A Fault Diagnosis System and Method for Photovoltaic Distribution Network

Abstract

The present invention discloses a fault diagnosis system and method for a photovoltaic distribution network, and relates to the technical field of fault diagnosis of distribution networks. The method includes collecting multi-dimensional real-time data of a photovoltaic array, adaptively partitioning the photovoltaic array, and performing multi-level classification of photovoltaic components based on the partitioning results; performing multi-dimensional feature extraction on the photovoltaic array, and constructing a multi-dimensional tensor model of photovoltaic component performance characteristics based on the multi-dimensional feature extraction results, extracting feature subspaces, and performing performance feature mapping on the photovoltaic array; based on the performance feature mapping results, performing intelligent diagnosis of local shading of photovoltaic components, and generating a dynamic heat map of shading effects; based on the dynamic heat map of shading effects, accurately locating faults of photovoltaic arrays. The present invention realizes the intelligence and automation of the entire process from data collection, performance analysis to fault location, and overcomes the limitations of traditional diagnostic methods such as reliance on manual investigation, low efficiency, and high false alarm rate.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault diagnosis, and in particular to a photovoltaic distribution network fault diagnosis system and method. Background Art

[0002] With the growing global demand for renewable energy, photovoltaic power generation has become one of the important ways to solve energy shortages and environmental problems. In this context, the large-scale application of photovoltaic power stations continues to expand, especially the integration in distribution networks. As the scale of photovoltaic arrays increases and their complexity increases, the diagnosis and maintenance of photovoltaic system faults become more complex and difficult.

[0003] Existing fault diagnosis methods have many deficiencies when dealing with the dynamic environmental changes of photovoltaic arrays. For example, current systems are often limited to local data analysis, making it difficult to accurately identify potential faults in the complex spatial structure and multi-dimensional environmental characteristics of photovoltaic arrays. More importantly, these methods usually lack a deep understanding of the interactions between photovoltaic modules and cannot accurately judge the operating status of photovoltaic modules under different environmental conditions (such as local shading, light fluctuations, temperature changes, etc.). Summary of the invention

[0004] In view of the problems existing in the existing fault diagnosis methods, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to solve the problems of photovoltaic module performance monitoring, fault location and local shading identification.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a fault diagnosis method for a photovoltaic distribution network, which includes collecting multi-dimensional real-time data of a photovoltaic array, adaptively partitioning the photovoltaic array, and performing multi-level classification of photovoltaic components based on the partitioning results; performing multi-dimensional feature extraction on the photovoltaic array, and constructing a multi-dimensional tensor model of photovoltaic component performance characteristics based on the multi-dimensional feature extraction results, extracting feature subspaces, and performing performance feature mapping on the photovoltaic array; based on the performance feature mapping results, performing intelligent diagnosis of local shading of photovoltaic components and generating a dynamic heat map of shading effects; based on the dynamic heat map of shading effects, accurately locating faults of the photovoltaic array;

[0008] The method of adaptively partitioning the photovoltaic array includes: collecting multi-dimensional data of the photovoltaic array in real time to generate a standardized multi-dimensional performance indicator matrix; based on the multi-dimensional performance indicator matrix, applying the self-organizing map SOM algorithm to adaptively partition the photovoltaic array, including: setting grid nodes, each node With the same input vector Same dimensions:

[0009] ,

[0010] Where N is the number of nodes; calculate the Euclidean distance between the node and the input, select the node with the smallest Euclidean distance as the winning node, and and neighborhood functions Update the weights of the node and its neighbors:

[0011] ,

[0012] Where t is the time step of the current iteration.

[0013] As a preferred solution of the photovoltaic power distribution network fault diagnosis method of the present invention, wherein: the neighborhood function Select the Gaussian distribution function.

