Coal mine electric leakage protection method, device, equipment and program product based on electric leakage fault identification model

Abstract

This disclosure provides a coal mine leakage current protection method based on a leakage current fault identification model, comprising: collecting electrical signals from various lines of the power grid; when the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, extracting the sampling values ​​of the electrical signals for N cycles before and after the fault occurrence time, and obtaining the data to be identified after preprocessing; inputting the data to be identified into a pre-trained leakage current identification model to obtain the leakage current fault identification result; and issuing a fault clearing command to the protection unit of the corresponding line according to the leakage current fault identification result. The scheme of this disclosure uses an AI model combined with the conditional attention mechanism of the power grid topology to dynamically adjust the feature weights, and introduces a dynamic routing controller in the feature learning stage to switch the optimized feature extraction branch in real time according to the fault severity. Compared with traditional methods, this scheme significantly improves the leakage current detection accuracy and significantly shortens the fault response time under complex underground working conditions.

Description

Technical Field

[0001] This disclosure relates to the field of power supply technology, and in particular to a method, device, equipment and program product for coal mine leakage protection based on a leakage fault identification model. Background Technology

[0002] Coal mine power grids, including 10kV, 6kV, and 1140V systems, often employ neutral-point non-effectively grounded systems, encompassing both ungrounded neutral systems and systems grounded via arc suppression coils. Single-phase leakage faults (or single-phase grounding faults) are the most frequent electrical faults in coal mine power grids. When a single-phase leakage fault occurs, it causes an increase in both the neutral point voltage and the voltage of the ungrounded phases, easily developing into a short-circuit fault that leads to widespread power outages in the mine. It can also generate electrical sparks that can trigger gas explosions, causing major safety accidents.

[0003] The detection and protection technology for single-phase grounding in neutral point non-effective grounding systems has always been a research hotspot. In recent years, the use of artificial intelligence to detect single-phase leakage current in coal mine power grids has also become a reality. Given the special nature of the coal mine production environment, highly reliable leakage current detection technology is required. Summary of the Invention

[0004] This invention provides a leakage current protection scheme for coal mines based on a leakage current fault identification model, in order to solve the problem of insufficient reliability of single-phase grounding detection and protection technology in coal mine neutral point non-effective grounding systems.

[0005] The present invention solves the above-mentioned technical problems through the following aspects:

[0006] In a first aspect, the present invention provides a method for leakage current protection in coal mines based on a leakage current fault identification model, comprising:

[0007] Collect electrical signals from each line of the power grid, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current;

[0008] If the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, the sampled values ​​of the power signal are extracted N cycles before and after the fault occurrence time, and the data to be identified is obtained after preprocessing.

[0009] The data to be identified is input into a pre-trained leakage current identification model to obtain leakage current fault identification results;

[0010] Based on the leakage fault identification result, a fault clearing command is issued to the protection unit of the corresponding line.

[0011] Secondly, the present invention provides a coal mine leakage protection device based on a leakage fault identification model, comprising:

[0012] The acquisition module is used to acquire electrical signals from various power grid lines, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current.

[0013] The data acquisition module is used to extract the sampled values ​​of the power signal N cycles before and after the fault occurrence when the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, and obtain the data to be identified after preprocessing.

[0014] The detection module is used to input the data to be identified into a pre-trained leakage current identification model to obtain leakage current fault identification results;

[0015] The control module is used to issue a fault clearing command to the protection unit of the corresponding line based on the leakage fault identification result.

[0016] Thirdly, this invention provides a coal mine leakage current protection device based on a leakage current fault identification model, installed in a power grid line. The protection device includes a power grid interface, a data interface, a processor, and a memory.

[0017] The power grid interface is connected to the processor and is used to connect to the power grid lines;

[0018] The data interface is connected to the processor and is used to communicate with the management platform;

[0019] The memory stores instructions that can be executed by the processor to implement the aforementioned protection method.

[0020] Fourthly, the present invention provides a computer program that, when executed by a processor, causes the processor to implement the aforementioned protection method.

