Rapid arc fault detection device and method for photovoltaic power station

Abstract

The invention relates to a rapid arc fault detection device and method for a photovoltaic power station. The device comprises a power station data acquisition module for acquiring circuit topology data and electric signals; the arc fault identification module is used for identifying an arc fault; the fault area positioning module is used for comparing and analyzing the electric signals according to circuit topology data and positioning an arc fault area; the abnormal discharge mode extraction module is used for extracting an abnormal discharge mode in allusion to the arc fault area and in combination with the electric signal; the fault evolution path determination module is used for determining a potential fault evolution path based on the abnormal discharge mode; and the arc fault alarm module is used for integrating the abnormal discharge mode and the potential fault evolution path and generating an arc fault alarm. By adopting the device, rapid identification, positioning and early warning of arc faults can be realized through cooperation of multiple modules, and the safety and reliability of a photovoltaic power station are improved.

Description

Technical Field

[0001] This invention belongs to the field of fault detection and intelligent diagnosis of new energy power generation equipment, and in particular relates to a rapid arc fault detection device and method for photovoltaic power plants. Background Technology

[0002] With the development of fault detection and intelligent diagnosis technologies for new energy power generation equipment, the safe and stable operation of photovoltaic power plants has received increasing attention. Arc faults, a common electrical hazard in photovoltaic power plants, are characterized by their high degree of concealment and rapid development; therefore, reliable detection and early warning are crucial for ensuring the safety of the power plant. Currently, this field mainly relies on single feature analysis of electrical signals in the time or frequency domain, using methods such as threshold comparison for fault judgment. Traditional technologies typically detect sudden changes in current or voltage, triggering alarms based on fixed thresholds. This method can identify obvious arc phenomena to some extent. However, current detection methods lack comprehensive analysis of circuit topology, making it difficult to quickly and accurately locate fault points. Traditional methods extract fault features in a relatively simplistic way, failing to analyze the evolution trend of the fault, resulting in long detection delays, insufficient location accuracy, and a lack of early warning capabilities, making it difficult to meet the actual needs of photovoltaic power plants for early fault detection and rapid response. Summary of the Invention

[0003] Therefore, it is necessary to provide a rapid arc fault detection device and method for photovoltaic power plants that can solve the above problems.

[0004] In a first aspect, this application provides a rapid arc fault detection device for a photovoltaic power station, comprising:

[0005] The power plant data acquisition module is used to acquire circuit topology data and electrical signals of the photovoltaic power plant;

[0006] The arc fault identification module is used to identify arc faults based on electrical signals and according to a preset arc fault detection system.

[0007] The fault area location module is used to locate the arc fault area by comparing and analyzing electrical signals based on circuit topology data in response to the detection of an arc fault.

[0008] The abnormal discharge mode extraction module is used to extract abnormal discharge modes in the arc fault area by combining electrical signals.

[0009] The fault evolution path determination module is used to determine potential fault evolution paths based on abnormal discharge modes and using a preset machine learning model.

[0010] The arc fault alarm module is used to integrate abnormal discharge modes and potential fault evolution paths to generate arc fault alarms.

[0011] In one embodiment, the arc fault identification module is further used for:

[0012] Based on electrical signals, time-domain features, frequency-domain features, and time-frequency-domain features are extracted.

[0013] By integrating time-domain features, frequency-domain features, and time-frequency-domain features, an electrical signal feature matrix is ​​constructed.

[0014] Obtain the working environment parameters and operating parameters of the photovoltaic power station;

[0015] Based on the working environment parameters and operating parameters, set the time-domain characteristic benchmark, frequency-domain characteristic benchmark, and time-frequency-domain characteristic benchmark;

[0016] The feature weights of the time-domain feature benchmark, frequency-domain feature benchmark, and time-frequency-domain feature benchmark are determined using the analytic hierarchy process.

[0017] Based on time-domain feature benchmarks, frequency-domain feature benchmarks, and time-frequency-domain feature benchmarks, and combined with feature weights, an electrical signal benchmark matrix is ​​constructed.

[0018] Calculate the feature similarity between the feature matrix of the electrical signal and the reference matrix of the electrical signal;

[0019] Based on feature similarity, an arc fault is identified using a preset similarity threshold.

[0020] In one embodiment, the fault area location module is further configured to:

[0021] Extract the topological relationships and location associations of each circuit branch from the circuit topology data;

[0022] In response to the detection of an arc fault, the electrical signals are divided according to the topological correlation to obtain the branch electrical signals;

[0023] Calculate the characteristic deviation between the characteristic matrix of each branch signal and the reference matrix of the signal;

[0024] Based on the preset deviation threshold, the characteristic deviation value of each branch's electrical signal is compared with the preset deviation threshold to determine the candidate branches with fault associations.

[0025] For candidate branches, a temporal correlation analysis is performed on the feature deviation values ​​in combination with topological relationships to determine the target associated branches;

[0026] Based on the target associated branch and combined with location association information, the arc fault area is located.

[0027] In one embodiment, the abnormal discharge mode extraction module is further configured to:

[0028] Based on the electrical signal corresponding to the arc fault area, the current extreme value difference, current standard deviation and current change rate within a preset time period are extracted from the time domain features; the energy mutation value and higher harmonic amplitude features of a preset frequency band are extracted from the frequency domain features; and wavelet transform features and intrinsic mode functions are extracted from the time and frequency domain features.

[0029] The abnormal discharge mode is generated by integrating the current extreme value difference, current standard deviation, current change rate, energy mutation value, higher harmonic amplitude characteristics, wavelet transform characteristics, and intrinsic mode function as characteristic parameters.

[0030] In one embodiment, the fault evolution path determination module presets a gradient boosting tree model as the machine learning model. The gradient boosting tree model presets the fault type corresponding to each feature parameter in the abnormal discharge mode, the mapping matching logic between feature parameters and fault types, and the path generation rules.

