A device anomaly detection method and a photovoltaic power generation device detection system

Abstract

The present application relates to photovoltaic power generation fault identification technical field, specifically be a kind of equipment abnormality detection method and photovoltaic power generation equipment detection system, in the present application, each node power offset ratio in path is convoluted by graph convolutional neural network, difference is conducted adjacent conduction in combination with the upstream and downstream connection order of node, so that the structure that there is sustained power growth trend in path has the continuity expression ability across node, through random forest in sudden jump synchronization judgment, with node as granularity, construct multi-channel label classification structure, through high-frequency channel category combination and change amplitude sorting, complete synchronous high-variable unit screening, so that the sudden jump synchronization judgment breaks through the mode of traditional three-channel single-variable difference threshold, through the incremental aggregation of offset trend between numbering structure, superimposed recovery period and the offset rate statistics of reference shading curve, improve the fault separability, response granularity and type description clarity in complex scene.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation fault identification technology, and in particular to a method for detecting equipment malfunctions and a photovoltaic power generation equipment detection system. Background Technology

[0002] The field of photovoltaic power generation fault identification technology aims to identify and locate abnormal power generation units in a timely manner by analyzing the electrical parameters of the operating status of various equipment in a photovoltaic system. This improves the reliability of system operation, power generation efficiency, and operation and maintenance response speed, and enables early detection and accurate location of typical faults such as component attenuation, shading, electrical faults, and connection abnormalities.

[0003] A photovoltaic power generation equipment detection system aims to automatically analyze the operating status of photovoltaic power generation equipment and identify abnormal units, identify possible faults in components or subarrays, including shading, attenuation, connection failure, etc., and output clear location information. The purpose is to improve system power generation efficiency, reduce maintenance costs, and ensure operational safety. Through quantitative efficiency evaluation or comparative analysis, it provides actionable diagnostic results to support operation and maintenance decisions.

[0004] Existing anomaly identification methods focus on data from a single node or component, neglecting the connectivity between nodes within the topology and the path trends of electrical parameter changes. They fail to establish power disturbance transmission chains between multiple nodes, resulting in a lack of identification of path-level anomaly diffusion characteristics. They also cannot assess the joint emergence of multi-channel jumps within a time window, leading to misjudgments and missed detections. Furthermore, the shading determination process relies on power decline characteristics or efficiency evaluation indicators, lacking modeling of the shading offset caused by the angle between the support orientation angle and the solar incidence angle. This makes it difficult to achieve geometric reconstruction of regional shading causes, thus limiting the ability to structurally classify shading types. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for detecting equipment malfunctions and a photovoltaic power generation equipment detection system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a photovoltaic power generation equipment testing system comprising:

[0007] Electrical topology construction module: Based on the established photovoltaic module numbering table, combiner box connection records and inverter wiring files, it performs number matching, power difference calculation, voltage difference judgment, constructs directed edge connection pairs and edge weight sequences, and generates an electrical directed connection graph;

[0008] Anomaly response aggregation module: Based on the power mutation trend of nodes in the electrical directed connection graph, it performs jump point synchronization detection, adjacent path identification and synchronization concentration screening to obtain the combination of mutation starting points with high distribution frequency and establish an anomaly propagation starting point candidate set;

[0009] Anomaly path identification module: Based on the candidate set of anomaly propagation starting points, a graph convolutional neural network is used to extract the ratio of power offset values ​​of nodes in the path, count continuously increasing paths, determine the path with the largest total offset and form a hop number path mapping, and generate a power offset propagation chain group.

[0010] The perturbation triggering judgment module: Based on the power offset propagation chain group, it extracts multi-channel jump values ​​of voltage, current and temperature channel data, uses random forest to judge the combination of synchronous jump nodes of multi-channel in the same period, filters high value points of index change rate, and establishes a multi-channel abnormal jump identification map.

[0011] The shading morphology deconstruction module: Based on the multi-channel abnormal jump identification map, it extracts the number of cycles that the power sequence of each node takes to recover to the previous state, calls the power change reference curve of the preset shading type, compares the point-by-point offset value with the recovery process curve, calculates the shading offset ratio by combining the bracket orientation angle and the solar incidence angle, and establishes a regional distribution map of shading categories.

[0012] The node power offset ratio is calculated by comparing the power value of each node on the path with the power value of its upstream node.

[0013] As a further aspect of the present invention, the electrical topology construction module includes:

[0014] Number verification submodule: Based on the established photovoltaic module number table, combiner box connection record and inverter wiring file, it performs item-by-item comparison and matching of module number with combiner box record, and filters out mismatched numbers. It performs item-by-item cross-comparison of module number with inverter file and marks the corresponding position of number. It calculates the difference of power output value between module and combiner box and generates difference matrix. It calculates the difference of power output value between combiner box and inverter and records the difference variation range. It measures the numerical difference between module output voltage and inverter input voltage and confirms voltage offset range to obtain electrical parameter difference mapping matrix.

[0015] Connection generation submodule: Based on the electrical parameter difference mapping matrix, compare the power difference between the module and the combiner box with the set interval, and eliminate connection combinations with abnormal differences. Compare the voltage difference between the combiner box and the inverter with the boundary of the permissible interval, and select connection pairs with voltage differences that meet the conditions. Generate triplet numbering combinations of photovoltaic modules, combiner boxes and inverter nodes, and construct connection edges with consistent directions. Extract the power difference, voltage difference and number position sequence, and generate the numerical set corresponding to the edge to obtain the edge connection attribute set.

[0016] The graph construction submodule constructs directed chains from component nodes to inverter nodes based on the edge connection attribute set, connects the node number paths, labels edge connection weights in order, associates them with the corresponding node numbers in the graph, organizes the outgoing and incoming edges of all nodes, and distinguishes and labels the directionality and node type to generate an electrical directed connection graph.

[0017] As a further aspect of the present invention, the anomaly response aggregation module includes:

[0018] Power mutation identification submodule: Based on the electrical directed connection graph, extract the power output value of each node at each time moment, construct a time series power array, calculate the adjacent difference of the node power output value, locate the position of power jump increase and drop, perform cross-retrieval of the difference of the jump time of multiple nodes, and mark the set of synchronous nodes with a change time difference less than a set window value to obtain the set of synchronous mutation nodes.

[0019] Path aggregation and extraction submodule: Based on the set of synchronous mutation nodes, perform path tracing for each synchronous node in the electrical directed connection graph, extract the path sequence from the start point to the end point, perform position index statistics of each jump node number in the path sequence, record the jump difference and connection level between the numbers, perform statistical counting of the start point numbers in all paths, and mark the path combination with the highest repetition frequency to obtain a centralized mutation path mapping table.

[0020] Starting point set generation submodule: Based on the centralized mutation path mapping table, sort the total frequency of occurrence of the starting numbers of all paths, generate a mapping set of numbers and corresponding frequencies, filter the top numbers after frequency sorting, extract the number combinations with the highest frequency, construct a candidate node number set and output it in the original order, and establish a candidate set of abnormal propagation starting points.

[0021] As a further aspect of the present invention, the abnormal path identification module includes:

[0022] Offset Extraction Submodule: Based on the candidate set of abnormal propagation starting points, a graph convolutional neural network is used to extract the sequential numbering of all nodes within the starting point numbering path, and to read the power data of each node within the path node by node. The difference between the power values ​​of the preceding and following nodes is calculated and the ratio is recorded. A list of node power offset value ratios is constructed and continuously increasing sequences are selected. A set of paths that meet the monotonicity requirement is extracted to generate a power increasing path sequence.