[0014] As a preferred solution of the fault diagnosis method of the photovoltaic distribution network of the present invention, the performance characteristic mapping of the photovoltaic array includes: comparing the characteristic vectors of each component with other components, calculating the similarity between the components, and generating a similarity matrix; using the similarity matrix as input, constructing a performance characteristic mapping relationship matrix D, in which each element Representation Components With components The performance similarity between them; the performance feature mapping relationship matrix D is a symmetric matrix.

[0015] As a preferred solution of the fault diagnosis method of the photovoltaic distribution network described in the present invention, the intelligent diagnosis of local shading of photovoltaic components based on the performance feature mapping results includes the following steps: based on the performance feature mapping relationship matrix D, a graph structure G=(V,E) is constructed, wherein the vertex set V represents the photovoltaic components and the edge set E represents the performance similarity relationship between the components; the spectral clustering algorithm is applied to segment the graph structure G to identify potential fault clusters; for each potential fault cluster, the local shading possibility index LOPI is calculated:

[0016] ,

[0017] in, , and They represent the comprehensive performance index, temperature and irradiance difference between the components in the cluster and the surrounding normal components, represents the spatial compactness of the cluster, ~ is the weight coefficient; based on the LOPI value, the adaptive threshold method is used to determine the final fault cluster.

[0018] As a preferred solution of the fault diagnosis method of the photovoltaic distribution network described in the present invention, the method of using an adaptive threshold method to determine the final fault cluster includes: setting the sum of the mean and standard deviation of the LOPI value as an adaptive threshold, and for each potential fault cluster, if the LOPI value is greater than or equal to the adaptive threshold, identifying the cluster as the final fault cluster.

[0019] As a preferred solution of the fault diagnosis method of the photovoltaic distribution network described in the present invention, the accurate fault location of the photovoltaic array based on the dynamic heat map of shading influence includes the following steps: based on the dynamic heat map of shading influence, the regional growing algorithm is applied to accurately locate the boundary of the fault area, and the geometric center and area of ​​each fault area are calculated; for each fault area, the performance data of the surrounding components, historical fault records and environmental data are combined, and the potential shading type is judged through statistical analysis and historical trend matching, and the results are transmitted to the management personnel.

[0020] As a preferred solution of the fault diagnosis method of the photovoltaic distribution network described in the present invention, the application of the regional growing algorithm to accurately locate the boundary of the fault area includes: selecting the point with the highest thermal value in the dynamic thermal map affected by the occlusion as the seed point, setting the growth threshold ; Check the 8 neighboring pixels of the seed point. If the heat value of the neighboring pixels is greater than the growth threshold , add it to the current area and use it as a new seed point, repeat this process until no new pixels can be added; use an edge detection algorithm to extract the precise boundary of the fault area at the edge of the growing area; the determination of potential occlusion types includes: calculating the performance difference percentage between the fault area and the normal area, and analyzing the time pattern of performance degradation; matching the current performance data with the historical fault records, and finding the potential occlusion type by comparing the performance degradation pattern; comparing various data indicators with the existing occlusion pattern, and determining the possible occlusion type according to the degree of similarity.

[0021] In a second aspect, an embodiment of the present invention provides a fault diagnosis system for a photovoltaic distribution network, which includes:

[0022] The photovoltaic array dynamic partitioning module is used to collect multi-dimensional real-time data of the photovoltaic array, adaptively partition the photovoltaic array, and perform multi-level classification of photovoltaic components based on the partitioning results; the dynamic mapping module is used to extract multi-dimensional features of the photovoltaic array, build a multi-dimensional tensor model of photovoltaic component performance characteristics based on the multi-dimensional feature extraction results, extract feature subspaces, and perform performance feature mapping of the photovoltaic array; the fault diagnosis dynamic model module is used to perform intelligent diagnosis of local shading of photovoltaic components based on the performance feature mapping results, and generate a dynamic thermal map of shading effects; the fault location and repair module is used to accurately locate faults in the photovoltaic array based on the dynamic thermal map of shading effects.

[0023] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the fault diagnosis method for the photovoltaic distribution network as described in the first aspect of the present invention are implemented.