[0031] Based on the abnormal discharge mode, a pre-set machine learning model is used to determine potential fault evolution paths, including:

[0032] Based on the characteristic parameters of the abnormal discharge mode, the entropy weight method is used to determine the contribution of each characteristic parameter to the arc fault.

[0033] Based on contribution, the top N feature parameters with the highest contribution are selected as core feature parameters.

[0034] Based on the core feature parameters, the corresponding fault type is matched through the mapping and matching logic of the gradient boosting tree model.

[0035] By combining the fault type with the topology association, multiple fault evolution paths are generated according to the path generation rules, forming a candidate path set;

[0036] The candidate path set is probabilistically ranked using a Bayesian network to obtain the probabilistic ranking results.

[0037] Based on the probability ranking results, potential failure evolution paths are determined from the candidate path set.

[0038] In one embodiment, the fault evolution path determination module is further configured to select the top N feature parameters based on their contribution as core feature parameters using the following formula:

[0039]

[0040] Where S is the set of core feature parameters, Let be the i-th characteristic parameter in the abnormal discharge mode, where i is the index of the characteristic parameter, and m is the total number of characteristic parameters in the abnormal discharge mode. Let be the contribution of the i-th characteristic parameter to the arc fault. for The ranking numbers are sorted in descending order, where N is the preset number of core feature parameters to be selected, and N is a positive integer. , This represents the contribution value of the kth feature parameter after sorting the contribution values ​​in descending order. The calibration coefficient for the effectiveness of contribution. The standard deviation of the contribution of all feature parameters The mean of the contributions of all feature parameters. It is a local minimum constant.

[0041] In one embodiment, the arc fault alarm module is further used for:

[0042] Based on the probability ranking results of the core characteristic parameters of abnormal discharge modes and potential fault evolution paths, key path nodes are extracted.

[0043] Based on core feature parameters and key path nodes, fault correlation values ​​are calculated according to preset node weight rules.

[0044] Based on the operating parameters of the photovoltaic power station, determine the threshold range of multiple alarm levels corresponding to the fault correlation values;

[0045] The alarm level is determined based on the fault correlation value and the level threshold range;

[0046] By integrating core feature parameters, critical path nodes, and alarm levels, an arc fault alarm is generated.

[0047] Secondly, this application also provides a method for rapid arc fault detection in photovoltaic power plants, comprising:

[0048] Acquire circuit topology data and electrical signals of photovoltaic power plants;

[0049] Based on electrical signals, and according to a pre-set arc fault detection system, arc faults are identified.

[0050] In response to the detection of an arc fault, the electrical signals are compared and analyzed based on the circuit topology data to locate the arc fault area;

[0051] For areas with electric arc faults, abnormal discharge patterns are extracted by combining electrical signals;

[0052] Based on the abnormal discharge mode, a pre-set machine learning model is used to determine the potential fault evolution path;

[0053] Integrate abnormal discharge modes and potential fault evolution paths to generate arc fault alarms.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to realize the function of the aforementioned rapid arc fault detection device for a photovoltaic power station.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the functions of the aforementioned rapid arc fault detection device for a photovoltaic power station.

[0056] The aforementioned rapid arc fault detection device and method for photovoltaic power plants acquires circuit topology data and electrical signals through a power plant data acquisition module, identifies faults based on electrical signals to solve the problem of long detection delays, locates fault areas by comparing and analyzing topology data to solve the problem of inaccurate location, extracts fault modes by an abnormal discharge mode extraction module, predicts fault evolution paths using machine learning models, and integrates and generates alarms by an arc fault alarm module to solve the problem of lack of early warning capabilities. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a structural diagram of a rapid arc fault detection device for a photovoltaic power station according to the present invention;

[0059] Figure 2 This is a flowchart of a rapid arc fault detection method for a photovoltaic power station according to the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0061] In one embodiment, such as Figure 1 As shown, a rapid arc fault detection device for a photovoltaic power station is provided. This embodiment uses the deployment of the device on a terminal as an example for illustration. It can be understood that the device can also be deployed on a server, or in a system including a terminal and a server, and can be implemented through the interaction between the terminal and the server.

[0062] In the implementation environment, the hardware includes electrical signal sensors, computing terminals, and servers. Terminal devices, such as embedded monitors, are responsible for real-time acquisition of current and voltage signals and topology data from the power plant. Servers provide high-performance computing resources for machine learning model training and big data storage. Application scenarios include: when a potential arc fault occurs during the operation of a photovoltaic power plant, the hardware interacts as follows: sensors continuously monitor electrical signals and transmit them to the terminal; the terminal integrates the circuit topology data and electrical signals and sends them to the server; the server, based on the electrical signals and circuit topology data, processes the data and transmits the generated alarm to the terminal, enabling remote diagnosis and early warning.

[0063] The data flow between the modules of this device is as follows: the power plant data acquisition module acquires circuit topology data and electrical signals, and transmits them to the arc fault identification module for multi-feature analysis to identify arc faults; after the arc fault is identified, the fault area location module is triggered to acquire topology data and electrical signals, and then locate the arc fault area; the abnormal discharge mode extraction module acquires the arc fault area, circuit topology data and electrical signals, and extracts the abnormal discharge mode; after the fault evolution path determination module acquires the abnormal discharge mode, it uses a gradient boosting tree model to predict the potential fault evolution path; the arc fault alarm module integrates the abnormal discharge mode and the potential fault evolution path to generate an arc fault alarm.

[0064] In this embodiment, the device includes:

[0065] The power station data acquisition module 101 is used to acquire the circuit topology data and electrical signals of the photovoltaic power station.

[0066] Among them, circuit topology data is the basic data describing the connection relationship, layout structure and topology association information of electrical components (such as inverters, combiner boxes, etc.) in the power station, and electrical signals cover real-time electrical parameters such as current and voltage; data can be acquired in real time or periodically through data acquisition devices (such as sensor networks, monitoring terminals or communication interfaces) to achieve comprehensive coverage of the information required for fault detection.