[0023] Path construction submodule: Based on the power increasing path sequence, construct the joint identifier of the node and its offset ratio in each path within the path number, sort the paths from high to low according to the cumulative offset value, record the corresponding path number, extract the set of paths with the largest offset value and perform hop count statistics, generate a path sequence index and hop count matching table, and generate a hop count path mapping set;

[0024] Chain group generation submodule: Based on the hop count path mapping set, the node numbers in each path are connected in sequence to form a directed node chain group. The chain group is constructed to build a consistent connection structure between nodes, generating a path topology set with increasing power as the connection condition. The chain group set is then organized and numbered, and a power offset propagation chain group is generated.

[0025] As a further aspect of the present invention, the graph convolutional neural network is configured according to the formula:

[0026]

[0027] in: Represents graph convolution. Layer nodes The power offset characteristics, Represents a non-linear activation function. Represents nodes Adjacent nodes , Represents a node The power-law adjustment value of the degree. Represents a node The power-law adjustment value of the degree. Represents graph convolution. The weight matrix of the layer, Represents graph convolution. Layer nodes The power ratio characteristic input, Represents a node Power fluctuation factor;

[0028] Execution process: First, historical power data of each node in the path is collected from the photovoltaic equipment nodes. A graph structure model is constructed based on the topological relationships between nodes, and each node is extracted sequentially. of Input simultaneously, then calculate. Through nodes The power values ​​along the path are normalized using the standard deviation, which is used to measure the power stability of the nodes. Then, the nodes are statistically analyzed. The path coverage factor is calculated by comparing the number of times a path appears in all power-increasing paths with the total number of paths. Then, targeting the current target node Connected path segments, calculate node pairs The derivative of the power difference with respect to the time difference yields the rate of change of power at the edge. Furthermore, the frequency of path segments occurring during anomaly propagation is statistically analyzed using a historical fault path database, and the path stability factor is obtained after normalization. Then extract the nodes. and connectivity and And substitute the empirical adjustment coefficient. Normalization is performed to control the bias of the influence of highly connected nodes during propagation, and finally the node is calculated. exist Power offset features in the layer Candidate path sequences that satisfy the monotonicity rule are selected.

[0029] As a further aspect of the present invention, the perturbation triggering determination module includes:

[0030] Sudden Jump Extraction Submodule: Based on the power offset propagation chain group, extract voltage, current and temperature channel data of each node in each chain and align them by time, compare the difference between each channel value at the same time point of the same node and the value at adjacent time points, mark the time points that are greater than the threshold, count the channel type and sudden jump amplitude data, and generate a multi-channel sudden jump dataset;

[0031] Synchronization Judgment Submodule: Based on the multi-channel jump dataset, random forest is used to count the types of jump channels in each time period and compare whether the three types of channels have common mutations. The node numbers that jump at the same time are filtered, the channel jump amplitude is extracted and the change rate per unit time is calculated. Nodes with change rates lower than the median value are filtered out, the numbers are retained and a set is generated to generate a set of synchronous high-change nodes.

[0032] The graph generation submodule generates connections between node pairs based on the synchronous high-variability node set, extracts edge connections between node pairs and labels them with channel names and change rate values, organizes connection edges with strong channel type consistency to form a subgraph, constructs a multi-node abrupt jump synchronization relationship graph structure, and establishes a multi-channel abnormal jump identification graph.

[0033] As a further aspect of the present invention, the random forest is defined according to the formula:

[0034]

[0035] in: This represents the total number of decision trees constructed in the random forest. Indicates the first The classification output function for each tree;

[0036] Execution process: First, multi-channel jump data from the photovoltaic equipment is collected, and then divided into basic input feature vectors according to time windows. This includes the jump amplitude, occurrence time, channel type code, and node number for each channel during a sudden jump. Next, extended feature parameters are calculated, and the change in the number of jumps per unit time is extracted to construct a frequency variation index. The proportion of synchronous jump nodes within the three-hop range of a jump node is statistically analyzed to generate a neighborhood collaboration factor. The jump energy ratio is obtained by integrating the jump power curve and dividing it by the total energy of the node period. The density factor is obtained by statistically analyzing the density of similar sudden jump events within the current time period. Then, the aforementioned extended features are multiplied by the adjustment weight coefficients respectively. and Perform linear combinations to form enhanced feature inputs, and then input them separately into the forest. The decision trees are used to make the judgment, and each tree is assigned a weight based on its normalized classification accuracy in the independent validation set. The output result is based on the weighted average of all trees' votes. It determines whether there is synchronous jump behavior of the three types of channels within a time period. If so, it retains the corresponding node number and then filters out nodes with a jump ratio higher than the median value within a unit time period, and finally establishes a set of synchronous high-change points.

[0037] As a further embodiment of the present invention, the perturbation triggering determination module is specifically an identification output structure for calibrating multi-channel jump behavior. The jump identification map includes a synchronous jump channel index set, a jump amplitude threshold distribution layer, and a jump node periodic grouping label. The synchronous high-change node is specifically a combination of voltage nodes, current nodes, and temperature channel data nodes whose jump amplitude exceeds a preset change rate threshold. The power offset propagation chain group specifically refers to a set of channel sequences that form a jump response path.

[0038] As a further aspect of the present invention, the occlusion morphology deconstruction module includes:

[0039] Occlusion response extraction submodule: Based on the multi-channel abnormal jump identification map, the number of all nodes in the map is read one by one and the path sequence is indexed. The power recovery time series of the corresponding nodes is extracted and the recovery segments are divided according to the time interval. The point-by-point difference between the recovery period of each segment and the occlusion reference curve is calculated and the relative offset rate is recorded. The period length of each node in the path is counted and the path is classified and numbered. The number set is aggregated according to the path label to generate the occlusion period response set.

[0040] The relative offset rate is obtained by subtracting the reference curve value from the actual power value at each time point, dividing by the reference value, and converting it into a percentage.

[0041] Angle offset calculation submodule: Based on the shading period response set, perform mapping and extraction of the support number corresponding to each path number and obtain the orientation angle parameter, extract the solar incidence angle sequence of the time period corresponding to the node jump time and match the path number, calculate the angle difference between the orientation angle and the incident angle of each path support and convert it into an angle ratio value, generate a list of path number and offset angle ratio combination, and generate a shading offset ratio table.

[0042] The region structure generation submodule extracts the coordinates of the nodes corresponding to each path number based on the occlusion offset ratio table and maps them to a two-dimensional space. It sets the hierarchical range according to the offset ratio and divides the nodes into corresponding level intervals. It aggregates the coordinates of nodes of the same level to form region blocks and records the region boundary numbers. It organizes the numbering structure of each region block and merges spatially adjacent blocks to establish an occlusion category region distribution map.

[0043] A method for detecting equipment anomalies, the method being executed based on the aforementioned photovoltaic power generation equipment detection system, includes the following steps:

[0044] S1: Based on the photovoltaic module number table, combiner box connection record and inverter wiring file, extract the correspondence between the node number and the wiring, match the module number and the wiring number, calculate the power value difference between number pairs, determine the degree of difference of the corresponding voltage value, establish a set of directed edges and edge weight sequence with number pairs as nodes, and obtain the electrical connection map;

[0045] S2: Based on the electrical connection map, identify the power change sequence of all nodes in the observation period, filter the jump points in the power sequence with an interval of no more than two sampling periods and an increase or decrease exceeding a set difference, identify the adjacent path nodes of each jump point, count the frequency of the same node in multiple paths in the jump path, filter the node combination with a frequency greater than a specified value, and obtain the abnormal propagation starting point combination set.