[0024] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the fault diagnosis method for the photovoltaic distribution network as described in the first aspect of the present invention are implemented.

[0025] The beneficial effects of the present invention are as follows: the present invention uses the self-organizing map (SOM) algorithm to partition the photovoltaic array in real time, so that the system can implement differentiated operation and maintenance measures for different efficiency zones, and the self-organizing map (SOM) algorithm and other methods significantly improve the speed and accuracy of local occlusion detection. The fault diagnosis system of the photovoltaic distribution network of the present invention realizes the intelligence and automation of the entire process from data collection, performance analysis to fault location, overcoming the limitations of traditional diagnostic methods such as reliance on manual investigation, low efficiency, and high false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 Flow chart of the fault diagnosis method for photovoltaic distribution network.

[0028] Figure 2 This is the structural diagram of the fault diagnosis system for photovoltaic distribution network. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0032] Example 1, reference Figure 1~Figure 2 , which is the first embodiment of the present invention, and provides a fault diagnosis method for a photovoltaic distribution network, comprising:

[0033] S1: Collect multi-dimensional real-time data of the photovoltaic array, including current, voltage, temperature and irradiance, etc., adaptively partition the photovoltaic array, and perform multi-level classification of photovoltaic modules based on the partitioning results.

[0034] Among them, the zones include high-efficiency zone, standard zone and low-efficiency zone.

[0035] Specifically, by deploying a high-precision sensor network in the photovoltaic array, multi-dimensional data of each photovoltaic module can be collected in real time. These data include not only basic operating parameters such as current, voltage and temperature, but also additional factors that affect power generation performance, such as irradiance, module tilt and ambient humidity.

[0036] By collecting these rich multi-dimensional data, the status of each PV module can be fully characterized and an accurate set of performance indicators can be formed.

[0037] The data preprocessing algorithm is used to denoise, normalize and time-synchronize the raw data to generate a standardized multi-dimensional performance indicator matrix.

[0038] Based on the multi-dimensional performance index matrix, the self-organizing map SOM algorithm is used to adaptively partition the photovoltaic array. The specific operations are as follows:

[0039] Set the grid nodes, each node With the same input vector Same dimensions, i.e., the initial nodes are random values:

[0040] ,

[0041] Where N is the number of nodes.

[0042] Compute the Euclidean distance between a node and the input:

[0043] ,

[0044] in, is the standardized value of the jth performance indicator of PV module i, is the weight value of the i-th node in the j-th dimension; nw is the number of dimensions of the multidimensional performance indicator.

[0045] The node with the smallest distance is selected as the winning node, and the learning rate and neighborhood functions Update the weights of the node and its neighbors:

[0046] ,

[0047] Among them, the learning rate Usually it decreases gradually with time, t is the time step of the current iteration, and the neighborhood function Determines the update range, usually using a Gaussian distribution function.

[0048] Furthermore, by calculating the comprehensive performance index (CPI) of each component, the photovoltaic components are divided into high efficiency area (CPI> 0.9), standard area (0.7≤CPI≤0.9) and low efficiency area (CPI<0.7). The component number sets of each area are generated and recorded as H1 (high efficiency area), H2 (standard area) and H3 (low efficiency area).

[0049] Optionally, different treatments are performed for components in different partitions. For example, for components in the high-efficiency zone: continue to maintain normal operation and monitor in real time whether there is a trend of performance degradation; for components in the standard zone: perform performance analysis regularly, monitor potential problems, and prevent further decline in component efficiency; for components in the low-efficiency zone: perform precise maintenance on components in the low-efficiency zone, which may include replacement, cleaning or optimization of components to restore their efficiency. This not only makes the operation of the entire PV array more efficient, but also effectively extends the life of the components and reduces the occurrence of failures.

[0050] S2: Perform multi-dimensional feature extraction on the photovoltaic array, build a multi-dimensional tensor model of photovoltaic module performance characteristics based on the feature extraction results, extract the feature subspace, and perform performance feature mapping on the photovoltaic array.