[0067] Arc fault identification module 102 is used to identify arc faults based on electrical signals and according to a preset arc fault detection system.

[0068] Among them, the preset arc fault detection system (which includes comprehensive fault judgment logic based on feature analysis, threshold comparison or pattern recognition, used to identify arc faults) can quickly detect faults by extracting multi-dimensional features of electrical signals (such as time domain, frequency domain or time-frequency domain features) and performing similarity or deviation analysis with reference parameters (such as feature references dynamically set based on environment or operating conditions).

[0069] The fault area location module 103 is used to locate the arc fault area by comparing and analyzing electrical signals based on circuit topology data in response to the detection of an arc fault.

[0070] In response to the arc fault identification module detecting an arc fault, the electrical signals can be compared and analyzed based on circuit topology data (such as by dividing branch signals, calculating characteristic deviation values, or performing time-series correlation analysis) to locate the arc fault area. The electrical signals can be divided into branch units by extracting topological correlation relationships, and candidate branches can be screened based on a preset deviation threshold and a reference matrix. The target associated branch can be determined by combining time-series analysis, and the area can be located based on the location information of the target associated branch.

[0071] The abnormal discharge mode extraction module 104 is used to extract abnormal discharge modes in the arc fault area by combining electrical signals.

[0072] Among them, the abnormal discharge mode (a set of multi-dimensional feature parameters derived from the electrical signal to characterize the typical discharge behavior of the fault) can be generated by extracting feature parameters (such as statistical features, spectral features or transform features) from the electrical signal through signal processing techniques (such as time domain analysis, frequency domain analysis and time-frequency domain analysis) and integrating these parameters. The feature extraction methods such as adaptive feature selection, machine learning feature extraction or dynamic threshold adjustment can be used to achieve the representativeness and adaptability of the abnormal discharge mode.

[0073] The fault evolution path determination module 105 is used to determine potential fault evolution paths based on abnormal discharge modes and using a preset machine learning model.

[0074] Among them, a pre-set machine learning model (such as gradient boosting tree, neural network or other data-driven algorithm, and pre-set fault type mapping logic and path generation rules corresponding to feature parameters) is used to determine the potential fault evolution path (the possible development sequence or influence chain of the fault); by analyzing the feature parameters of the abnormal discharge mode (such as using the entropy weight method to evaluate the contribution of each parameter to the arc fault), and screening the core feature parameters, the model's matching logic is used to associate the fault type, and combined with the circuit topology relationship to generate a candidate path set. Based on the candidate path set, the high probability path, i.e. the potential fault evolution path, is determined by probability ranking (such as Bayesian network).

[0075] Arc fault alarm module 106 is used to integrate abnormal discharge modes and potential fault evolution paths to generate arc fault alarms.

[0076] Among them, the arc fault alarm (which includes structured output information on fault characteristics, evolution trends and warning levels) can be implemented by combining abnormal discharge modes with potential fault evolution paths through data fusion and analysis technologies (such as correlation calculation based on feature parameters and path nodes, fault level assessment or information integration logic) to output readable alarms and achieve accurate real-time early warning of arc faults.

[0077] The aforementioned rapid arc fault detection device for photovoltaic power plants acquires circuit topology data and electrical signals in real time through a power plant data acquisition module, providing a comprehensive data foundation for fault analysis. The arc fault identification module uses a preset detection system based on electrical signals to quickly identify arc faults, solving the problem of long detection delays and shortening response time through dynamic feature comparison. The fault area location module responds to the identification results, using topology data to divide and compare electrical signals into branches, accurately locating the fault area through feature deviation value calculation and temporal correlation, overcoming the shortcomings of traditional location accuracy. The abnormal discharge mode extraction module extracts multi-dimensional discharge features from the fault area, and the fault evolution path determination module uses a machine learning model to predict potential paths, achieving early warning capabilities. The arc fault alarm module integrates information to generate alarms, comprehensively improving the safety and reliability of the photovoltaic power plant.

[0078] In one embodiment, the arc fault identification module 102 is further configured to:

[0079] Based on electrical signals, time-domain features, frequency-domain features, and time-frequency-domain features are extracted.

[0080] By integrating time-domain features, frequency-domain features, and time-frequency-domain features, an electrical signal feature matrix is ​​constructed.

[0081] Obtain the working environment parameters and operating parameters of the photovoltaic power station;

[0082] Based on the working environment parameters and operating parameters, set the time-domain characteristic benchmark, frequency-domain characteristic benchmark, and time-frequency-domain characteristic benchmark;

[0083] The feature weights of the time-domain feature benchmark, frequency-domain feature benchmark, and time-frequency-domain feature benchmark are determined using the analytic hierarchy process.

[0084] Based on time-domain feature benchmarks, frequency-domain feature benchmarks, and time-frequency-domain feature benchmarks, and combined with feature weights, an electrical signal benchmark matrix is ​​constructed.

[0085] Calculate the feature similarity between the feature matrix of the electrical signal and the reference matrix of the electrical signal;

[0086] Based on feature similarity, an arc fault is identified using a preset similarity threshold.