[0021] This invention monitors electrical signals in the power grid and uses AI to identify faults by capturing sampled values ​​from multiple periods before and after a fault occurs. It fuses multi-domain features through spatiotemporal joint encoding and dynamically adjusts feature weights using a conditional attention mechanism based on the power grid topology. Innovatively, a dynamic routing controller is introduced in the feature learning stage to switch optimized feature extraction branches in real time based on fault severity. Through a dynamic parameter adaptation mechanism using an extreme learning machine, collaborative optimization of fault classification and understanding is achieved. The AI ​​model employed in this invention features a dynamic attention weighting layer that encodes power grid structural information into attention weights, ensuring that the feature learning process conforms to the propagation laws of the power system and overcoming the shortcomings of traditional methods that neglect topological correlations. The AI ​​model designed in this invention incorporates multiple learning branches, which can focus on current change rate features for minor leakage, harmonic distortion features for severe leakage, and waveform abrupt changes for transient faults. This flexible feature adaptation mechanism achieves a leap from "perceiving faults" to "understanding faults." Compared with traditional methods, this solution improves the accuracy of leakage current detection by 15.7% under complex downhole conditions, achieves topology node-level positioning accuracy, and reduces the response time to less than 30ms. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of a leakage current protection method based on a leakage current fault identification model provided in this disclosure embodiment;

[0024] Figure 2 This is a schematic diagram of the structure of the leakage fault identification model in the embodiments of this disclosure;

[0025] Figure 3 This is a schematic diagram of the ELM network structure;

[0026] Figure 4 Structural block diagram of a coal mine leakage protection device 300 based on a leakage fault identification model provided in this embodiment of the present disclosure;

[0027] Figure 5 This is a schematic diagram of the structure of the protection device 400 provided in an embodiment of the present disclosure. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this disclosure. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure. Furthermore, for clarity, parts unrelated to the described exemplary embodiments have been omitted from the drawings.

[0029] In this specification, it should be understood that terms such as "comprising" or "having" are intended to indicate the presence of features, figures, steps, behaviors, components, portions, or combinations thereof disclosed in this disclosure, and are not intended to exclude the possibility of one or more other features, figures, steps, behaviors, components, portions, or combinations thereof being present or added. It should also be noted that, unless otherwise specified, embodiments and features within embodiments of this disclosure can be combined with each other.

[0030] Leakage faults in coal mine power supply systems include arcing ground faults, high-resistance ground faults, metallic ground faults, transient ground faults, and intermittent ground faults. Commonly identifiable fault characteristics when leakage faults occur include: harmonic characteristics, fifth harmonic characteristics, first half-wave characteristics, zero-sequence voltage amplitude characteristics, zero-sequence current amplitude characteristics, phase characteristics, zero-sequence active power characteristics, zero-sequence admittance characteristics, S-injection method characteristics, and medium-resistance ground fault characteristics.

[0031] The manifestations and functions of these fault characteristics are explained below.

[0032] Zero-sequence voltage amplitude characteristics: When a single-phase grounding occurs in a neutral point ungrounded system or a neutral point grounded system via an arc suppression coil, the zero-sequence voltage can be detected throughout the entire power supply system. The magnitude of the zero-sequence voltage is related to the grounding resistance. When the grounding resistance is high, the zero-sequence voltage is small. When the grounding is metallic, the zero-sequence voltage is the largest, close to the voltage of the power supply line, and cannot locate faulty or non-faulty lines.

[0033] Zero-sequence current amplitude characteristics: In a neutral-point ungrounded system, when a single-phase ground fault occurs, the zero-sequence current of the faulted line is equal to the sum of the ground capacitance currents of all non-faulted lines in the system; in a neutral-point grounded system via an arc suppression coil, when a single-phase ground fault occurs, the zero-sequence current of the faulted line is equal to the sum of the capacitive current of the faulted line itself and the compensation current of the arc suppression coil of the power supply system minus the sum of the capacitive currents of other non-faulted lines.

[0034] Phase characteristics: In a neutral-point ungrounded system, when a single-phase ground fault occurs, the direction of the zero-sequence current in the faulty line is from the line to the busbar, while the direction of the zero-sequence current in the non-faulty line is from the busbar to the line. The directions of the zero-sequence current in the faulty and non-faulty lines are exactly opposite. In a neutral-point grounded system with an arc-suppression coil, when a single-phase ground fault occurs, the arc-suppression coil is generally overcompensated. The direction of the zero-sequence current in both the faulty and non-faulty lines is from the busbar to the line, making it impossible to distinguish between the faulty and non-faulty lines.