[0051] S2.1: Perform multi-dimensional feature extraction on each photovoltaic module, including:

[0052] Time domain characteristics, such as mean, variance, skewness, kurtosis and other statistics;

[0053] Frequency domain features, such as using fast Fourier transform (FFT) to analyze signal spectrum characteristics;

[0054] Time-frequency domain features, including the application of wavelet transform to analyze the time-frequency characteristics of signals;

[0055] Nonlinear features, such as calculating sample entropy, approximate entropy and other complexity indicators.

[0056] Optionally, for components in different partitions of H1, H2, and H3, different data collection and analysis strategies are adopted according to their partition characteristics. For example, H1 (high efficiency zone): perform monitoring at a lower frequency, because the performance of the components is relatively stable, and focus on monitoring whether there are signs of sudden performance decline; H2 (standard zone): perform regular monitoring, pay special attention to performance fluctuations, and apply feature extraction methods to detect potential efficiency decline trends; H3 (low efficiency zone): perform high-frequency and detailed feature extraction for these components, focus on monitoring the operating status and health of the components, and find the causes of low efficiency.

[0057] S2.2: Construct a multidimensional tensor model of PV module performance characteristics.

[0058] The extracted multidimensional features are organized into a fourth-order tensor, including component number, feature type, time window, and data dimension, and tensor decomposition techniques (such as Tucker decomposition) are used to reduce the data dimension and extract the feature subspace. .

[0059] S2.3: Feature subspace based on dimensionality reduction , calculate the similarity between each component and the ideal performance template, and generate the performance feature mapping relationship matrix D, where Representation Components With components The performance similarity between them is as follows:

[0060] Compare the feature vectors of each component with those of other components, calculate the similarity between components, and generate a similarity matrix, where the similarity can be calculated by Euclidean distance or cosine similarity, and construct a performance feature mapping relationship matrix D, where each element Representation Components With components The performance similarity between them, and the performance feature mapping relationship matrix D is a symmetric matrix.

[0061] It can be seen that in the performance feature mapping process, the system will form a mapping relationship based on the similarity matrix. This mapping relationship can be represented by the matrix D, which represents the performance correlation between the components in the photovoltaic array.

[0062] S3: Based on the performance characteristic mapping results, intelligent diagnosis of local shading of photovoltaic modules is carried out to generate a dynamic heat map of shading impact.

[0063] Based on the performance feature mapping relationship matrix D, a graph structure G=(V,E) is constructed, where the vertex set V represents photovoltaic modules and the edge set E represents the performance similarity relationship between modules.

[0064] The spectral clustering algorithm is applied to segment the graph structure G and identify potential fault clusters.

[0065] For each potential fault cluster, the Local Occlusion Probability Index (LOPI) is calculated:

[0066] ,

[0067] in, , and They represent the comprehensive performance index, temperature and irradiance difference between the components in the cluster and the surrounding normal components, represents the spatial compactness of the cluster, ~ is the weight coefficient;

[0068] Based on the LOPI value, the adaptive threshold method is used to determine the final fault cluster, and then the Kriging interpolation method is used to generate the dynamic thermal map of occlusion impact, where the thermal value Indicates location The severity of local occlusion at the location.

[0069] Optionally, H1 components usually appear normal during the diagnostic process, so their diagnostic priority can be lowered; H2 components need to be regularly checked for local occlusion, with a focus on identifying whether there is a potential cluster of faults; H3 components need to be prioritized for diagnosis of occlusion or other faults to facilitate a quick response to restore their operating efficiency.

[0070] It should be noted that the use of the adaptive threshold method to determine the final fault cluster includes: setting the sum of the mean and standard deviation of the LOPI value as the adaptive threshold, and for each potential fault cluster, if its LOPI value is greater than or equal to the adaptive threshold, then the cluster is identified as the final fault cluster. At this time, the photovoltaic modules in the cluster may be affected by local shading or other performance problems.