[0046] S3: Based on the set of abnormal propagation starting points, a graph convolutional neural network is used to construct the numbering order pairs of each pair of upstream and downstream nodes in the path, calculate the ratio of the power value of each node in the path to the power value of its upstream node, form a ratio matrix indexed by numbering pairs, and perform weighted superposition and normalization of the ratio matrix along the connection direction under the whole graph structure. Extract all numbering combinations in the connection path where the ratio is continuously increasing, calculate the cumulative sum of the ratios within the combination, and mark the number jump values ​​to obtain the offset propagation path mapping group.

[0047] S4: Based on the offset propagation path mapping group, a random forest is used to extract the continuous value sequence of voltage, current and temperature of each node in the same period within the path. The abrupt change amplitude values ​​of the three types of sequences in adjacent periods are calculated respectively. A multi-dimensional vector set of the three types of abrupt change amplitude is established. The multi-dimensional vectors of all nodes in the same path are aggregated according to the path structure. The number combination of the peak points of the three types of abrupt change amplitude in multiple paths is identified. The frequency of occurrence of each number combination and its corresponding maximum amplitude are recorded to obtain the multi-channel synchronous jump node set.

[0048] S5: Based on the multi-channel synchronous jump node set, extract the number of cycles that the power sequence of each node takes to recover to the previous state, call the power change reference curve of the preset shading type, compare the point-by-point offset value with the recovery process curve, combine the orientation angle of the component support where the numbered node is located with the solar incidence angle of the time period, calculate the ratio of the total offset value of each node to the total offset value of the reference curve, generate a matching table of number and shading type, and obtain the shading pattern distribution map.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0050] 1. In this invention, a graph convolutional neural network is used to perform convolution processing on the power offset ratio of each node in the path, and the difference is propagated by adjacency in combination with the upstream and downstream connection order of the nodes, so that the structure with a continuous power growth trend in the path has the ability to express the continuity across nodes, and the path recognition has improved the perception range of the global relationship of the graph structure.

[0051] 2. In this invention, a multi-channel label classification structure is constructed at the node level in the sudden jump synchronization judgment by using random forest. The high-frequency channel category combination and change amplitude sorting are used to complete the screening of high-variable synchronization units, so that the sudden jump synchronization judgment breaks through the traditional three-channel single variable difference threshold mode.

[0052] 3. In this invention, by increasing the aggregation of the offset trends between numbered structures, the path propagation direction forms a sequence-driven hop number mapping. The offset rate statistics of the recovery period and the reference occlusion curve are superimposed. The conversion ratio of the angle between the orientation angle and the incident angle is used as the occlusion morphology calculation factor, so that local occlusion events have the characteristics of mapping to regional offset groups, which improves the fault separability, response granularity and type characterization clarity in complex scenarios. Attached Figure Description

[0053] Figure 1 This is a system flowchart of the present invention;

[0054] Figure 2 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0056] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0057] Please see Figure 1 The present invention provides a technical solution: a photovoltaic power generation equipment testing system comprising:

[0058] Electrical topology construction module: Based on the established photovoltaic module numbering table, combiner box connection records and inverter wiring files, it performs number matching, power difference calculation, voltage difference judgment, constructs directed edge connection pairs and edge weight sequences, and generates an electrical directed connection graph;

[0059] Anomaly Response Aggregation Module: Based on the power mutation trend of nodes in the electrical directed connection graph, it performs jump point synchronization detection, adjacent path identification and synchronization concentration screening to obtain combinations of mutation starting points with high distribution frequency and establish a candidate set of anomaly propagation starting points;

[0060] Anomaly path identification module: Based on the candidate set of anomaly propagation starting points, a graph convolutional neural network is used to extract the ratio of power offset values ​​of nodes in the path, count continuously increasing paths, determine the path with the largest total offset and form a hop number path mapping, and generate a power offset propagation chain group.

[0061] The perturbation triggering judgment module: Based on the power offset propagation chain group, it extracts multi-channel jump values ​​of voltage, current and temperature channel data, uses random forest to judge the combination of synchronous jump nodes of multi-channel in the same period, filters high value points of index change rate, and establishes a multi-channel abnormal jump identification map.

[0062] The shading morphology deconstruction module: Based on the multi-channel abnormal jump identification map, it extracts the number of cycles that the power sequence of each node takes to recover to the previous state, calls the power change reference curve of the preset shading type, compares the point-by-point offset value with the recovery process curve, calculates the shading offset ratio by combining the bracket orientation angle and the solar incidence angle, and establishes a regional distribution map of shading categories.

[0063] The node power offset ratio is calculated by comparing the power value of each node on the path with the power value of its upstream node.

[0064] The electrical topology construction module includes:

[0065] Number verification submodule: Based on the established photovoltaic module number table, combiner box connection record and inverter wiring file, it performs item-by-item comparison and matching of module number with combiner box record, and filters out mismatched numbers. It performs item-by-item cross-comparison of module number with inverter file and marks the corresponding position of number. It calculates the difference of power output value between module and combiner box and generates difference matrix. It calculates the difference of power output value between combiner box and inverter and records the difference variation range. It measures the numerical difference between module output voltage and inverter input voltage and confirms voltage offset range to obtain electrical parameter difference mapping matrix.

[0066] Connection generation submodule: Based on the electrical parameter difference mapping matrix, it compares the power difference between the module and the combiner box with the set interval and eliminates connection combinations with abnormal differences. It compares the voltage difference between the combiner box and the inverter with the boundary of the permissible interval and selects connection pairs with voltage differences that meet the conditions. It generates triplet numbering combinations of photovoltaic modules, combiner boxes and inverter nodes, and constructs connection edges with consistent directions. It extracts the power difference, voltage difference and number position sequence, and generates the numerical set corresponding to the edge to obtain the edge connection attribute set.

[0067] The graph construction submodule constructs directed chains from component nodes to inverter nodes based on the edge connection attribute set, connects the node number paths, labels the edge connection weights in order, associates them with the corresponding node numbers in the graph, organizes the outgoing and incoming edges of all nodes, and distinguishes and labels the directionality and node type to generate an electrical directed connection graph.

[0068] The number verification submodule, based on the established photovoltaic module number table, combiner box connection records, and inverter wiring files, employs a bidirectional hash mapping method to map the photovoltaic module numbers one-to-one with the number fields in the combiner box records. It uses key-value pairs (key: module number, value: combiner box number) in the hash table for initial index matching, and then uses the combiner box number as the reverse key to build a reverse table structure for bidirectional verification and comparison, filtering out unmatched numbers. Finally, it uses a Boolean intersection filtering method to filter for number consistency between the module numbers and the access numbers in the inverter files, generating a Boolean matrix according to the index position, calculating the intersection Boolean matrix, and marking the corresponding numbers for all matching index positions. For wiring positions, a numerical difference matrix generation method is used to subtract the component output power value from the corresponding combiner box power input value point by point and construct a difference array. The difference array is then filled into a two-dimensional matrix according to the index dimension to generate a power difference matrix. The same operation is used to construct the difference between the combiner box output power value and the inverter input power value. All elements of the matrix are traversed and the difference variation range of the corresponding numbered pairs is recorded. The voltage vector difference comparison method is called, using the component terminal voltage measurement sequence and the inverter input voltage sequence as the input array. A one-to-one sequence subtraction operation is performed on each node number position to output the voltage offset difference vector and count the maximum and minimum values ​​to obtain the voltage offset range and generate an electrical parameter difference mapping matrix.