[0087] For example, for electrical signals, signal processing techniques can be used to extract features such as amplitude changes and peak values ​​through time-domain analysis, extract features such as frequency distribution and harmonic content through Fourier transform through frequency-domain analysis, and extract joint features of the signal at different times and frequencies through wavelet transform and other methods through time-frequency domain analysis, thus obtaining time-domain features, frequency-domain features, and time-frequency-domain features. These three types of features are then integrated to construct an electrical signal feature matrix according to a preset matrix construction rule with feature type as rows and feature value as columns. Working environment parameters such as temperature, humidity, and light intensity, as well as operating parameters such as rated power, output voltage range, and operating time, can be obtained through the sensors or monitoring system of the photovoltaic power station. Combined with historical operating data and safety operating standards under different operating conditions of the photovoltaic power station, time-domain feature benchmarks, frequency-domain feature benchmarks, and time-frequency-domain feature benchmarks adapted to the current operating conditions can be set according to the working environment parameters and operating parameters. A hierarchical structure of three types of feature benchmarks is established using the analytic hierarchy process (AHP). Pairwise comparisons are used to determine the importance of each feature benchmark for arc fault identification, and a judgment matrix is ​​constructed. After consistency verification, the feature weights of each feature benchmark are calculated. Following the same dimensions and arrangement rules as the electrical signal feature matrix, the three types of feature benchmarks are used as matrix elements. Weighted calculations are performed on the corresponding features and their corresponding weights, and the matrix elements are then filled into their corresponding positions. The electrical signal benchmark matrix is ​​constructed by combining these feature weights. A cosine similarity algorithm or a Euclidean distance transformation similarity calculation method can be used to calculate the feature similarity between the corresponding elements in the electrical signal feature matrix and the electrical signal benchmark matrix. This feature similarity is then compared with a preset similarity threshold (determined based on statistical analysis of numerous arc fault simulation experiments and normal operation data). If the feature similarity is lower than the preset similarity threshold, an arc fault is determined to exist, thus achieving arc fault identification.

[0088] In one embodiment, the fault area location module 103 is further configured to:

[0089] Extract the topological relationships and location associations of each circuit branch from the circuit topology data;

[0090] In response to the detection of an arc fault, the electrical signals are divided according to the topological correlation to obtain the branch electrical signals;

[0091] Calculate the characteristic deviation between the characteristic matrix of each branch signal and the reference matrix of the signal;

[0092] Based on the preset deviation threshold, the characteristic deviation value of each branch's electrical signal is compared with the preset deviation threshold to determine the candidate branches with fault associations.

[0093] For candidate branches, a temporal correlation analysis is performed on the feature deviation values ​​in combination with topological relationships to determine the target associated branches;

[0094] Based on the target associated branch and combined with location association information, the arc fault area is located.

[0095] Specifically, by analyzing the electrical component connection identifiers, port mapping relationships, and branch hierarchy structure in the circuit topology data, the topological relationships between each circuit branch (such as upstream and downstream branch connections, parallel branch associations, etc.) can be extracted. From the physical layout description accompanying the circuit topology data and the installation location log corresponding to the branch numbers, the location association information of each branch (such as the equipment installation area, coordinate range, component number, etc.) can be extracted. After the arc fault identification module detects an arc fault, in response to the fault signal, it determines the signal transmission of each branch based on the extracted topological relationships. Path and impedance characteristics can be determined using a signal separation algorithm based on branch affiliation (such as a signal decomposition model based on nodal voltage method) to divide the acquired overall electrical signal and assign it to the corresponding circuit branches, resulting in independent branch electrical signals for each branch. Following the same rules used in the arc fault identification module to construct the electrical signal feature matrix, a branch electrical signal feature matrix is ​​constructed for each branch signal. Using the Euclidean distance formula or relative deviation calculation method, the deviation of corresponding elements in each branch electrical signal feature matrix is ​​calculated compared to the already constructed electrical signal reference matrix to obtain the characteristic deviation of each branch. This value quantifies the deviation of a branch signal from a normal reference signal. The characteristic deviation value of each branch is compared with a preset deviation threshold (set based on a 99% confidence interval through statistical analysis of a large amount of normal operation data and simulated fault data from photovoltaic power plants to ensure the accuracy of fault judgment). Branches with characteristic deviation values ​​greater than the preset deviation threshold are identified as candidate branches with fault associations. For the selected candidate branches, the connection logic of the branches in the topological relationship (such as the signal propagation timing of series branches and the signal synchronization of parallel branches) is used to extract the continuous time... The feature deviation values ​​within the segment are used to construct a time-series deviation sequence. By calculating indicators such as Pearson correlation coefficient and trend slope matching degree between sequences, time-series correlation analysis is performed to screen out branches that match the occurrence and development time of the fault and conform to the topological transmission logic as target associated branches. The branch number and association identifier of the determined target associated branches are matched with the extracted location association information. By querying the branch-location mapping table, the physical installation area and equipment component location range corresponding to the target associated branches are determined, and the area where the arc fault is located is accurately located, realizing rapid and accurate location of the fault area.

[0096] In one embodiment, the abnormal discharge mode extraction module 104 is further configured to:

[0097] Based on the electrical signal corresponding to the arc fault area, the current extreme value difference, current standard deviation and current change rate within a preset time period are extracted from the time domain features; the energy mutation value and higher harmonic amplitude features of a preset frequency band are extracted from the frequency domain features; and wavelet transform features and intrinsic mode functions are extracted from the time and frequency domain features.

[0098] The abnormal discharge mode is generated by integrating the current extreme value difference, current standard deviation, current change rate, energy mutation value, higher harmonic amplitude characteristics, wavelet transform characteristics, and intrinsic mode function as characteristic parameters.

[0099] For example, the preset time period is set to 10ms-1s based on the typical duration of arc faults in photovoltaic power plants and the signal sampling frequency, and the preset frequency band is set to 2kHz-10kHz based on the high-frequency characteristics of arc fault discharge signals. For time-domain characteristics, the difference between the maximum and minimum current values ​​is calculated by iterating through the current sampling data of the electrical signal within the preset time period to obtain the current extreme value difference. The standard deviation of the current is calculated using the statistical standard deviation formula for the current data within this time period. The rate of change of the current is obtained by dividing the difference between two adjacent current sampling points by the sampling interval. For frequency-domain characteristics, a Fourier transform is performed on the electrical signal to obtain the frequency distribution. The signal energy within the preset frequency band is extracted and its difference is calculated with the reference energy of the same frequency band under normal operating conditions. The energy mutation value can be used to screen out the third and higher harmonic components and measure their amplitudes to obtain the higher harmonic amplitude characteristics. For the time-frequency domain characteristics, the db4 wavelet basis can be used to perform three-level wavelet decomposition on the electrical signal, and the detail coefficients and approximation coefficients after decomposition can be extracted as wavelet transform feature quantities. The time-frequency domain signal can be decomposed into multiple intrinsic mode functions by the empirical mode decomposition (EMD) method, and the first three intrinsic mode functions with the highest energy proportion can be selected as effective features. According to the fixed order of current extreme value difference - current standard deviation - current change rate - energy mutation value - higher harmonic amplitude characteristics - wavelet transform feature quantity - intrinsic mode function, the above seven types of feature parameters are integrated to construct a seven-dimensional feature vector. This feature vector is the abnormal discharge mode that can comprehensively characterize the discharge characteristics of arc faults.