[0071] S4: Accurately locate faults in photovoltaic arrays based on dynamic thermal maps of shading effects.

[0072] S4.1: Based on the dynamic heat map of occlusion effects, the region growing algorithm is used to accurately locate the fault region boundary and calculate the geometric center and area of ​​each fault region. The specific operations are as follows:

[0073] Select the point with the highest thermal value in the dynamic thermal map affected by occlusion as the seed point and set the growth threshold , usually the average value of the thermal value plus one standard deviation; check the 8 neighboring pixels of the seed point, if the thermal value of the neighboring pixel is greater than the growth threshold , add it to the current area and use it as a new seed point, and repeat this process until no new pixels can be added.

[0074] Use an edge detection algorithm (such as the Canny algorithm) to extract the exact boundaries of the fault area at the edge of the growing area.

[0075] The geometric center of each fault area is calculated as:

[0076] ,

[0077] ,

[0078] in, is the coordinate of each pixel in the region, is the total number of pixels.

[0079] Calculate the area of ​​the fault region:

[0080] ,

[0081] in, The actual area represented by each pixel.

[0082] S4.2: For each fault area, combine the performance data of surrounding components, historical fault records, and environmental data, and determine the potential types of obstructions, such as dust, leaves, or snow, through statistical analysis and historical trend matching, and transmit the results to management personnel.

[0083] Collect performance data of components in the fault area (such as output power, current, voltage), obtain performance data of normal components in the surrounding area as a reference, extract relevant historical fault records, and obtain current and recent meteorological data (temperature, humidity, wind speed, precipitation, etc.).

[0084] Calculate the percentage of performance difference between the faulty area and the normal area, and analyze the time pattern of performance degradation (sudden or gradual).

[0085] Match the current performance data with historical fault records, and find potential types of obstructions by comparing similar performance degradation patterns. For example, if the voltage continues to drop and is accompanied by an increase in temperature, it may be dust obstruction; if the irradiance fluctuates violently and there is snow obstruction in the historical records, it may be obstruction caused by snow accumulation.

[0086] Based on performance data, environmental data, and historical fault records, statistical analysis, such as linear regression and anomaly detection, is performed to evaluate the impact of current shading on component performance.

[0087] Compare various data indicators with existing occlusion patterns (dust, leaves, snow, etc.), and determine the possible occlusion type based on the degree of similarity.

[0088] For example: if the irradiance is stable but the voltage and current continue to decrease and the temperature rises, it may be dust; if the irradiance and current fluctuate intermittently and the irradiance fluctuation is large, it may be leaves; if the irradiance and temperature decrease simultaneously and the current fluctuation is more significant, it may be snow accumulation.

[0089] Occlusion impact dynamic heat map and actual performance data, by comparing the intensity of occlusion impact in different areas, generating an occlusion severity distribution map.

[0090] The statistical analysis results, shading type judgment, and shading severity information are summarized to generate a report, where the report content includes the specific shading type and severity of each fault area, the performance data of the affected components and historical trend comparison, and suggestions for maintenance measures that may be needed in the future.

[0091] Among them, H1 components only require routine maintenance due to the low probability of failure; H2 components need to focus on whether the boundaries in the regional growth algorithm extend to the H2 area to ensure that potential failures do not spread; H3 components need to focus on locating the fault areas of these inefficient components and quickly take maintenance measures based on real-time data, such as cleaning, replacing components or repairing. The analysis results are transmitted to management personnel through the management system as a basis for subsequent maintenance decisions. Relevant personnel can be reminded through text messages, emails or system notifications to deal with the faulty areas in a timely manner.

[0092] Furthermore, this embodiment also provides a fault diagnosis system for a photovoltaic power distribution network, comprising:

[0093] The photovoltaic array dynamic partitioning module collects multi-dimensional real-time data of the photovoltaic array, including current, voltage, temperature and irradiance, etc., and performs adaptive partitioning of the photovoltaic array. Based on the partitioning results, the photovoltaic modules are classified into multiple levels.