[0069] The connection generation submodule, based on the electrical parameter difference mapping matrix, uses a partitioned interval comparison method. It calls a set power tolerance interval array to perform element-level range comparison on the power difference matrix between the module and the combiner box. Matrix cells exceeding the tolerance boundary are marked as invalid connections and the corresponding connection pairs are removed. A boundary condition logic judgment method is used to perform a double boundary logic comparison operation on the voltage difference sequence between the combiner box and the inverter. The upper and lower limits are the preset permissible voltage offset values ​​of 8.5 and 13.2, respectively. Logical masking is performed on connection number combinations that do not meet the boundary conditions. A ternary combination generation function module is used to use the photovoltaic module number, combiner box number, and inverter number as ternary number one, number two, and number three, respectively. The numbering structure is spliced ​​in tuple format to form connection number ternary groups that conform to the direction label. A sequence extraction function group is used to extract the power difference, voltage difference, and physical location index number between the module and the inverter from the above ternary connection edges in sequence. The three indicators are generated into floating-point value combinations to form an edge attribute index vector group, resulting in the edge connection attribute set.

[0070] The graph construction submodule, based on the edge connection attribute set, adopts a directed graph structure construction method. Using numbered triples as the path basis, it sets the starting node as the component number and the ending node as the inverter number. It defines numbered paths in the graph structure according to the connection direction, calls the numbered mapping chain generation operation, establishes an array index sequence of node connection order in the path, and constructs numbered paths in sequence. A weighted mapping labeling strategy is used to label the weight values ​​in the edge connection attributes onto the connecting edges according to the edge connection sequence, and records the corresponding node numbers for graph indexing. A node connection table organization logic is used, organizing the outgoing and incoming node numbers of all nodes into two list structures to generate a node direction structure mapping table. Finally, it traverses by node number, sets the directional marker value to input as 0 and output as 1, and concatenates it with the node type codes 0, 1, and 2 to generate an electrical directed connection graph.

[0071] The exception response aggregation module includes:

[0072] Power mutation identification submodule: Based on the electrical directed connection graph, extract the power output value of each node at each time moment, construct a time series power array, calculate the adjacent difference of the node power output value, locate the position of power jump increase and drop, perform cross-retrieval of the difference of the jump time of multiple nodes, and mark the set of synchronous nodes with a change time difference less than a set window value to obtain the set of synchronous mutation nodes.

[0073] Path aggregation and extraction submodule: Based on the set of synchronous mutation nodes, it performs path tracing for each synchronous node in the electrical directed connection graph, extracts the path sequence from the start point to the end point, performs position index statistics of each hopping node number in the path sequence, records the hop difference and connection level between the numbers, performs statistical counting of the start point numbers in all paths, and marks the path combination with the highest repetition frequency to obtain a centralized mutation path mapping table.

[0074] Starting point set generation submodule: Based on the centralized mutation path mapping table, sort the total frequency of occurrence of the starting numbers of all paths, generate a mapping set of numbers and corresponding frequencies, filter the top numbers after frequency sorting, extract the number combinations with the highest frequency, construct a candidate node number set and output it in the original order, and establish a candidate set of anomaly propagation starting points.

[0075] Power mutation identification submodule: Based on the electrical directed connection graph, the local range mutation detection algorithm is used to extract the power output value of each node at each time step according to the sampling time sequence. A two-dimensional time series matrix is ​​constructed with the node number as the main index and the power value of each sampling point as the horizontal data. The power values ​​are subtracted at adjacent sampling times, and the absolute value of all results is taken. The positions of elements with a difference greater than a set threshold of 9.7 are selected and recorded as the jump point index set. The jump times of different nodes in the set are combined and cross-referenced one by one. The synchronous time window matching algorithm is used, and the synchronous judgment time difference threshold is set to 3 sampling periods. The jump time difference between nodes is calculated, and the number combination with a difference less than or equal to 3 is selected. The node group that mutates at the same time is selected as the synchronous mutation node set element. Finally, the synchronous mutation node set is obtained by aggregating the number combinations.

[0076] The path aggregation and extraction submodule, based on the set of synchronous mutation nodes, uses a graph path backtracking algorithm. Starting from each number in the synchronous mutation node, it traces all reachable paths level by level along the connection direction of the electrical directed connection graph. The maximum path tracing depth is set to 12 layers. The numbers of each layer of connection are recorded in sequence as a path number sequence. The number sequence difference is calculated according to the appearance position of each pair of jump node numbers in the path. All number sequence differences are uniformly converted into positive numbers to represent jump difference. The nodes in the path are assigned values ​​of 0, 1, and 2 according to the three categories of component, combiner box, and inverter, respectively. The connection level sequence of the corresponding number is recorded. The number frequency statistics are performed on the starting number of each path. The frequency values ​​are compared and sorted according to the starting number number table. The starting number combination with the highest frequency is used as the key identifier for path aggregation. The corresponding number path combination structure is recorded, and a centralized mutation path mapping table is generated.

[0077] Starting point set generation submodule: Based on the centralized mutation path mapping table, the module uses a number frequency ranking algorithm to extract the frequency value of the starting node numbers in all paths, counts the total number of times each number appears in all paths, sorts them from high to low frequency, and sets a selection threshold of the top 15% of the number groups by frequency. The module summarizes the position numbers of the top-ranked node number combinations in the sorting results, recombines them according to the number arrangement order in the path mapping table, constructs a number set in a continuous output format, generates number pairs for all number combinations, forms a non-repeating candidate number combination table, and finally obtains the candidate set of anomaly propagation starting points.

[0078] The abnormal path identification module includes:

[0079] Offset Extraction Submodule: Based on the candidate set of anomaly propagation starting points, a graph convolutional neural network is used to extract the sequential numbering of all nodes within the starting point path, and to read the power data of each node within the path node by node. The difference between the power values ​​of the preceding and following nodes is calculated and the ratio is recorded. A list of node power offset ratios is constructed and continuously increasing sequences are selected. A set of paths that meet the monotonicity requirement is extracted, and a power increasing path sequence is generated.

[0080] The path construction submodule is based on the power-increasing path sequence. It constructs a joint identifier for each node in each path within the path number and its offset ratio. It sorts the paths from high to low according to the cumulative offset value and records the corresponding path number. It extracts the set of paths with the largest offset value and performs hop count statistics. It generates a path sequence index and hop count matching table and generates a hop count path mapping set.

[0081] Chain group generation submodule: Based on the hop count path mapping set, the nodes in each path are connected in order to form a directed node chain group. The chain group is constructed to build a consistent connection structure between nodes, generating a path topology set with increasing power as the connection condition. The chain group set is then organized and numbered, and a power offset propagation chain group is generated.

[0082] Offset Extraction Submodule: Based on the candidate set of anomaly propagation starting points, a graph convolutional neural network is used to construct a graph structure with path numbers as node indices and the weights of the connecting edges between nodes as the input features of the graph convolution. The starting number in the path number sequence is called as the first node, and the node number extraction operation is performed sequentially on all paths in the graph structure starting from the starting number. Each number is regarded as a row in the network input feature matrix, and the feature dimension is set as the power value. The convolution operation stride is 1. The adjacency matrix is ​​constructed using a Boolean structure in the edge connection attribute matrix where the existence of a connection relationship is 1 and the absence of a connection relationship is 0. Graph convolution propagation is performed on all path numbers, and the difference between the power value of the current node and the power value of the upstream node is extracted at each node. All differences are constructed into a one-dimensional difference vector according to the node order. The ratio operation is performed on the power difference of any three consecutive nodes in the difference vector. All ratios are recorded and the increasing ratio sequence is selected. A set of path numbers that satisfy the positive growth relationship between any two ratios in the sequence is established. The set of path sequences whose number order does not reverse is extracted, generating a path number structure with a continuous and consistent power offset direction, and generating a power increasing path sequence.