[0035] Zero-sequence active current characteristics: The active current formed based on the resistive characteristics of the grounding point of the faulted line is related to the grounding resistance of the grounding line. When the grounding resistance is high, the zero-sequence active current is small, and when the grounding resistance is metallic, the zero-sequence active current is large.

[0036] Zero-sequence admittance characteristics: In a neutral-point ungrounded system, when a single-phase ground fault occurs, the zero-sequence admittance of the faulty line is equal to the sum of the zero-sequence admittances to ground of all non-faulty lines in the system. In a neutral-point grounded system via an arc-suppression coil, when a single-phase ground fault occurs, the zero-sequence admittance of the faulty line is equal to the sum of the zero-sequence admittance of the faulty line itself and the zero-sequence admittance of the arc-suppression coil in the power supply system, minus the sum of the zero-sequence admittances of other non-faulty lines.

[0037] Harmonic characteristics or high-frequency component characteristics: During metallic grounding faults, the fundamental current increases significantly, while the harmonic components are relatively few. During arc grounding, instantaneous grounding, or intermittent grounding faults, high-frequency harmonic components are mainly generated, primarily odd harmonics, with the third, fifth, and seventh harmonics having the highest harmonic content.

[0038] Fifth harmonic characteristics: The fifth harmonic in coal mine power supply is mainly generated by leakage faults in equipment such as frequency converters and rectifiers. In a neutral ungrounded system, when a single-phase ground fault occurs, the fifth harmonic of the faulty line is much larger than that of the non-faulty line.

[0039] First Half-Wave Characteristics: When a single-phase ground fault occurs, a significant abrupt change occurs in the first half-wave of the current or voltage waveform. This abrupt change is caused by an abnormal increase in current or an abnormal decrease in voltage at the fault point. The first half-wave characteristics mainly refer to the changes in the current or voltage waveform at the instant the fault occurs. During the first half-wave, in systems with an ungrounded neutral point or systems with a neutral point grounded through an arc suppression coil, the zero-sequence current waveform of the faulted branch differs significantly from that of the non-faulted branch. In systems with a neutral point grounded through an arc suppression coil, the zero-sequence current in the first half-wave is less affected by the arc suppression coil (after the arc suppression coil compensation time is greater than half a cycle).

[0040] The characteristics of the S-injection method are as follows: In the event of a single-phase ground fault, a non-power frequency signal, mainly between 80Hz and 200Hz, is injected into the power supply system through a PT (Power Placement) system. The injected signal is detected and tracked by the protection terminal to achieve line selection and tripping. However, the injected non-power frequency signal is easily interfered with by fault signals and harmonics. Especially during a single-phase ground fault, the non-power frequency signal generated by the system is much larger than the injected signal, making it impossible to accurately locate the faulty and non-faulty lines.

[0041] Medium-resistance grounding characteristics: In single-phase grounding, a medium-resistance resistor is connected between the neutral point and ground. A current loop is formed through the earth and the grounding point of the faulty line, limiting the magnitude of the single-phase grounding fault current, preventing arc reignition, and suppressing system overvoltage. The magnitude of the resistive current at the grounding point of the faulty line is used to locate the faulty and non-faulty lines, but this method has limitations for high-resistance grounding.

[0042] This invention establishes a feature library based on a large number of leakage faults, pre-trains an AI model, and collects three-phase voltage and current, as well as zero-sequence voltage and current. When the monitored value of the zero-sequence voltage exceeds a preset range, the pre-trained AI model is used for intelligent identification, achieving accurate determination of leakage faults using artificial intelligence methods. The implementation scheme of this invention is described in detail below.

[0043] Figure 1 A flowchart of a leakage current protection method based on a leakage current fault identification model provided in an embodiment of this disclosure.

[0044] like Figure 1 As shown, the method includes steps S110 to S140.

[0045] S110: Collects electrical signals from various power grid lines, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current.

[0046] S120: When the zero-sequence voltage amplitude of any line is detected to be greater than the preset value, the sampled values ​​of the power signal are extracted N cycles before and after the fault occurs, and the data to be identified is obtained after preprocessing.

[0047] S130: Input the data to be identified into the pre-trained leakage current identification model to obtain the leakage current fault identification result.

