Photovoltaic monitoring system anomaly test data generation method and device, equipment and medium

Abstract

Embodiments of the present application relate to the field of photovoltaic technology, and disclose a photovoltaic monitoring system abnormal test data generation method, device, equipment and medium, the method comprises the following steps: constructing a GAN model, constructing a fusion framework of a knowledge graph and a probabilistic graph model, including constructing an ontology model in the field of photovoltaic, and establishing a probabilistic graph model according to the constraint relationship in the ontology model, iteratively training the GAN model to convergence, and then outputting a preset number of evaluation test samples of a predetermined photovoltaic fault type, inputting the preset number of evaluation test samples into the probabilistic graph model, and outputting a multidimensional influence probability corresponding to each evaluation test sample, obtaining a business impact degree corresponding to the evaluation test sample according to the multidimensional influence probability, and obtaining test data by screening the evaluation test sample according to the business impact degree. The test data generated by the embodiments of the present application can be widely applied to the test and verification of data cleaning, fault tolerance processing, fault detection and other functions of the photovoltaic monitoring system.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the field of photovoltaic technology, and particularly relates to a photovoltaic monitoring system abnormal test data generation method and system. BACKGROUND

[0002] The data monitoring system of a photovoltaic power station needs to process a large amount of environmental parameters, equipment operation parameters and geographic position information and the like data every day. When an equipment abnormality occurs in the actual operation process of the photovoltaic power station, if the data monitoring system fails to identify the abnormal data in time and process it, the data monitoring system will make a wrong judgment, affect the operation and maintenance efficiency of the power station, and cause economic loss.

[0003] The existing data processing capacity test of the photovoltaic monitoring system mainly relies on limited abnormal scenes generated in a real environment, or artificially constructs abnormal data or randomly generates abnormal data based on simple rules to form test data. The data in the abnormal scenes or the artificially generated test data are input into the photovoltaic monitoring system for processing, and the data processing capacity of the photovoltaic monitoring system is evaluated according to the processing result. The test data generated by relying on limited abnormal scenes generated in a real environment has high authenticity and reliability, but has the problems of limitation and incompleteness. The test data generated by artificially constructing or randomly generating based on simple rules can provide diversified abnormal scenes, but lacks authenticity, and the data amount of the test data generated by this method is not large enough, which cannot comprehensively evaluate the data processing capacity of the photovoltaic monitoring system. Therefore, it is urgent to generate test data with comprehensiveness and authenticity to meet the requirement of evaluating the data processing capacity of the photovoltaic monitoring system. SUMMARY

[0004] In view of the above problems, the embodiment of the present application provides a photovoltaic monitoring system abnormal test data generation method, device, equipment and medium, which is used to solve the problem that the test data for testing the photovoltaic monitoring system is small in quantity and not comprehensive in the prior art.

[0005] According to one aspect of the embodiment of the present application, a photovoltaic monitoring system abnormal test data generation method is provided, which comprises:

[0006] The GAN model is constructed, including constructing a generator network and a discriminator network. The generator network comprises a physical constraint module, a multi-modal encoding-decoding structure and a time sequence generation module. The discriminator network comprises a physical rule filtering layer, a multi-modal discrimination branch and a time sequence analysis module.

[0007] The fusion framework of the knowledge graph and the probabilistic graph model is constructed, including constructing an ontology model in the photovoltaic field to form a knowledge graph in the photovoltaic field, and establishing a probabilistic graph model according to the constraint relationship in the knowledge graph.