[0083] The path construction submodule, based on the power-increasing path sequence, employs a path incremental accumulation algorithm to sum the power ratios of all nodes in each path number sequence node by node and construct a list of cumulative path offset values. The path numbers and total offset values ​​are combined into a two-dimensional index set. The index set is sorted in descending order of offset values, and the path number corresponding to each sorted position is recorded. The path number groups with the highest cumulative offset values ​​are extracted as high-offset path structures. For each number sequence in the path number structure, the number of hops between the first and last node numbers is calculated, and the hop value is recorded as a path span index. All path number indices and their corresponding hop counts form a key-value pair, establishing a path extension structure with a one-to-one correspondence between path numbers and hop count indices, and generating a hop count path mapping set.

[0084] The chain group generation submodule, based on the hop count path mapping set, adopts a node chain topology generation method. It sequentially connects the node numbers in each path number into directed number pairs, and then concatenates all directed number pairs into a complete path chain structure. The chain connection conditions are set as follows: the power value of the upstream node is less than the power value of the downstream node, the node numbers are continuously incremented, and each node is connected only once. For all path chain sets that meet the connection conditions, it constructs the directed edge structure between nodes and assigns three structural attributes to each directed connection in the graph structure: node position difference, path number index, and cumulative offset value. It completes the structural registration and numbering of each chain, and finally outputs all numbered chain groups in the path number order, generating a set of numbered chain paths with unidirectional consistent power offset direction, and generating a power offset propagation chain group.

[0085] Graph convolutional neural networks, according to the formula:

[0086]

[0087] in: Represents graph convolution. Layer nodes The power offset characteristics, Represents a non-linear activation function. Represents nodes Adjacent nodes , Represents a node The power-law adjustment value of the degree. Represents a node The power-law adjustment value of the degree. Represents graph convolution. The weight matrix of the layer, Represents graph convolution. Layer nodes The power ratio characteristic input, Represents a node Power fluctuation factor;

[0088] Execution process: First, historical power data of each node in the path is collected from the photovoltaic equipment nodes. A graph structure model is constructed based on the topological relationships between nodes, and each node is extracted sequentially. of Input simultaneously, then calculate. Through nodes The power values ​​along the path are normalized using the standard deviation, which is used to measure the power stability of the nodes. Then, the nodes are statistically analyzed. The path coverage factor is calculated by comparing the number of times a path appears in all power-increasing paths with the total number of paths. Then, targeting the current target node Connected path segments, calculate node pairs The derivative of the power difference with respect to the time difference yields the rate of change of power at the edge. Furthermore, the frequency of path segments occurring during anomaly propagation is statistically analyzed using a historical fault path database, and the path stability factor is obtained after normalization. Then extract the nodes. and connectivity and And substitute the empirical adjustment coefficient. Normalization is performed to control the bias of the influence of highly connected nodes during propagation, and finally the node is calculated. exist Power offset features in the layer Candidate path sequences that satisfy the monotonicity rule are selected.

[0089] The perturbation trigger determination module includes:

[0090] Sudden Jump Extraction Submodule: Based on the power offset propagation chain group, it extracts voltage, current and temperature channel data of each node in each chain and aligns them by time. It compares the difference between each channel value at the same time point of the same node and the value at adjacent time points, marks the time points that are greater than the threshold, and counts the channel type and sudden jump amplitude data to generate a multi-channel sudden jump dataset.

[0091] Synchronization Judgment Submodule: Based on the multi-channel jump dataset, random forest is used to count the types of jump channels in each time period and compare whether the three types of channels have common mutations. The node numbers that jump at the same time are filtered, the channel jump amplitude is extracted and the change rate per unit time is calculated. Nodes with change rates lower than the median value are filtered out, the numbers are retained and a set is generated to generate a set of synchronous high-change nodes.

[0092] The graph generation submodule generates a graph based on a set of synchronous high-variable nodes, generates a connection between node pairs based on the channel mutation type, extracts the edge connection of node pairs and labels the channel name and change rate value, organizes the connection edges with strong channel type consistency to form a subgraph, constructs a multi-node jump synchronization relationship graph structure, and establishes a multi-channel abnormal jump identification graph.

[0093] The jump extraction submodule, based on the power offset propagation chain, employs a channel timing difference extraction method. It extracts the original sequence data of the three channels (voltage, current, and temperature) for each node in each chain according to their numerical order. The data for each channel is aligned by node number, and a time axis structure is established with the sampling period as the horizontal axis. The data length for each channel is set to a fixed value of 120, and the sampling frequency is 5 seconds. For each node, the difference between the three channel values ​​at the same time point and the values ​​of the two adjacent sampling points in the channel is calculated, forming the difference between the current time point and the previous value, and the current time... Two difference vectors, one for the point and one for the subsequent value, are used to store the three types of channel differences into a matrix according to the node number. Each value in the matrix is ​​compared with the channel jump threshold. The voltage threshold is set to 3.4, the current threshold to 0.85, and the temperature rise threshold to 1.3. The time points that exceed the threshold are marked, and the field group consisting of the marked channel type, power value, and time index position is stored in the multi-channel difference identifier table. All jump records for each node number are statistically analyzed, and a multi-dimensional array consisting of channel type, jump amplitude, and jump position is output to generate a multi-channel jump dataset.

[0094] Synchronization Judgment Submodule: Based on a multi-channel jump dataset, a random forest is used to construct a sample set of node channel jump situations for each time period. Each sample contains channel number, jump amplitude, jump time index, node number, and channel type label. The training features are set as channel type and amplitude value. A model is constructed using 100 subtrees with a maximum tree depth of 6. Boolean judgment is performed on whether the three channel types of each node are simultaneously marked as jumps at each sampling time point. All node numbers that jump simultaneously are recorded and the jump channel type is marked. The value corresponding to the jump amplitude is extracted, and the change ratio is calculated by dividing the jump amplitude by the time interval between two sampling points. The ratio values ​​of the three types of channels are stored in the channel ratio vector table. The median value of the ratio is calculated according to the channel type. Records with a channel ratio of each node number that is less than the corresponding median value are set as invalid and the numbers are removed. Only the node numbers that meet the condition that all three channels jump and the change rate is higher than the median value are retained. Finally, a set of synchronous high-variable nodes is generated.

[0095] The graph generation submodule, based on a synchronous high-variability node set, employs a channel type mapping group construction method. It sequentially extracts number pairs from the node set, generating connection structures between nodes pair by pair. Connection generation conditions are set as follows: there is a directed connection between the number pairs within the chain group structure; the channel jump occurrence time index interval does not exceed two sampling periods. For number pairs meeting these conditions, a channel type intersection extraction operation is performed. The average jump change rate of each channel type in the intersection is calculated. A consistency judgment threshold is set to a jump ratio deviation of no more than 15%. All node connection structures meeting the conditions are used to form an edge set, and edge attribute fields are constructed using channel names and average change rates. Subgraphs are established based on connection directions, with all edges in the subgraph exhibiting channel type consistency stronger than the threshold. Finally, the subgraph number index, channel type group, and change rate field are organized into a structural graph form, generating a multi-channel abnormal jump identification graph.