[0048] S140: Issues a fault clearing command to the protection unit of the corresponding line based on the leakage fault identification result.

[0049] According to the method of this embodiment, the power signal can be acquired by a protection unit installed in the power grid line, or by other sensors or acquisition devices in the power grid. Since zero-sequence voltage is usually only related to grounding resistance, while zero-sequence current (leakage current) is related to grounding resistance, grounding method (whether the neutral point has an arc suppression coil), and the magnitude of the capacitive current of the system cable to ground, the amplitude of the zero-sequence voltage can be used to monitor whether leakage has occurred. For example, when the grounding resistance is 2kΩ, the zero-sequence voltage of the open delta connection is slightly higher than 15V. Therefore, the preset value of the zero-sequence voltage amplitude can be set to 15V to detect minor leakage faults. When a leakage fault occurs, it is always accompanied by an increase in zero-sequence voltage. A zero-sequence voltage of 15V can reflect the occurrence of leakage, but it cannot accurately determine the line where leakage has occurred. The leakage line still needs to be identified through a leakage fault identification model.

[0050] The coal mine power grid system is powered by three-phase AC, consisting of three sinusoidal AC currents with the same frequency, equal amplitude, and a phase difference of 120°. These currents are transmitted through primary three-phase voltage transformers, current transformers, and zero-sequence current transformers in the high-voltage switch, and through peripheral signal conditioning hardware in the secondary protection terminals. When acquiring electrical signals, an analog-to-digital converter (ADC) is used to convert the analog signals of the power grid into digital signals. Higher sampling frequencies reflect waveform accuracy more accurately, but also require higher computational power. For example, selecting a 12800Hz sampling frequency allows for 256 points to be collected per cycle. The acquired zero-sequence voltage sampling signal can be used to calculate the zero-sequence voltage amplitude in real time using a discrete Fourier series algorithm. When the zero-sequence voltage amplitude exceeds a preset value, such as 15V, N cycles of sampling values ​​of all voltage and current signals (three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current) before the fault and N cycles of sampling values ​​of all voltage and current signals after the fault are recorded. These 2N cycles of sampling values ​​are then used as raw data and preprocessed to obtain the data to be identified. Preferably, N can be set to 5, meaning that 10 cycles of sampled values ​​are preprocessed. Acquiring N cycles of signals before and after a fault can reduce false alarms caused by equipment startup or strong electrical disturbances in the power grid. The preprocessing of the sampled signals includes bandpass filtering, wavelet denoising, sliding window alignment, and 3σ outlier removal for each electrical signal sample value.

[0051] Figure 2 This is a schematic diagram of the structure of the leakage fault identification model in this embodiment of the present disclosure.

[0052] like Figure 2 As shown, the leakage fault identification model includes a feature extraction module, a spatiotemporal joint coding module, a dynamic routing controller, and multiple feature learning branches.

[0053] The feature extraction module is used to extract multi-domain features from the data to be identified. These multi-domain features include time-domain waveforms, frequency-domain harmonic energy distribution, and statistical domain abrupt change parameters.

[0054] Multi-scale, multi-directional convolution operations are performed on the leakage current data to be identified, and the basic elements of the fault features extracted mainly include:

[0055] 1) Waveform Changes: Waveform changes refer to the shape of current or voltage signals as they change over time. Leakage faults often cause waveform distortion, jitter, or abrupt changes. These changes can be captured by the model through convolution operations and serve as an important basis for fault diagnosis.

[0056] 2) Frequency characteristics: Frequency characteristics are the signal's behavior in the frequency domain, reflecting the distribution and intensity of different frequency components in the signal. Leakage faults may cause changes in frequency components, such as increased harmonics and shifts in the dominant frequency. These characteristics can also be extracted by the model through convolution operations.

[0057] 3) Current Intensity and Rate of Change: By monitoring the current intensity and its rate of change over time, the model can detect abnormal current fluctuations, which are important indicators of leakage faults. Convolutional kernels can be designed as filters sensitive to current intensity and rate of change over time to extract these features.

[0058] 4) Voltage Level and Fluctuations: Voltage stability is a key indicator for power grid operation. Model analysis examines voltage levels and fluctuations to identify potential voltage anomalies. Convolutional kernels can capture the stability and volatility of voltage signals, providing strong evidence for fault diagnosis.