[0094] Dynamic mapping module: extract multi-dimensional features of the photovoltaic array, build a multi-dimensional tensor model of photovoltaic module performance characteristics based on the feature extraction results, and map the performance characteristics of the photovoltaic array;

[0095] Fault diagnosis dynamic model module: Based on the performance feature mapping results, it conducts intelligent diagnosis of local shading of photovoltaic modules and generates dynamic heat maps of shading impacts;

[0096] Fault location and repair module: Based on the dynamic thermal map of shading effects, accurate fault location of the photovoltaic array is performed.

[0097] This embodiment also provides a computer device, which is suitable for the fault diagnosis system of a photovoltaic distribution network, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the fault diagnosis system of the photovoltaic distribution network proposed in the above embodiment.

[0098] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0099] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the fault diagnosis system for a photovoltaic power distribution network as proposed in the above embodiment is implemented.

[0100] In summary, the present invention uses the self-organizing map (SOM) algorithm to partition the photovoltaic array in real time, so that the system can implement differentiated operation and maintenance measures for different efficiency zones, and through the self-organizing map (SOM) algorithm and other methods, the speed and accuracy of local occlusion detection are significantly improved. The fault diagnosis system of the photovoltaic distribution network of the present invention realizes the intelligence and automation of the whole process from data collection, performance analysis to fault location, overcoming the limitations of traditional diagnostic methods such as reliance on manual investigation, low efficiency, and high false alarm rate.

[0101] Embodiment 2, referring to Table 1 and Table 2, is the second embodiment of the present invention. This embodiment provides a fault diagnosis method for a photovoltaic distribution network. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0102] In order to ensure the comprehensiveness and representativeness of the experimental data, 1,000 photovoltaic modules of a photovoltaic power station were selected, and 7 typical modules (numbered P1 to P7) were randomly selected from them for long-term operating status monitoring.

[0103] The historical data of each component includes the following parameters: operating temperature (°C), power generation (kW), radiation intensity (W / m²), shading ratio (%), component health (%), efficiency loss (%). The data collection frequency is recorded every 10 minutes.

[0104] The experiments are divided into two groups: the first group adopts the traditional fault detection method (detection based on static threshold setting), and the second group adopts the fault diagnosis method proposed by the present invention.

[0105] Before the experiment begins, each PV module is calibrated to its initial state to ensure that the health and performance of each module are in a relatively consistent baseline state.

[0106] Subsequently, the PV array was operated normally for 48 hours, during which partial shading of the component surface was randomly simulated (for example, 20% to 80% shading ratio) to simulate the impact of different environmental factors on the components.

[0107] In the experiment, the traditional method monitors faults by setting a fixed power generation threshold, while the present invention collects real-time multi-dimensional data and inputs it into the SOM algorithm for adaptive learning, generates dynamic heat maps and real-time partitioning results, and combines the multi-dimensional tensor analysis model to judge the fault situation.

[0108] The detailed steps of the experiment are as follows:

[0109] The basic operating data such as the operating temperature, power generation, radiation intensity, etc. of the components are collected in real time, and invalid data is filtered out through a denoising algorithm. To ensure the accuracy of the diagnosis, the shading conditions and health data of all components are monitored simultaneously.

[0110] Based on the traditional fault diagnosis method, the system monitors faults by setting fixed thresholds. If the power generation is lower than a fixed value, it is considered that the component may be faulty and an alarm is issued.

[0111] The present invention dynamically generates a thermal map of the photovoltaic array based on multidimensional data, and divides the components into zones according to health and power generation. Based on this, a multidimensional tensor is used to analyze the performance of the components in each zone to determine whether they are in an abnormal working state.

[0112] At the same time, through dynamic thermal maps and occlusion monitoring, the impact of local occlusion on component performance can be tracked in real time, thereby improving detection accuracy.

