action delay of the current protection node. During the delay time of the current protection node, it is continuously detected whether the grounding fault of the distribution network in the downstream area of ​​the current protection node has been isolated; If a ground fault in the distribution network is detected to be isolated before the delay timer of the current protection node ends, the current protection node will automatically reset. If the ground fault in the distribution network is not isolated when the delay timer of the current protection node ends, the current protection node will perform a trip operation.

2. The method according to claim 1, characterized in that, The process of acquiring three-phase current signals from the distribution network lines and determining whether the three-phase current signals meet the fault detection activation conditions includes: Real-time sampling of three-phase current signals on distribution network lines; calculation of current changes at adjacent sampling points. Within a preset sliding time window, the steady-state average value and steady-state standard deviation of the current change are statistically analyzed, and a dynamic start-up threshold is determined based on the steady-state average value, the steady-state standard deviation, and a preset sensitivity coefficient. When the change in current at the current sampling point is greater than the dynamic start threshold, it is determined that the three-phase current signal meets the fault detection start condition.

3. The method according to claim 2, characterized in that, The determination of the dynamic start-up threshold based on the steady-state average value, the steady-state standard deviation, and a preset sensitivity coefficient includes: Based on the steady-state average value, the steady-state standard deviation, and the preset sensitivity coefficient, the following formula is constructed to determine the dynamic start-up threshold: Where T is the dynamic startup threshold. 稳态 The steady-state average value is... 稳态 Let be the steady-state standard deviation. I k Let K be the change in current, K be the sensitivity coefficient, N be the number of sampling points in a sliding time window, and k be the number of sampling points.

4. The method according to claim 1, characterized in that, The step of inputting three-phase current waveform data within a preset time period before and after the start-up time into a preset convolutional neural network model, wherein the convolutional neural network model directly outputs the fault prediction probability of the distribution network line being in a ground fault state based on the input tensor formed by the three-phase current waveform data, includes: The timing representation corresponding to the three-phase current signal is obtained to obtain the three-phase current waveform data. The three-phase current waveform data is normalized, and the data within a preset time before and after the start-up time is extracted to form an input tensor. The input tensor is input into a preset convolutional neural network model, wherein the convolutional neural network model includes a one-dimensional convolutional layer, a fully connected layer, and an output layer; The coupling transient features of the three-phase current signal are extracted in the time dimension through the one-dimensional convolutional layer, and a linear score is output through the fully connected layer. The output layer uses the Softmax function to convert the linear score into a normal prediction probability of the distribution network being in normal operation and a fault prediction probability of the distribution network being in a ground fault state, wherein the sum of the normal prediction probability and the fault prediction probability is 1.

5. The method according to claim 1, characterized in that, Whether the grounding fault in the power distribution network has been isolated is determined by the following methods: Whether the fault current detected by the current protection node has disappeared.

6. A grounding fault detection device for a power distribution network, characterized in that, The device is used to implement the power distribution network grounding fault detection method as described in any one of claims 1 to 5, comprising the following series connections: The current sampling module is used to synchronously acquire three-phase current signals; A large-capacity memory is used to store current waveform data before and after the time of the ground fault when a ground fault is detected. The main control processing module is used to run a preset convolutional neural network model to identify grounding faults and output the fault judgment results. The relay output module is used to output a trip command based on the fault judgment result. The power supply module is used to convert externally supplied AC or DC power into digital power for digital circuits, analog power for analog circuits, and operating power for relay ports. The wireless communication module is used to upload operating status data to the remote distribution network master station and receive setting adjustment instructions, remote control opening instructions and closing instructions issued by the remote distribution network master station; The self-organizing network communication module is used to network multiple fault detection devices and exchange data with a remote computer.

7. A power distribution network grounding fault detection system, characterized in that, The system includes: The fault detection start-up unit is used to collect the three-phase current signals on the distribution network line and determine whether the three-phase current signals meet the fault detection start-up conditions. The prediction probability output unit is used to input the three-phase current waveform data within a preset time before and after the start time into a preset convolutional neural network model when the three-phase current signal meets the fault detection start condition. The convolutional neural network model directly outputs the fault prediction probability that the distribution network line is in a ground fault state based on the input tensor formed by the three-phase current waveform data. The fault line disconnection unit is used to compare the fault prediction probability with a preset probability threshold. When the fault prediction probability is greater than the probability threshold, a trip command is generated. The trip command is used to control the circuit breaker connected to the fault line in the distribution network line to trip. The differential backup trip unit is used to trigger the current protection node to perform backup trip operation based on a preset time differential delay if no ground fault is detected and isolated within a preset time. The differential backup tripping unit is also used to arrange multiple protection nodes sequentially along the direction from the load side to the power supply side of the distribution network. The protection nodes are configured with progressively increasing action delays, wherein the action delay of the adjacent protection node located downstream of the current protection node is less than the action delay of the current protection node. During the delay time of the current protection node, it is continuously detected whether the grounding fault of the distribution network in the downstream area of ​​the current protection node has been isolated; If a ground fault in the distribution network is detected to be isolated before the delay timer of the current protection node ends, the current protection node will automatically reset. If the ground fault in the distribution network is not isolated when the delay timer of the current protection node ends, the current protection node will perform a trip operation.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images