[0096] Random forest, according to the formula:

[0097]

[0098] in: This represents the total number of decision trees constructed in the random forest. Indicates the first The classification output function for each tree;

[0099] Execution process: First, multi-channel jump data from the photovoltaic equipment is collected, and then divided into basic input feature vectors according to time windows. This includes the jump amplitude, occurrence time, channel type code, and node number for each channel during a sudden jump. Next, extended feature parameters are calculated, and the change in the number of jumps per unit time is extracted to construct a frequency variation index. The proportion of synchronous jump nodes within the three-hop range of a jump node is statistically analyzed to generate a neighborhood collaboration factor. The jump energy ratio is obtained by integrating the jump power curve and dividing it by the total energy of the node period. The density factor is obtained by statistically analyzing the density of similar sudden jump events within the current time period. Then, the aforementioned extended features are multiplied by the adjustment weight coefficients respectively. and Perform linear combinations to form enhanced feature inputs, and then input them separately into the forest. The decision trees are used to make the judgment, and each tree is assigned a weight based on its normalized classification accuracy in the independent validation set. The output result is based on the weighted average of all trees' votes. It determines whether there is synchronous jump behavior of the three types of channels within a time period. If so, it retains the corresponding node number and then filters out nodes with a jump ratio higher than the median value within a unit time period, and finally establishes a set of synchronous high-change points.

[0100] The perturbation trigger determination module is specifically an identification output structure used to calibrate the sudden jump behavior of multiple channels. The sudden jump identification map includes a set of synchronous sudden jump channel indexes, a sudden jump amplitude threshold distribution layer, and a sudden jump node periodic grouping label. The synchronous high change node is specifically a combination of voltage nodes, current nodes, and temperature rise nodes whose sudden jump amplitude exceeds the preset change rate threshold. The power offset propagation chain group specifically refers to the set of channel sequences that form the sudden jump response path.

[0101] The occlusion shape deconstruction module includes:

[0102] Occlusion response extraction submodule: Based on the multi-channel abnormal jump identification map, the number of all nodes in the map is read one by one and the path sequence is indexed. The power recovery time series of the corresponding nodes is extracted and the recovery segments are divided according to the time interval. The point-by-point difference between the recovery period of each segment and the occlusion reference curve is calculated and the relative offset rate is recorded. The period length of each node in the path is counted and the path is classified and numbered. The number set is aggregated according to the path label to generate the occlusion period response set.

[0103] The relative offset rate is obtained by subtracting the reference curve value from the actual power value at each time point, dividing by the reference value, and converting it into a percentage.

[0104] Angle offset calculation submodule: Based on the shading period response set, it performs mapping and extraction of the support number corresponding to each path number and obtains the orientation angle parameter, extracts the solar incidence angle sequence corresponding to the node jump time and matches it with the path number, calculates the angle difference between the orientation angle and the incident angle of each path support and converts it into an angle ratio value, generates a list of path number and offset angle ratio combinations, and generates a shading offset ratio table.

[0105] Region structure generation submodule: Based on the occlusion offset ratio table, extract the location coordinates of the nodes corresponding to each path number and map them to a two-dimensional space. Set the hierarchical range according to the offset ratio and divide the nodes into the corresponding level intervals. Aggregate the coordinates of nodes of the same level to form region blocks and record the region boundary numbers. Organize the numbering structure of each region block and merge spatial neighboring blocks to establish an occlusion category region distribution map.

[0106] The occlusion response extraction submodule, based on a multi-channel abnormal jump identification map, employs a power recovery difference matching method to read each node number in the map structure one by one. A numbered path list is then created sequentially according to the path number index. Power recovery time series data is extracted for each node, with a sampling interval of 5 seconds. The recovery cycle start point is set as the first rising point after the jump point, and the end point is set as the last sampling point where the power fluctuation in the stable segment is no greater than 0.5. Power recovery segments are divided according to the above time period, and point-by-point difference calculations are performed between the data of each recovery segment and the preset occlusion reference curve. The actual power value corresponding to each time point is subtracted from the reference curve value, divided by the reference value, and converted into a percentage, recorded as the relative offset rate. All recovery cycle lengths in the path are stored as an array according to the node number. Node numbers within the path are grouped by recovery segment length, aggregated in numerical order, and a label field is added to each path. The resulting data structure set, consisting of each path number and the number list, is output, generating an occlusion cycle response set.

[0107] Angle offset calculation submodule: Based on the shading period response set, the angle calculation method is adopted. The support direction parameters are extracted for each support number corresponding to each number in the path number index set. The direction angle data is referenced with true north, and the angle unit is degrees, ranging from 0 to 360 degrees. The direction angle vector list is constructed in the order of the numbers. For the time period of each number jump time point, the solar incidence angle time series is extracted. The time series range is set to 30 minutes before and after the jump point, and sampling is performed once every 5 minutes. For each time point, the angle difference between the orientation angle and the solar incidence angle corresponding to all numbers in the path number is extracted. The difference is the difference between the absolute values ​​of the two angles. If it exceeds 180 degrees, 360 is subtracted from the difference. For each node, the angle offset ratio is constructed by dividing the difference by 90, and four decimal places are retained. A two-dimensional list with a one-to-one mapping between path number and offset ratio is established. The number index table and the offset ratio value comparison structure are output to generate the shading offset ratio table.

[0108] The regional structure generation submodule, based on the occlusion offset ratio table, employs a spatial hierarchical clustering method to obtain the geographical coordinates of each component of each numbered node in the path numbering structure. The coordinate format is two-dimensional Cartesian coordinates in meters. A list of location coordinates is constructed according to the numbering order. Four-level classification intervals are set for each offset ratio, with interval boundary values ​​set to 0.00 to 0.25, 0.26 to 0.50, 0.51 to 0.75, and 0.76 to 1.00. Each node is assigned to the corresponding level interval according to the comparison results between the node number and the corresponding ratio. Number aggregation is performed on the node coordinate set within the same interval, and the aggregated number list is recorded to construct a regional boundary number set. The regional boundary number is the boundary area identifier formed by the minimum and maximum numbers in each group of nodes. An index table is built for the numbering structure within all level regions. Spatially adjacent nodes are merged, with the merging condition being that the difference between the horizontal and vertical coordinates of two nodes does not exceed 1.8 meters. The merging result is output as a numbered block index, generating an occlusion category regional distribution map.

[0109] Please see Figure 2 A method for detecting equipment malfunctions, which is based on the aforementioned photovoltaic power generation equipment detection system, includes the following steps:

[0110] S1: Based on the photovoltaic module number table, combiner box connection record and inverter wiring file, extract the correspondence between the node number and the wiring, match the module number and the wiring number, calculate the power value difference between number pairs, determine the degree of difference of the corresponding voltage value, establish a set of directed edges and edge weight sequence with number pairs as nodes, and obtain the electrical connection map;

[0111] S2: Based on the electrical connection map, identify the power change sequence of all nodes within the observation period, filter the jump points in the power sequence with an interval of no more than two sampling periods and an increase or decrease exceeding a set difference, identify the adjacent path nodes of each jump point, count the frequency of the same node in multiple paths in the jump path, filter the node combination with a frequency greater than a specified value, and obtain the abnormal propagation starting point combination set.

[0112] S3: Based on the abnormal propagation starting point combination set, a graph convolutional neural network is used to construct the numbering order pairs of each pair of upstream and downstream nodes in the path, calculate the ratio of the power value of each node in the path to the power value of its upstream node, form a ratio matrix indexed by numbering pairs, and perform weighted superposition and normalization of the ratio matrix along the connection direction under the whole graph structure. Extract all numbering combinations in the connection path where the ratio is continuously increasing, calculate the cumulative sum of the ratios within the combination, and mark the number jump values ​​to obtain the offset propagation path mapping group;

[0113] S4: Based on the offset propagation path mapping group, a random forest is used to extract the continuous value sequence of voltage, current and temperature of each node in the same period within the path. The abrupt change amplitude values ​​of the three types of sequences in adjacent periods are calculated respectively. A multi-dimensional vector set of the three types of abrupt change amplitude is established. According to the path structure, the multi-dimensional vectors of all nodes in the same path are aggregated. The number combination of the peak points of the three types of abrupt change amplitude in multiple paths is identified. The frequency of occurrence of each number combination and its corresponding maximum amplitude are recorded to obtain the multi-channel synchronous jump node set.