[0100] In one embodiment, the fault evolution path determination module 105 presets a machine learning model as a gradient boosting tree model. The gradient boosting tree model presets the fault type corresponding to each feature parameter in the abnormal discharge mode, the mapping matching logic between feature parameters and fault types, and the path generation rules.

[0101] Based on the abnormal discharge mode, a pre-set machine learning model is used to determine potential fault evolution paths, including:

[0102] Based on the characteristic parameters of the abnormal discharge mode, the entropy weight method is used to determine the contribution of each characteristic parameter to the arc fault.

[0103] Based on contribution, the top N feature parameters with the highest contribution are selected as core feature parameters.

[0104] Based on the core feature parameters, the corresponding fault type is matched through the mapping and matching logic of the gradient boosting tree model.

[0105] By combining the fault type with the topology association, multiple fault evolution paths are generated according to the path generation rules, forming a candidate path set;

[0106] The candidate path set is probabilistically ranked using a Bayesian network to obtain the probabilistic ranking results.

[0107] Based on the probability ranking results, potential failure evolution paths are determined from the candidate path set.

[0108] Specifically, the pre-set machine learning model can be a gradient boosting tree model, which is pre-trained using a large amount of historical data on arc faults in photovoltaic power plants. It includes fault types corresponding to seven types of feature parameters in abnormal discharge modes (such as abnormal current extreme values ​​corresponding to line insulation damage faults, abnormal high-order harmonic amplitudes corresponding to loose joint faults, etc.), mapping and matching logic between feature parameters and fault types (setting threshold ranges for each feature parameter, matching the corresponding fault type when the core feature parameter value falls into a certain threshold range), and path generation rules (such as fault propagation along upstream and downstream branches of the topology, and priority of fault propagation between parallel branches, etc.). Based on seven characteristic parameters of abnormal discharge modes, the contribution of each characteristic parameter can be calculated using the entropy weight method. By statistically analyzing the numerical distribution of each characteristic parameter across multiple fault samples, its information entropy is calculated. According to the conversion formula between information entropy and weight, the information entropy is transformed into the contribution of each characteristic parameter to the arc fault. This contribution quantifies the influence of each characteristic parameter on fault identification and evolution. Based on this contribution, all characteristic parameters are sorted in descending order. A preset number N of core characteristic parameters is selected (N is a positive integer of 1 ≤ N ≤ 7, preferably 3-5 to balance computational efficiency and characteristic representativeness). The top N characteristic parameters that meet the core characteristic parameter selection criteria are then selected. The data is used as the core feature parameter. The selected core feature parameters are input into the gradient boosting tree model. Through the model's built-in mapping matching logic, the values ​​of the core feature parameters are compared with the preset threshold range and feature combination relationship to match the corresponding arc fault type. Combining the matched fault type with the topological association relationship extracted by the fault area localization module, and based on path generation rules (such as faults being preferentially propagated along directly connected branches of the topological association, and fault propagation priority of high-load branches being higher than that of low-load branches), multiple possible fault evolution paths are generated. For example, a line insulation failure fault may spread from the target associated branch to its upstream combiner box, and then spread to the inverter. This can directly affect parallel associated branches, forming a candidate path set containing multiple different transmission paths. The candidate path set is input into a pre-trained Bayesian network. This Bayesian network uses the historical failure probability, topological association strength, and equipment operating years of each branch in the candidate path as input variables. The network's built-in probabilistic inference algorithm calculates the failure probability of each candidate path and obtains the probability ranking result. Based on this probability ranking result, the top 2-3 candidate paths (which can be adjusted according to the safety requirements of the photovoltaic power station) are selected as potential failure evolution paths that can reflect the failure development trend, thereby achieving effective prediction of the subsequent evolution direction of the failure.

[0109] In one embodiment, the fault evolution path determination module 105 is further configured to select the top N feature parameters based on their contribution as core feature parameters using the following formula:

[0110]

[0111] Where S is the set of core feature parameters, Let be the i-th characteristic parameter in the abnormal discharge mode, where i is the index of the characteristic parameter, and m is the total number of characteristic parameters in the abnormal discharge mode. Let be the contribution of the i-th characteristic parameter to the arc fault. for The ranking numbers are sorted in descending order, where N is the preset number of core feature parameters to be selected, and N is a positive integer. , This represents the contribution value of the kth feature parameter after sorting the contribution values ​​in descending order. The calibration coefficient for the effectiveness of contribution. The standard deviation of the contribution of all feature parameters The mean of the contributions of all feature parameters. It is a local minimum constant.