[0059] 5) Power Factor and Reactive Power: The power factor reflects the ratio of usable power to apparent power in the power grid, while the flow of reactive power can lead to a decrease in grid efficiency. By monitoring these parameters, the model helps to detect power factor changes caused by leakage and abnormal reactive power flows. Convolutional kernels can be designed as filters sensitive to changes in power factor and reactive power.

[0060] 6) Harmonic Content and Distribution: Harmonics in the power grid are generated by nonlinear loads and can damage grid equipment. Model analysis is used to determine the harmonic content and distribution to assess the impact of leakage faults on the harmonic characteristics of the power grid. Convolutional kernels can capture the frequency components and amplitude of harmonic signals, providing harmonic characteristics for fault diagnosis.

[0061] 7) Phase Angle and Phase Difference: The phase angle is the angle between current and voltage, while the phase difference is the angular difference between different phases. By monitoring these parameters, the model can identify phase anomalies, which may be related to leakage faults. Convolutional kernels can be designed as filters sensitive to changes in phase angle and phase difference.

[0062] In this embodiment, feature extraction mainly involves extracting the current rise rate and voltage drop depth from the time-domain waveform, calculating the energy proportion of the 3rd to 11th harmonics in the frequency domain, detecting abrupt changes within a sliding window using the CUSUM algorithm, and calculating the peak-to-peak variance to obtain the statistical domain abrupt change parameters. Adaptive notch filtering is also applied to the time-domain waveform to eliminate power frequency interference; frequency domain analysis employs windowed FFT and subband energy normalization.

[0063] The spatiotemporal joint coding module is used to jointly encode multi-domain features to obtain coded features. Multi-dimensional coded features can be generated through temporal kurtosis analysis, frequency harmonic analysis, and statistical domain abrupt change parameters.

[0064] The dynamic routing controller activates corresponding feature learning branches based on real-time evaluation metrics. These branches learn the encoded features. Multiple feature learning branches are implemented using an SAE (Stacked Auto Encoder) model. These branches respectively learn features for minor leakage, severe leakage, and transient faults. Specifically, the first branch focuses on current change rate features, using a convolutional neural network to extract these features and activating when the current change rate exceeds a first threshold; the second branch focuses on harmonic distortion features, using a frequency domain attention network to extract these features and activating when the total harmonic distortion rate exceeds a second threshold; and the third branch captures waveform abrupt changes, using a long short-term memory network to extract these features and activating when the density of abrupt change points exceeds a third threshold.

[0065] The activation thresholds for the dynamic routing controller can be set as follows: for example, a current change rate threshold of 5A / ms indicates a minor leakage, activating the first branch; a total harmonic distortion (THD) greater than 5% indicates a more serious leakage, activating the second branch; and a sudden change point density threshold of 3 per cycle indicates a transient fault, activating the third branch. Alternatively, the activation thresholds for the dynamic routing controller can be set based on the range of the zero-sequence voltage amplitude. For example, a zero-sequence voltage of 15V to 30V indicates a minor leakage, activating the first branch; a zero-sequence voltage of 30V to 90V indicates a more serious leakage, activating the second branch; and a zero-sequence voltage greater than 90V (severe leakage, the fault cannot persist for a long time), indicates a transient fault, activating the third branch.

[0066] In the leakage fault identification model of this embodiment, the spatiotemporal joint coding module is also used to generate conditional attention weights based on the adjacency matrix of the power grid topology, dynamically weighting the encoded features. The adjacency matrix is ​​a 0-1 matrix generated based on the coal mine power supply network topology (adjacent nodes are 1, non-adjacent nodes are 0). Dynamic weighting using the power grid topology effectively adapts to the power grid structure, calculates the distance to the fault node based on the fault feature propagation delay, focuses on lines along the fault propagation path, improves the accuracy of fault location, and reduces the computation time for irrelevant lines. In the conditional attention mechanism, the key is the basic element of the fault feature, and the value is the corresponding leakage fault. By calculating the similarity between the key and the query, the model can dynamically adjust the attention given to different features, thereby more accurately identifying leakage faults. The conditional attention weights are calculated using the following formula:

[0067]

[0068] Where, α ij W is the attention weight of node i to node j. q It is a query matrix, W k It is a bond matrix, A ij σ is the element of the grid topology adjacency matrix, λ is the topology influence coefficient, and σ is the nonlinear activation function, preferably GELU.