[0114] S5: Based on the multi-channel synchronous jump node set, extract the number of cycles that the power sequence of each node takes to recover to the previous state, call the power change reference curve of the preset shading type, compare the point-by-point offset value with the recovery process curve, combine the orientation angle of the component support where the numbered node is located with the solar incidence angle of the time period, calculate the ratio of the total offset value of each node to the total offset value of the reference curve, generate a matching table of number and shading type, and obtain the shading pattern distribution map.

[0115] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A photovoltaic power generation equipment testing system, characterized in that: The system includes: Electrical topology construction module: Based on the established photovoltaic module numbering table, combiner box connection records and inverter wiring files, it performs number matching, power difference calculation, voltage difference judgment, constructs directed edge connection pairs and edge weight sequences, and generates an electrical directed connection graph; Anomaly response aggregation module: Based on the power mutation trend of nodes in the electrical directed connection graph, it performs jump point synchronization detection, adjacent path identification and synchronization concentration screening to obtain the combination of mutation starting points with high distribution frequency and establish an anomaly propagation starting point candidate set; Anomaly path identification module: Based on the candidate set of anomaly propagation starting points, a graph convolutional neural network is used to extract the ratio of power offset values ​​of nodes in the path, count continuously increasing paths, determine the path with the largest total offset and form a hop number path mapping, and generate a power offset propagation chain group. The perturbation triggering judgment module: Based on the power offset propagation chain group, it extracts multi-channel jump values ​​of voltage, current and temperature channel data, uses random forest to judge the combination of synchronous jump nodes of multi-channel in the same period, filters high value points of index change rate, and establishes a multi-channel abnormal jump identification map. The shading morphology deconstruction module: Based on the multi-channel abnormal jump identification map, it extracts the number of cycles that the power sequence of each node takes to recover to the previous state, calls the power change reference curve of the preset shading type, compares the point-by-point offset value with the recovery process curve, calculates the shading offset ratio by combining the bracket orientation angle and the solar incidence angle, and establishes a regional distribution map of shading categories. The node power offset ratio is obtained by calculating the ratio of the power value of each node on the path to the power value of its upstream node.

2. The photovoltaic power generation equipment testing system according to claim 1, characterized in that, The electrical topology construction module includes: Number verification submodule: Based on the established photovoltaic module number table, combiner box connection record and inverter wiring file, it performs item-by-item comparison and matching of module number with combiner box record, and filters out mismatched numbers. It performs item-by-item cross-comparison of module number with inverter file and marks the corresponding position of number. It calculates the difference of power output value between module and combiner box and generates difference matrix. It calculates the difference of power output value between combiner box and inverter and records the difference variation range. It measures the numerical difference between module output voltage and inverter input voltage and confirms voltage offset range to obtain electrical parameter difference mapping matrix. Connection generation submodule: Based on the electrical parameter difference mapping matrix, compare the power difference between the module and the combiner box with the set interval, and eliminate connection combinations with abnormal differences. Compare the voltage difference between the combiner box and the inverter with the boundary of the permissible interval, and select connection pairs with voltage differences that meet the conditions. Generate triplet numbering combinations of photovoltaic modules, combiner boxes and inverter nodes, and construct connection edges with consistent directions. Extract the power difference, voltage difference and number position sequence, and generate the numerical set corresponding to the edge to obtain the edge connection attribute set. The graph construction submodule constructs directed chains from component nodes to inverter nodes based on the edge connection attribute set, connects the node number paths, labels edge connection weights in order, associates them with the corresponding node numbers in the graph, organizes the outgoing and incoming edges of all nodes, and distinguishes and labels the directionality and node type to generate an electrical directed connection graph.

3. The photovoltaic power generation equipment testing system according to claim 1, characterized in that, The anomaly response aggregation module includes: Power mutation identification submodule: Based on the electrical directed connection graph, extract the power output value of each node at each time moment, construct a time series power array, calculate the adjacent difference of the node power output value, locate the position of power jump increase and drop, perform cross-retrieval of the difference of the jump time of multiple nodes, and mark the set of synchronous nodes with a change time difference less than a set window value to obtain the set of synchronous mutation nodes. Path aggregation and extraction submodule: Based on the set of synchronous mutation nodes, perform path tracing for each synchronous node in the electrical directed connection graph, extract the path sequence from the start point to the end point, perform position index statistics of each jump node number in the path sequence, record the jump difference and connection level between the numbers, perform statistical counting of the start point numbers in all paths, and mark the path combination with the highest repetition frequency to obtain a centralized mutation path mapping table. Starting point set generation submodule: Based on the centralized mutation path mapping table, sort the total frequency of occurrence of the starting numbers of all paths, generate a mapping set of numbers and corresponding frequencies, filter the top numbers after frequency sorting, extract the number combinations with the highest frequency, construct a candidate node number set and output it in the original order, and establish a candidate set of abnormal propagation starting points.

4. The photovoltaic power generation equipment testing system according to claim 1, characterized in that, The abnormal path identification module includes: Offset Extraction Submodule: Based on the candidate set of abnormal propagation starting points, a graph convolutional neural network is used to extract the sequential numbering of all nodes within the starting point numbering path, and to read the power data of each node within the path node by node. The difference between the power values ​​of the preceding and following nodes is calculated and the ratio is recorded. A list of node power offset value ratios is constructed and continuously increasing sequences are selected. A set of paths that meet the monotonicity requirement is extracted to generate a power increasing path sequence. Path construction submodule: Based on the power increasing path sequence, construct the joint identifier of the node and its offset ratio in each path within the path number, sort the paths from high to low according to the cumulative offset value, record the corresponding path number, extract the set of paths with the largest offset value and perform hop count statistics, generate a path sequence index and hop count matching table, and generate a hop count path mapping set; Chain group generation submodule: Based on the hop count path mapping set, the node numbers in each path are connected in sequence to form a directed node chain group. The chain group is constructed to build a consistent connection structure between nodes, generating a path topology set with increasing power as the connection condition. The chain group set is then organized and numbered, and a power offset propagation chain group is generated.

5. The photovoltaic power generation equipment testing system according to claim 4, characterized in that, The graph convolutional neural network is constructed according to the formula: ； in: Represents graph convolution. Layer nodes The power offset characteristics, Represents a non-linear activation function. Represents nodes Adjacent nodes , Represents a node The power-law adjustment value of the degree. Represents a node The power-law adjustment value of the degree. Represents graph convolution. The weight matrix of the layer, Represents graph convolution. Layer nodes The power ratio characteristic input, Represents a node Power fluctuation factor; Execution process: First, historical power data of each node in the path is collected from the photovoltaic equipment nodes. A graph structure model is constructed based on the topological relationships between nodes, and each node is extracted sequentially. of Input simultaneously, then calculate. Through nodes The power values ​​along the path are normalized using the standard deviation, which is used to measure the power stability of the nodes. Then, the nodes are statistically analyzed. The path coverage factor is calculated by comparing the number of times a path appears in all power-increasing paths with the total number of paths. Then, targeting the current target node Connected path segments, calculate node pairs The derivative of the power difference with respect to the time difference yields the rate of change of power at the edge. Furthermore, the frequency of path segments occurring during anomaly propagation is statistically analyzed using a historical fault path database, and the path stability factor is obtained after normalization. Then extract the nodes. and connectivity and And substitute the empirical adjustment coefficient. Normalization is performed to control the bias of the influence of highly connected nodes during propagation, and finally the node is calculated. exist Power offset features in the layer Candidate path sequences that satisfy the monotonicity rule are selected.