[0112] For example, the contribution of each characteristic parameter in the abnormal discharge mode to the arc fault. The contribution of all m feature parameters is calculated using the quicksort algorithm. Sort in descending order, and record each feature parameter. Corresponding ranking number The preset number of core feature parameters to be screened, N, is set according to the balance between the accuracy requirements and computational efficiency of photovoltaic power station fault detection (N is a positive integer of 1≤N≤m, preferably 3-5); the contribution effectiveness calibration coefficient λ is set to 0.8-1.2 (preferably 0.9, to ensure that the screened core parameters have effective contribution) after verification by multiple arc fault simulation experiments and actual operating data; the minimum constant ε is set to 1×10^-6 (to avoid a denominator of 0); and the standard deviation of the contribution of all feature parameters is... with the mean Statistical calculations were performed to determine the contribution of the first N characteristic parameters after sorting them in descending order. Summing is performed (k from 1 to N), and the sum is multiplied by the ratio of λ to N, then multiplied by... and( The ratio of ) is used to obtain the contribution threshold for screening; for each feature parameter Perform a dual condition check, that is, check its corresponding index identifier i. Is it less than or equal to N? Whether it is greater than or equal to the contribution threshold; all feature parameters that simultaneously meet both conditions. The core feature parameters are integrated to form a set S. The selected core feature parameters have both a high contribution ranking and meet the contribution effectiveness requirements, which can provide accurate and efficient feature support for subsequent fault type matching and evolution path prediction.

[0113] In one embodiment, the arc fault alarm module 106 is further configured to:

[0114] Based on the probability ranking results of the core characteristic parameters of abnormal discharge modes and potential fault evolution paths, key path nodes are extracted.

[0115] Based on core feature parameters and key path nodes, fault correlation values ​​are calculated according to preset node weight rules.

[0116] Based on the operating parameters of the photovoltaic power station, determine the threshold range of multiple alarm levels corresponding to the fault correlation values;

[0117] The alarm level is determined based on the fault correlation value and the level threshold range;

[0118] By integrating core feature parameters, critical path nodes, and alarm levels, an arc fault alarm is generated.

[0119] Specifically, based on the probability ranking results of the core characteristic parameters and potential fault evolution paths already screened in the abnormal discharge mode, nodes that play a decisive role in fault propagation and diffusion in the evolution paths with high probability rankings (such as the branch node where the fault source is located, the key merging node in the topology association, and the connection node of equipment with a high probability of being affected) are selected as critical path nodes. The corresponding influence relationship between the core characteristic parameters and the nodes is simultaneously associated (e.g., the fault risk of a specific critical node corresponding to an abnormal core characteristic parameter). Based on the contribution of each core characteristic parameter, combined with preset node weight rules (set through the analysis of the degree of fault impact of photovoltaic power plants, such as setting the weight of the fault source node to 0.8, the weight of the main propagation node to 0.5, and the weight of the secondary association node to 0.3), a weighted summation algorithm is used to calculate the fault association value, which is obtained by summing the products of the contribution of the core characteristic parameter and the corresponding critical path node weight, i.e.:

[0120]

[0121] This system quantifies the severity and impact of faults. Based on the photovoltaic power plant's operating parameters (such as rated power, operating years, and current load rate), combined with historical fault handling data and safe operation standards, it determines the threshold ranges for multiple alarm levels (such as early warning, general fault, and severe fault). For example, when the power plant's rated power is 50MW and its operating years are ≤3 years, a fault correlation value of 0-0.3 is set as the early warning threshold range, 0.3-0.7 as the general fault threshold range, and above 0.7 as the severe fault threshold range, ensuring that the threshold ranges are adapted to the actual operating conditions of the power plant. The calculated fault correlation value is compared with the determined threshold ranges to determine the alarm level. The system integrates core characteristic parameters (including abnormal parameter names and values), critical path nodes (including node numbers, corresponding equipment and location information), and the determined alarm level to supplement the possible impact range of the fault and preliminary handling suggestions, generating an arc fault alarm. This alarm can be output through power plant monitoring terminals, mobile management platforms, and other channels, enabling staff to quickly grasp key fault information and respond promptly.

[0122] The aforementioned rapid arc fault detection device for a photovoltaic power station acquires circuit topology data and electrical signals through a power station data acquisition module. An arc fault identification module extracts time-domain, frequency-domain, and time-frequency-domain features of the electrical signals to construct a feature matrix. It then sets feature benchmarks based on the photovoltaic power station's operating environment and parameters, determines weights using the analytic hierarchy process (AHP), and constructs an electrical signal benchmark matrix. Arc faults are identified through feature similarity comparison. In response to the fault identification results, a fault area location module divides branch electrical signals based on topological relationships and locates the fault area through feature deviation value comparison and time-series correlation analysis. An abnormal discharge mode extraction module extracts feature parameters such as current extreme value difference and higher harmonic amplitude from multi-domain features and integrates them to generate abnormal discharge modes. A fault evolution path determination module uses entropy weighting to screen core feature parameters, matches fault types using a gradient boosting tree model, generates a candidate path set based on topological relationships, and determines potential fault evolution paths through Bayesian network probability ranking. An arc fault alarm module extracts key path nodes, calculates fault correlation values, matches multiple alarm levels, and integrates core feature parameters, key nodes, and alarm levels to generate an alarm. By integrating multi-module collaboration, multi-domain feature analysis, topology application, and machine learning models, the system addresses the problems of long detection delays, inaccurate positioning, and lack of early warning in traditional methods. This enables rapid identification, accurate positioning, and effective early warning of arc faults, thereby improving the safety and reliability of photovoltaic power plants.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides a method for detecting rapid arc faults in a photovoltaic power station, which implements the aforementioned rapid arc fault detection device for a photovoltaic power station. The solution provided by this method is similar to the solution described in the aforementioned device. Therefore, the specific limitations in one or more embodiments of the rapid arc fault detection method for a photovoltaic power station provided below can be found in the above-described limitations of the rapid arc fault detection device for a photovoltaic power station, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 2 As shown, a rapid arc fault detection method for photovoltaic power plants is provided, including:

[0126] S01, acquire the circuit topology data and electrical signals of the photovoltaic power station;

[0127] S02, based on electrical signals, identifies arc faults according to a preset arc fault detection system;

[0128] S03, in response to the detection of an arc fault, compares and analyzes the electrical signals based on the circuit topology data to locate the arc fault area;

[0129] S04, for the arc fault area, extract the abnormal discharge mode by combining the electrical signal;

[0130] S05, based on the abnormal discharge mode, uses a preset machine learning model to determine the potential fault evolution path;

[0131] S06 integrates abnormal discharge modes and potential fault evolution paths to generate arc fault alarms.