[0113] The system records the test results of each component after 48 hours of experiment, and compares and analyzes the test results of the traditional method and the method of the present invention. Some data are as follows:

[0114] Table 1 Operating status of photovoltaic modules (traditional method)

[0115] Component Number Working temperature(℃) Power generation (kW) Radiant intensity (W / m²) Occlusion ratio (%) Component health (%) Fault detection results P1 45 8.5 850 20 95 normal P2 50 6.8 780 50 88 Fault P3 43 7.9 820 35 90 normal P4 47 6.5 760 65 85 Fault P5 42 9.2 900 10 98 normal P6 49 7.1 770 60 87 Fault P7 46 8.7 880 15 96 normal

[0116] Table 2 Operation status of photovoltaic modules (method of the present invention)

[0117] Component Number Working temperature(℃) Power generation (kW) Radiant intensity (W / m²) Occlusion ratio (%) Component health (%) SOM Partitions Fault detection results P1 45 8.5 850 20 95 A normal P2 50 6.8 780 50 88 B Fault P3 43 7.9 820 35 90 A normal P4 47 6.5 760 65 85 B Fault P5 42 9.2 900 10 98 A normal P6 49 7.1 770 60 87 B Fault P7 46 8.7 880 15 96 A normal

[0118] It can be seen from the above two tables that the present invention can partition the components in real time according to parameters such as the health of the photovoltaic components and the power generation capacity. The components with a large shading ratio, such as P2, P4, and P6 in Table 2, are divided into "Zone B" (a lower health interval) by SOM. The component failure probability corresponding to this partition is higher, and it is finally accurately judged as a faulty component after multi-dimensional tensor analysis. In contrast, the traditional method is only based on static threshold detection, which cannot fully identify the dynamics and complexity of shading faults, which may lead to misjudgment or missed judgment.

[0119] Experiments also show that traditional methods cannot distinguish and manage components under different occlusion conditions, while the present invention can not only identify the impact of occlusion through SOM dynamic partitioning, but also automatically adjust the threshold within the partition. The table shows that components such as P1, P3, P5, and P7 are in normal working condition due to their high health, thus avoiding unnecessary maintenance and replacement operations and significantly improving the system's operation and maintenance efficiency.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for a photovoltaic distribution network, characterized in that: include, Collect multi-dimensional real-time data of photovoltaic arrays, adaptively partition the photovoltaic arrays, and perform multi-level classification of photovoltaic modules based on the partition results; The method of adaptively partitioning the photovoltaic array and performing multi-level classification of photovoltaic components based on the partitioning results includes: Collect multi-dimensional data of photovoltaic arrays in real time and generate a standardized multi-dimensional performance index matrix; Based on the multi-dimensional performance index matrix, the self-organizing map (SOM) algorithm is used to adaptively partition the photovoltaic array, including: Set the grid nodes, each node With the same input vector Same dimensions: ， Where N is the number of nodes; Calculate the Euclidean distance between the node and the input, select the node with the smallest Euclidean distance as the winning node, and and neighborhood functions Update the weights of the node and its neighbors: ， Among them, t is the time step of the current iteration, and the neighborhood function Select Gaussian distribution function; Furthermore, by calculating the comprehensive performance index CPI of each photovoltaic module, the photovoltaic modules are divided into high efficiency zone, standard zone and low efficiency zone; Performing multi-dimensional feature extraction on the photovoltaic array, wherein the performance characteristics of photovoltaic modules in the standard area and the low-efficiency area are extracted, and based on the multi-dimensional feature extraction results, a multi-dimensional tensor model of photovoltaic module performance characteristics is constructed, the feature subspace is extracted, and performance characteristic mapping is performed on the photovoltaic array; Based on the performance characteristic mapping results, an intelligent diagnosis of local shading of photovoltaic modules is performed to generate a dynamic thermal map of shading impact; Based on the dynamic thermal map of shading effects, the photovoltaic array faults can be accurately located; The performance characteristic mapping of the photovoltaic array comprises: Compare the feature vectors of each component with those of other components, calculate the similarity between components, and generate a similarity matrix; Taking the similarity matrix as input, we construct the performance feature mapping relationship matrix D, where each element Representation Components With components The performance similarity between The performance characteristic mapping relationship matrix D is a symmetric matrix; The method of performing intelligent diagnosis of partial shading of photovoltaic components based on the performance characteristic mapping result comprises the following steps: Based on the performance feature mapping relationship matrix D, a graph structure G=(V,E) is constructed, where the vertex set V represents the photovoltaic modules and the edge set E represents the performance similarity relationship between the modules; Apply spectral clustering algorithm to segment the graph structure G and identify potential fault clusters; For each potential fault cluster, the local occlusion probability index LOPI is calculated: ， in, , and They represent the comprehensive performance index, temperature and irradiance difference between the components in the cluster and the surrounding normal components, represents the spatial compactness of the cluster, ~ is the weight coefficient; Based on the LOPI values, an adaptive threshold method is used to determine the final fault clusters.