[0069] During calculation, Used to stabilize gradient (d) k W (where W is the dimension of the key vector) q and W k All are trainable parameters, λA ij This allows the model to autonomously learn the correlation between electrical parameters and topology, and GELU activation enhances its nonlinear expressive power. ij The value of is 0 or 1, representing the connectivity of the node. The topology influence coefficient λ is a hyperparameter, an explicit coupling term in the grid adjacency matrix, and its value can range from [0.1, 0.5]. Through A ij Embedding the physical connections of the power grid into attention calculations (such as bus and feeder connection points) avoids over-response to irrelevant nodes. For example, if node A and node B are physically connected in the power grid (A... AB =1), then when processing the features of node A, the model will strengthen the attention calculation of the features of node B. The λ coefficient controls the strength of topology influence and can adapt to the differences in wiring density in different mines. When the power grid topology is complex, such as when there are multiple branches for power supply, increasing λ can strengthen topology constraints; when the feature correlation is significant, such as when high-frequency harmonics are prominent, decreasing λ can focus on data-driven approaches.

[0070] The weighting coefficient α obtained from the above attention weighting formula ijThe input features change in real time. By combining explicit topological coding with implicit feature learning, a weight allocation that better conforms to the physical laws of the power grid than traditional attention mechanisms is achieved, which is beneficial to improving the accuracy of leakage current location. Specifically, this is reflected in:

[0071] (1) Enhanced spatial correlation: In the analysis of leakage fault propagation paths, priority is given to the characteristics of adjacent lines through which the fault current flows (such as busbar → feeder → load node). For example, when an arc grounding occurs in a branch, the model automatically strengthens the harmonic characteristic weights of the branch and adjacent nodes, while suppressing the noise signals of irrelevant branches.

[0072] (2) Topology filtering: assign low weight to abnormal features of non-adjacent nodes, such as instantaneous fluctuations in remote load, to reduce the false positive rate.

[0073] (3) Hierarchical weight allocation: In deep networks, a large λ value is used for the high-pressure side (sparse topological connections) to strengthen structural constraints; a small λ value is used for the low-pressure side (dense connections) to emphasize feature association.

[0074] The leakage fault identification model in this embodiment may further include a feature fusion module and an ELM (extreme learning machine) classifier.

[0075] The feature fusion module is used to weight and fuse the feature vectors output from multiple feature learning branches through a dynamic gating mechanism, and dynamically adjusts the hidden layer activation function and regularization parameters of the ELM classifier based on the activated feature learning branch. If the first branch is activated, the hidden layer of the ELM uses the ReLU activation function and applies L1 regularization constraints; if the second branch is activated, the hidden layer uses the Sigmoid activation function and applies L2 regularization constraints; if the third branch is activated, the hidden layer uses the LeakyReLU activation function and enables the Dropout mechanism.

[0076] ELM is a feedforward neural network algorithm consisting of an input layer, hidden layers, and an output layer. ELM has strong generalization ability and high learning efficiency, and is widely used in various fields. Its network structure is as follows: Figure 3 As shown. Deep learning methods typically employ the Softmax classifier when solving classification problems. Based on probability principles, it strengthens the largest feature value while weakening the influence of other features, resulting in poor performance in identifying fault types with small feature differences. To overcome the shortcomings of Softmax, this embodiment uses the ELM classifier, which has good nonlinear fitting capabilities, strong self-learning ability on fault data, and is suitable for identifying leakage fault types with small feature differences.

[0077] The ELM classifier processes steady-state and transient fault criteria using a fault metric function, then fuses the information to select the faulty line using multiple fault features, thus improving the accuracy of fault location for leakage faults. The number of hidden layer nodes in the ELM can be determined experimentally. Preferably, nine hidden layer nodes are used, achieving high line selection accuracy with a relatively small number of nodes while maintaining detection efficiency.

[0078] During implementation, the preprocessed data to be identified is input. Figure 2 The leakage fault identification model shown obtains the fault identification result and sends a fault clearing command to the corresponding protection device. The monitoring platform can then execute the corresponding maintenance plan based on the identification result.