[0132] In one embodiment, based on an electrical signal and according to a preset arc fault detection system, an arc fault is identified, including:

[0133] S11, based on electrical signals, extracts time-domain features, frequency-domain features, and time-frequency-domain features;

[0134] S12 integrates time-domain features, frequency-domain features, and time-frequency-domain features to construct an electrical signal feature matrix;

[0135] S13, obtain the working environment parameters and operating parameters of the photovoltaic power station;

[0136] S14, Based on the working environment parameters and operating parameters, set the time domain characteristic reference, frequency domain characteristic reference, and time-frequency domain characteristic reference;

[0137] S15, the feature weights of the time-domain feature benchmark, frequency-domain feature benchmark, and time-frequency-domain feature benchmark are determined by the analytic hierarchy process;

[0138] S16, based on time-domain feature benchmarks, frequency-domain feature benchmarks, and time-frequency-domain feature benchmarks, combined with feature weights, constructs an electrical signal benchmark matrix;

[0139] S17, calculate the feature similarity between the electrical signal feature matrix and the electrical signal reference matrix;

[0140] S18. Based on feature similarity, an arc fault is identified using a preset similarity threshold.

[0141] In one embodiment, in response to the detection of an arc fault, the electrical signals are compared and analyzed based on circuit topology data to locate the arc fault region, including:

[0142] S21, Extract the topological relationships and location association information of each circuit branch from the circuit topology data;

[0143] S22, in response to the detection of an arc fault, divides the electrical signals according to the topological correlation to obtain the branch electrical signals;

[0144] S23, calculate the characteristic deviation value between the characteristic matrix of each branch signal and the reference matrix of the signal;

[0145] S24. Based on the preset deviation threshold, the characteristic deviation value of each branch electrical signal is compared with the preset deviation threshold to determine the candidate branch with fault association.

[0146] S25. For candidate branches, based on topological relationships, perform temporal correlation analysis on feature deviation values ​​to determine target associated branches;

[0147] S26. Based on the target associated branch and combined with the location association information, locate the arc fault area.

[0148] In one embodiment, for the arc fault area, abnormal discharge patterns are extracted by combining electrical signals, including:

[0149] S31, based on the electrical signal corresponding to the arc fault area, extract the current extreme value difference, current standard deviation and current change rate within a preset time period from the time domain features; extract the energy mutation value and higher harmonic amplitude features of a preset frequency band from the frequency domain features; extract wavelet transform feature quantities and intrinsic mode functions from the time and frequency domain features;

[0150] S32 integrates current extreme value difference, current standard deviation, current change rate, energy mutation value, higher harmonic amplitude characteristics, wavelet transform characteristics, and intrinsic mode function as characteristic parameters to generate abnormal discharge modes.

[0151] In one embodiment, the preset machine learning model is a gradient boosting tree model. The gradient boosting tree model presets the fault type corresponding to each feature parameter in the abnormal discharge mode, the mapping matching logic between feature parameters and fault types, and the path generation rules.

[0152] S41, based on the abnormal discharge mode, uses a pre-set machine learning model to determine potential fault evolution paths, including:

[0153] S42, based on the characteristic parameters of the abnormal discharge mode, the contribution of each characteristic parameter to the arc fault is determined by the entropy weight method.

[0154] S43, based on contribution, select the top N feature parameters by contribution as core feature parameters;

[0155] S44, based on the core feature parameters, uses the gradient boosting tree model mapping matching logic to match the corresponding fault type;

[0156] S45, combining the fault type and topology association, generates multiple fault evolution paths according to the path generation rules, forming a candidate path set;

[0157] S46, The candidate path set is probabilistically ranked using a Bayesian network to obtain the probabilistic ranking results;

[0158] S47. Based on the probability ranking results, potential failure evolution paths are determined from the candidate path set.

[0159] In one embodiment, S51, based on contribution, the top N feature parameters with the highest contribution are selected as core feature parameters, which is achieved by the following formula:

[0160]

[0161] Where S is the set of core feature parameters, Let be the i-th characteristic parameter in the abnormal discharge mode, where i is the index of the characteristic parameter, and m is the total number of characteristic parameters in the abnormal discharge mode. Let be the contribution of the i-th characteristic parameter to the arc fault. for The ranking numbers are sorted in descending order, where N is the preset number of core feature parameters to be selected, and N is a positive integer. , This represents the contribution value of the kth feature parameter after sorting the contribution values ​​in descending order. The calibration coefficient for the effectiveness of contribution. The standard deviation of the contribution of all feature parameters The mean of the contributions of all feature parameters. It is a local minimum constant.

[0162] In one embodiment, an arc fault alarm is generated by integrating abnormal discharge modes and potential fault evolution paths, including:

[0163] S61, based on the core characteristic parameters of the abnormal discharge mode and the probability ranking results of potential fault evolution paths, extract key path nodes;

[0164] S62, based on core feature parameters and combined with critical path nodes, calculates fault association values ​​according to preset node weight rules;

[0165] S63, based on the operating parameters of the photovoltaic power station, determine the threshold range of multiple alarm levels corresponding to the fault-related values;

[0166] S64, determine the alarm level based on the fault correlation value and the level threshold range;

[0167] S65 integrates core characteristic parameters, critical path nodes, and alarm levels to generate arc fault alarms.

[0168] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the function of a rapid arc fault detection device for a photovoltaic power station as described above.

[0169] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs the functions described in the above-described device embodiments.

[0170] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0171] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A rapid arc fault detection device for a photovoltaic power station, characterized in that, The device includes: The power plant data acquisition module is used to acquire circuit topology data and electrical signals of the photovoltaic power plant; An arc fault identification module is used to identify arc faults based on the electrical signal and according to a preset arc fault detection system. The fault area location module is used to locate the arc fault area by comparing and analyzing the electrical signal according to the circuit topology data in response to the detection of the arc fault. An abnormal discharge mode extraction module is used to extract abnormal discharge modes for the arc fault area in combination with the electrical signal. The fault evolution path determination module is used to determine potential fault evolution paths based on the abnormal discharge mode and using a preset machine learning model. An arc fault alarm module is used to integrate the abnormal discharge mode and the potential fault evolution path to generate an arc fault alarm.