2. The fault diagnosis method for a photovoltaic power distribution network according to claim 1, characterized in that: The method of using the adaptive threshold method to determine the final fault cluster includes: The sum of the mean and standard deviation of the LOPI value is set as the adaptive threshold. For each potential fault cluster, if the LOPI value is greater than or equal to the adaptive threshold, the cluster is identified as the final fault cluster.

3. The fault diagnosis method for a photovoltaic power distribution network according to claim 2, characterized in that: The method of accurately locating faults of a photovoltaic array based on the dynamic thermal map of shading effects comprises the following steps: Based on the dynamic heat map of occlusion effects, the region growing algorithm is used to accurately locate the fault region boundary and calculate the geometric center and area of ​​each fault region; For each fault area, the performance data of surrounding components, historical fault records and environmental data are combined to determine the potential type of obstruction through statistical analysis and historical trend matching, and the results are transmitted to management personnel.

4. The fault diagnosis method for a photovoltaic power distribution network as claimed in claim 3, characterized in that: The application of the region growing algorithm to accurately locate the fault region boundary includes: Select the point with the highest thermal value in the dynamic thermal map affected by occlusion as the seed point and set the growth threshold ; Check the neighboring pixels of the seed point, if the thermal value of the neighboring pixels is greater than the growth threshold , add it to the current area and use it as a new seed point, and repeat this process until no new pixels can be added; An edge detection algorithm is used to extract the precise boundary of the fault area at the edge of the growth area; The determination of potential occlusion types includes: Calculate the percentage of performance difference between the faulty area and the normal area, and analyze the time pattern of performance degradation; Match current performance data with historical fault records to find potential occlusion types by comparing performance degradation patterns; Compare various data indicators with existing occlusion patterns, and determine the possible occlusion type based on the degree of similarity.

5. A photovoltaic distribution network fault diagnosis system, based on the photovoltaic distribution network fault diagnosis method according to claim 4, characterized in that: The fault diagnosis system of the photovoltaic distribution network comprises: Photovoltaic array dynamic partitioning module, used to collect multi-dimensional real-time data of photovoltaic array, adaptively partition photovoltaic array, and perform multi-level classification of photovoltaic modules based on the partitioning results; The dynamic mapping module is used to extract multi-dimensional features of the photovoltaic array, build a multi-dimensional tensor model of photovoltaic module performance characteristics based on the multi-dimensional feature extraction results, extract feature subspaces, and perform performance feature mapping on the photovoltaic array; Fault diagnosis dynamic model module, which is used to perform intelligent diagnosis of local shading of photovoltaic modules based on performance feature mapping results and generate dynamic heat maps of shading impact; The fault location and repair module is used to accurately locate faults in photovoltaic arrays based on the dynamic thermal map of shading effects.

Images