[0079] The method in this embodiment can identify different types of leakage faults. For minor leakage, it focuses on the current change rate characteristics; for severe leakage, it focuses on harmonic distortion characteristics; and for transient faults, it captures waveform abrupt changes. This flexible feature adaptation mechanism achieves a leap from "perceiving the fault" to "understanding the fault." Compared to traditional methods, this solution improves leakage detection accuracy by 15.7% under complex downhole conditions, achieves topology node-level positioning accuracy, and reduces response time to less than 30ms.

[0080] The embodiments of the coal mine leakage protection method based on the leakage fault identification model have been described above. Correspondingly, this disclosure also provides embodiments of the coal mine leakage protection device based on the leakage fault identification model.

[0081] Figure 4 This is a structural block diagram of a coal mine leakage current protection device 300 based on a leakage current fault identification model, provided in an embodiment of this disclosure. The device 300 can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0082] like Figure 4 As shown, the device 300 includes a data acquisition module 310, a data acquisition module 320, a detection module 330, and a control module 340. The device 300 can perform the protection method described above.

[0083] The acquisition module 310 is used to acquire electrical signals from various power grid lines, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current.

[0084] The data acquisition module 320 is used to extract the sampled values ​​of the power signal before and after the fault occurrence time N cycles when the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, and obtain the data to be identified after preprocessing.

[0085] The detection module 330 is used to input the data to be identified into a pre-trained leakage current identification model to obtain leakage current fault identification results;

[0086] The control module 340 is used to issue a fault clearing command to the protection unit of the corresponding line based on the leakage fault identification result.

[0087] Based on the same inventive concept, this disclosure also provides a coal mine leakage current protection device 400 based on a leakage current fault identification model, which is installed in the power grid line. This protection device 400 can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0088] Figure 5 This is a schematic diagram of the structure of the protection device 400 provided in an embodiment of this disclosure. Figure 5 As shown, the protection device includes a power grid interface 410, a data interface 420, a processor 430, and a memory 440.

[0089] The power grid interface 410 is connected to the processor 430 and is used to connect to the power grid lines.

[0090] Data interface 420 is connected to processor 430 and is used to communicate with management platform.

[0091] The memory 440 stores instructions that can be executed by the processor 430 to implement the various methods described above.

[0092] The protection device 400 can exist in various physical forms, including smart IoT protection terminals, mine-use explosion-proof and intrinsically safe high-voltage vacuum power distribution devices, or low-voltage smart boxes. These coal mine leakage protection devices communicate with pre-trained AI models. In some implementations, the processor 430 can be a processor with AI computing power, and the device has a built-in pre-trained AI model that can perform real-time calculations. In other implementations, the pre-trained AI model is configured on an edge computing platform, and the protection device 400 is connected to it via a transmission line.

[0093] Embodiments of this disclosure also provide a computer program that, when executed by a processor, causes the processor to implement the various protection methods described above. This computer program can be embedded in power protection equipment, power distribution devices, or other electrical equipment as part of an intelligent device, or it can be placed in an edge server or a central server and begin operation upon receiving data to be identified.

[0094] This disclosure's embodiments monitor electrical signals in the power grid, extracting sampled values ​​from multiple periods before and after a leakage fault for AI identification to determine the fault. It fuses multi-domain features through spatiotemporal joint encoding, dynamically adjusts feature weights using a conditional attention mechanism based on the power grid topology, and innovatively introduces a dynamic routing controller in the feature learning stage, switching optimized feature extraction branches in real time according to fault severity. Through the extreme learning machine's parameter dynamic adaptation mechanism, it achieves collaborative optimization of fault classification and understanding. The AI ​​model used in this disclosure employs a dynamic attention weighting layer, encoding power grid structure information into attention weights, ensuring the feature learning process conforms to the propagation laws of the power system and overcoming the shortcomings of traditional methods that neglect topological correlations. The AI ​​model used in this disclosure is designed with multiple learning branches, focusing on current change rate features for minor leakage, harmonic distortion features for severe leakage, and waveform abrupt changes for transient faults. This flexible feature adaptation mechanism achieves a leap from "perceiving faults" to "understanding faults." Compared with traditional methods, this solution improves the accuracy of leakage current detection by 15.7% under complex downhole conditions, achieves topology node-level positioning accuracy, and reduces the response time to less than 30ms.