6. The photovoltaic power generation equipment testing system according to claim 1, characterized in that, The perturbation trigger determination module includes: Sudden Jump Extraction Submodule: Based on the power offset propagation chain group, extract voltage, current and temperature channel data of each node in each chain and align them by time, compare the difference between each channel value at the same time point of the same node and the value at adjacent time points, mark the time points that are greater than the threshold, count the channel type and sudden jump amplitude data, and generate a multi-channel sudden jump dataset; Synchronization Judgment Submodule: Based on the multi-channel jump dataset, random forest is used to count the types of jump channels in each time period and compare whether the three types of channels have common mutations. The node numbers that jump at the same time are filtered, the channel jump amplitude is extracted and the change rate per unit time is calculated. Nodes with change rates lower than the median value are filtered out, the numbers are retained and a set is generated to generate a set of synchronous high-change nodes. The graph generation submodule generates connections between node pairs based on the synchronous high-variability node set, extracts edge connections between node pairs and labels them with channel names and change rate values, organizes connection edges with strong channel type consistency to form a subgraph, constructs a multi-node abrupt jump synchronization relationship graph structure, and establishes a multi-channel abnormal jump identification graph.

7. The photovoltaic power generation equipment testing system according to claim 6, characterized in that, The random forest is defined according to the formula: ； in: This represents the total number of decision trees constructed in the random forest. Indicates the first The classification output function for each tree; Execution process: First, multi-channel jump data from the photovoltaic equipment is collected, and then divided into basic input feature vectors according to time windows. This includes the jump amplitude, occurrence time, channel type code, and node number for each channel during a sudden jump. Next, extended feature parameters are calculated, and the change in the number of jumps per unit time is extracted to construct a frequency variation index. The proportion of synchronous jump nodes within the three-hop range of a jump node is statistically analyzed to generate a neighborhood collaboration factor. The jump energy ratio is obtained by integrating the jump power curve and dividing it by the total energy of the node period. The density factor is obtained by statistically analyzing the density of similar sudden jump events within the current time period. Then, the aforementioned extended features are multiplied by the adjustment weight coefficients respectively. and Perform linear combinations to form enhanced feature inputs, and then input them separately into the forest. The decision trees are used to make the judgment, and each tree is assigned a weight based on its normalized classification accuracy in the independent validation set. The output result is based on the weighted average of all trees' votes. It determines whether there is synchronous jump behavior of the three types of channels within a time period. If so, it retains the corresponding node number and then filters out nodes with a jump ratio higher than the median value within a unit time period, and finally establishes a set of synchronous high-change points.

8. The photovoltaic power generation equipment testing system according to claim 6, characterized in that, The perturbation triggering determination module is specifically an identification output structure used to calibrate the sudden jump behavior of multiple channels. The sudden jump identification map includes a set of synchronous sudden jump channel indexes, a sudden jump amplitude threshold distribution layer, and a sudden jump node periodic grouping label. The synchronous high-change node is specifically a combination of voltage nodes, current nodes, and temperature channel data nodes whose sudden jump amplitude exceeds a preset change rate threshold. The power offset propagation chain group specifically refers to the set of channel sequences that form the sudden jump response path.

9. The photovoltaic power generation equipment testing system according to claim 1, characterized in that, The occlusion morphology deconstruction module includes: Occlusion response extraction submodule: Based on the multi-channel abnormal jump identification map, the number of all nodes in the map is read one by one and the path sequence is indexed. The power recovery time series of the corresponding nodes is extracted and the recovery segments are divided according to the time interval. The point-by-point difference between the recovery period of each segment and the occlusion reference curve is calculated and the relative offset rate is recorded. The period length of each node in the path is counted and the path is classified and numbered. The number set is aggregated according to the path label to generate the occlusion period response set. The relative offset rate is obtained by subtracting the reference curve value from the actual power value at each time point, dividing by the reference value, and converting it into a percentage. Angle offset calculation submodule: Based on the shading period response set, perform mapping and extraction of the support number corresponding to each path number and obtain the orientation angle parameter, extract the solar incidence angle sequence of the time period corresponding to the node jump time and match the path number, calculate the angle difference between the orientation angle and the incident angle of each path support and convert it into an angle ratio value, generate a list of path number and offset angle ratio combination, and generate a shading offset ratio table. The region structure generation submodule extracts the coordinates of the nodes corresponding to each path number based on the occlusion offset ratio table and maps them to a two-dimensional space. It sets the hierarchical range according to the offset ratio and divides the nodes into corresponding level intervals. It aggregates the coordinates of nodes of the same level to form region blocks and records the region boundary numbers. It organizes the numbering structure of each region block and merges spatially adjacent blocks to establish an occlusion category region distribution map.

10. A method for detecting equipment malfunctions, characterized in that, The photovoltaic power generation equipment testing system according to any one of claims 1-9 shall be executed. Includes the following steps: S1: Based on the photovoltaic module number table, combiner box connection record and inverter wiring file, extract the correspondence between the node number and the wiring, match the module number and the wiring number, calculate the power value difference between number pairs, determine the degree of difference of the corresponding voltage value, establish a set of directed edges and edge weight sequence with number pairs as nodes, and obtain the electrical connection map; S2: Based on the electrical connection map, identify the power change sequence of all nodes in the observation period, filter the jump points in the power sequence with an interval of no more than two sampling periods and an increase or decrease exceeding a set difference, identify the adjacent path nodes of each jump point, count the frequency of the same node in multiple paths in the jump path, filter the node combination with a frequency greater than a specified value, and obtain the abnormal propagation starting point combination set. S3: Based on the set of abnormal propagation starting points, a graph convolutional neural network is used to construct the numbering order pairs of each pair of upstream and downstream nodes in the path, calculate the ratio of the power value of each node in the path to the power value of its upstream node, form a ratio matrix indexed by numbering pairs, and perform weighted superposition and normalization of the ratio matrix along the connection direction under the whole graph structure. Extract all numbering combinations in the connection path where the ratio is continuously increasing, calculate the cumulative sum of the ratios within the combination, and mark the number jump values ​​to obtain the offset propagation path mapping group. S4: Based on the offset propagation path mapping group, a random forest is used to extract the continuous value sequence of voltage, current and temperature of each node in the same period within the path. The abrupt change amplitude values ​​of the three types of sequences in adjacent periods are calculated respectively. A multi-dimensional vector set of the three types of abrupt change amplitude is established. The multi-dimensional vectors of all nodes in the same path are aggregated according to the path structure. The number combination of the peak points of the three types of abrupt change amplitude in multiple paths is identified. The frequency of occurrence of each number combination and its corresponding maximum amplitude are recorded to obtain the multi-channel synchronous jump node set. S5: Based on the multi-channel synchronous jump node set, extract the number of cycles that the power sequence of each node takes to recover to the previous state, call the power change reference curve of the preset shading type, compare the point-by-point offset value with the recovery process curve, combine the orientation angle of the component support where the numbered node is located with the solar incidence angle of the time period, calculate the ratio of the total offset value of each node to the total offset value of the reference curve, generate a matching table of number and shading type, and obtain the shading pattern distribution map.

Images