2. The apparatus according to claim 1, characterized in that, The arc fault identification module is also used for: Based on the electrical signal, time-domain features, frequency-domain features, and time-frequency-domain features are extracted; By integrating the time-domain features, the frequency-domain features, and the time-frequency-domain features, an electrical signal feature matrix is ​​constructed. Obtain the working environment parameters and operating parameters of the photovoltaic power station; Based on the working environment parameters and the operating parameters, set the time-domain characteristic benchmark, the frequency-domain characteristic benchmark, and the time-frequency-domain characteristic benchmark; The feature weights of the time-domain feature benchmark, the frequency-domain feature benchmark, and the time-frequency-domain feature benchmark are determined using the analytic hierarchy process (AHP). Based on the time-domain feature reference, the frequency-domain feature reference, and the time-frequency-domain feature reference, and combined with the feature weights, an electrical signal reference matrix is ​​constructed; Calculate the feature similarity between the electrical signal feature matrix and the electrical signal reference matrix; Based on the feature similarity, the arc fault is identified using a preset similarity threshold.

3. The apparatus according to claim 2, characterized in that, The fault area location module is also used for: Extract the topological relationships and location associations of each circuit branch from the circuit topology data; In response to the detection of the arc fault, the electrical signal is divided according to the topological correlation to obtain the branch electrical signal; Calculate the characteristic deviation value between the electrical signal characteristic matrix of each branch and the electrical signal reference matrix; Based on a preset deviation threshold, the characteristic deviation value of each branch electrical signal is compared with the preset deviation threshold to determine candidate branches with fault associations. For the candidate branches, and in conjunction with the topological association, a temporal correlation analysis is performed on the feature deviation values ​​to determine the target associated branches; Based on the target associated branch and the location association information, the arc fault area is located.

4. The apparatus according to claim 3, characterized in that, The abnormal discharge mode extraction module is also used for: Based on the electrical signal corresponding to the arc fault region, the current extreme value difference, current standard deviation, and current change rate within a preset time period are extracted from the time domain features; the energy mutation value and higher harmonic amplitude features of a preset frequency band are extracted from the frequency domain features; and wavelet transform features and intrinsic mode functions are extracted from the time and frequency domain features. The abnormal discharge mode is generated by integrating the current extreme value difference, the current standard deviation, the current change rate, the energy mutation value, the higher harmonic amplitude characteristics, the wavelet transform characteristics, and the intrinsic mode function as feature parameters.

5. The apparatus according to claim 4, characterized in that, In the fault evolution path determination module, the preset machine learning model is a gradient boosting tree model. The gradient boosting tree model presets the fault type corresponding to each feature parameter in the abnormal discharge mode, the mapping matching logic between the feature parameters and the fault type, and the path generation rules. The step of determining potential fault evolution paths based on the abnormal discharge mode using a preset machine learning model includes: Based on the characteristic parameters of the abnormal discharge mode, the contribution of each characteristic parameter to the arc fault is determined by the entropy weight method. Based on the contribution, the top N feature parameters with the highest contribution are selected as core feature parameters. Based on the core feature parameters, the corresponding fault type is matched through the mapping and matching logic of the gradient boosting tree model; Combining the fault type with the topology association, multiple fault evolution paths are generated according to the path generation rules to form a candidate path set; The candidate path set is probabilistically ranked using a Bayesian network to obtain the probabilistic ranking result. Based on the probability ranking results, the potential failure evolution path is determined from the candidate path set.

6. The apparatus according to claim 5, characterized in that, The fault evolution path determination module is also used to select the top N feature parameters based on the contribution degree as core feature parameters using the following formula: Where S is the set of core feature parameters, Let be the i-th characteristic parameter in the abnormal discharge mode, where i is the index of the characteristic parameter, and m is the total number of characteristic parameters in the abnormal discharge mode. Let be the contribution of the i-th characteristic parameter to the arc fault. for The ranking numbers are sorted in descending order, where N is the preset number of core feature parameters to be selected, and N is a positive integer. , This represents the contribution value of the kth feature parameter after sorting the contribution values ​​in descending order. The calibration coefficient for the effectiveness of contribution. The standard deviation of the contribution of all feature parameters The mean of the contributions of all feature parameters. It is a local minimum constant.

7. The apparatus according to claim 5, characterized in that, The arc fault alarm module is also used for: Based on the core characteristic parameters of the abnormal discharge mode and the probability ranking results of the potential fault evolution path, key path nodes are extracted. Based on the core feature parameters and the critical path nodes, the fault association value is calculated according to the preset node weight rules. Based on the operating parameters of the photovoltaic power station, determine the threshold range of multiple alarm levels corresponding to the fault correlation value; The alarm level is determined based on the fault correlation value and the level threshold range; The arc fault alarm is generated by integrating the core feature parameters, the critical path nodes, and the alarm level.

8. A method for rapid arc fault detection in a photovoltaic power station, characterized in that, The method includes: Acquire circuit topology data and electrical signals of photovoltaic power plants; Based on the electrical signal, an arc fault is identified according to a preset arc fault detection system; In response to the detection of the arc fault, the electrical signals are compared and analyzed based on the circuit topology data to locate the arc fault area; For the arc fault area, the abnormal discharge mode is extracted based on the electrical signal; Based on the abnormal discharge mode, a preset machine learning model is used to determine the potential fault evolution path; By integrating the abnormal discharge mode and the potential fault evolution path, an arc fault alarm is generated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it performs the functions of the apparatus according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the function of the apparatus according to any one of claims 1 to 7.

Images