[0095] The various embodiments in this disclosure are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and computer programs are relatively simple in description because they are substantially similar to the method embodiments; relevant details can be found in the description of the method embodiments.

[0096] The foregoing has described specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0097] The above description is merely an embodiment of this disclosure and is not intended to limit the scope of this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of the claims of this disclosure.

Claims

1. A method for leakage current protection in coal mines based on a leakage current fault identification model, characterized in that, include: Collect electrical signals from each line of the power grid, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current; If the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, the sampled values ​​of the power signal are extracted N cycles before and after the fault occurrence time, and the data to be identified is obtained after preprocessing. The data to be identified is input into a pre-trained leakage fault identification model to obtain leakage fault identification results; Based on the leakage fault identification results, a fault clearing command is issued to the protection unit of the corresponding line; The leakage fault identification model includes: The feature extraction module is used to extract multi-domain features from the data to be identified. The multi-domain features include time-domain waveform, frequency-domain harmonic energy distribution, and statistical domain abrupt change parameters. A spatiotemporal joint coding module is used to fuse and encode the multi-domain features to obtain coded features; The system includes a dynamic routing controller and multiple feature learning branches. The dynamic routing controller activates corresponding feature learning branches based on real-time evaluation metrics. The feature learning branches learn the encoded features. The multiple feature learning branches are based on a stacked autoencoder (SAE) model and include: a first branch that uses a convolutional neural network to extract current change rate features and is activated when the current change rate exceeds a first threshold; a second branch that uses a frequency domain attention network to extract harmonic distortion features and is activated when the total harmonic distortion rate exceeds a second threshold; and a third branch that uses a long short-term memory network to extract waveform abrupt change features and is activated when the density of abrupt change points exceeds a third threshold. The system includes a feature fusion module and an Extreme Learning Machine (ELM) classifier. The feature fusion module is used to weight and fuse the feature vectors output by the multiple feature learning branches through a dynamic gating mechanism, and then input them into the Extreme Learning Machine (ELM) classifier. The module also dynamically adjusts the hidden layer activation function and regularization parameters of the Extreme Learning Machine (ELM) classifier based on the feature learning branches activated by the dynamic routing controller.

2. The method according to claim 1, characterized in that, The spatiotemporal joint coding module is also used to generate conditional attention weights based on the adjacency matrix of the power grid topology and to dynamically weight the coding features.

3. The method according to claim 2, characterized in that, The conditional attention weight calculation satisfies: Where, α ij W is the attention weight of node i to node j. q It is a query matrix, W k It is a key matrix, A ij These are the elements of the power grid topology adjacency matrix, where λ is the topology influence coefficient, σ is the nonlinear activation function, and d... k The dimension of the key vector.

4. A coal mine leakage current protection device based on a leakage current fault identification model, used to implement the method described in any one of claims 1-3, characterized in that, include: The acquisition module is used to acquire electrical signals from various power grid lines, including three-phase voltage, three-phase current, zero-sequence voltage, and zero-sequence current. The data acquisition module is used to extract the sampled values ​​of the power signal N cycles before and after the fault occurrence when the zero-sequence voltage amplitude of any line is detected to be greater than a preset value, and obtain the data to be identified after preprocessing. The detection module is used to input the data to be identified into a pre-trained leakage fault identification model to obtain leakage fault identification results; The control module is used to issue a fault clearing command to the protection unit of the corresponding line based on the leakage fault identification result.

5. A coal mine leakage current protection device based on a leakage current fault identification model, installed in the power grid line, characterized in that, The protection device includes a power grid interface, a data interface, a processor, and a memory, wherein, The power grid interface is connected to the processor and is used to connect to the power grid lines; The data interface is connected to the processor and is used to communicate with the management platform; The memory stores instructions that can be executed by the processor to implement the protection method according to any one of claims 1-3.

6. The coal mine leakage protection device according to claim 5, wherein the coal mine leakage protection device is an intelligent IoT protection terminal, a mine explosion-proof and intrinsically safe high-voltage vacuum power distribution device or a low-voltage intelligent box, and the coal mine leakage protection device is communicatively connected to a pre-trained AI model.

7. A computer program product, when executed by a processor, causes the processor to implement the protection method as described in any one of claims 1 to 3.

Images