[0008] iteratively training the generator network and the discriminator network according to the photovoltaic fault type, the feature parameter corresponding to the photovoltaic fault type, and the key feature corresponding to the historical operation data of the photovoltaic monitoring system until convergence; the generator network outputting a preset number of evaluation test samples of a predetermined photovoltaic fault type after training is completed;

[0009] inputting the preset number of evaluation test samples into the probabilistic graph model, outputting a multidimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtaining a business influence degree corresponding to the evaluation test sample according to the multidimensional influence probability;

[0010] screening the evaluation test samples according to the business influence degree to obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

[0011] In an optional manner, the inputting the preset number of evaluation test samples into the probabilistic graph model, outputting a multidimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtaining a business influence degree corresponding to the evaluation test sample according to the multidimensional influence probability further include:

[0012] inputting the evaluation test sample into the probabilistic graph model to output an influence path corresponding to the evaluation test sample;

[0013] mapping the influence path into a domain entity relationship according to the constraint relationship of the ontology model, and outputting an influence path graph corresponding to the evaluation test sample according to the domain entity relationship;

[0014] outputting a test report of the evaluation test sample according to the influence path graph, the multidimensional influence probability, and the business influence degree, the test report including: a fault type, an influence path graph, a multidimensional influence probability, and a business influence degree.

[0015] In an optional manner, the fusion framework of the knowledge graph and the probabilistic graph model includes constructing an ontology model in the photovoltaic field to form a knowledge graph in the photovoltaic field, and establishing the probabilistic graph model according to the constraint relationship in the knowledge graph, including:

[0016] constructing an ontology model Q=(E, R, A) in the photovoltaic system, E being an entity set, R being a relationship set, and A being an attribute set, the constraint relationship of the knowledge graph including an entity set, a relationship set, and an attribute set;

[0017] establishing the probabilistic graph model according to the constraint relationship in the knowledge graph includes:

[0018] According to the relationship set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein a node of the Bayesian structure represents a system parameter X or a business index I, and an edge represents a conditional probability distribution derived from the relationship set R ;

[0019] The Bayesian structure is jointly mapped with the ontology model through a shared variable node to form a factor graph model , A set of factor nodes of relationship-attribute in the knowledge graph, A set of variable nodes composed of nodes in the Bayesian structure, and the factor graph model is the probabilistic graph model;

[0020] According to historical operation data of a photovoltaic system, parameters of a factor function corresponding to the factor nodes in the factor graph model are calculated through a maximum likelihood estimation method, and a trained probabilistic graph model is obtained.

[0021] In an optional manner, according to the relationship set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein a node of the Bayesian structure represents a system parameter X or a business index I, and an edge represents a conditional probability distribution derived from the relationship set R , comprising:

[0022] Mapping an attribute in the attribute set A to a system parameter node in the Bayesian structure;

[0023] Mapping a predefined business index to a business index node in the Bayesian structure;

[0024] According to the relationship set R, determining the existence and direction of a directed edge between the system parameter nodes and between the system parameter nodes and the business index nodes, and obtaining the Bayesian structure.

[0025] In an optional manner, the Bayesian structure is jointly mapped with the ontology model through a shared variable node to form a factor graph model , A set of factor nodes of relationship-attribute in the knowledge graph, A set of variable nodes composed of nodes in the Bayesian structure, and the factor graph model is the probabilistic graph model, comprising:

[0026] Mapping each conditional probability distribution in the Bayesian structure to a factor node in the factor graph model, and connecting the factor node to all variable nodes involved in the conditional probability distribution;

[0027] The relationships in the relationship set R are mapped to the factor nodes in the factor graph model, and the factor nodes are connected to the variable nodes corresponding to the entities participating in the relationship; wherein the variable nodes are composed of the system parameter nodes and the business indicator nodes, thus completing the construction of the factor graph model.

[0028] In one optional approach, a preset number of evaluation test samples are input into the probabilistic graphical model. Based on the mapping of constraints in the knowledge graph, a multi-dimensional impact probability corresponding to each evaluation test sample is output. The business impact degree corresponding to the evaluation test sample is obtained based on the multi-dimensional impact probability, including:

[0029] An improved belief propagation algorithm is executed in the factor graph model to obtain the multi-dimensional probability and influence path of the evaluation test sample on the business indicator node;

[0030] The improvements to the belief propagation algorithm include:

[0031] An attenuation factor is introduced into the message passing rules of the belief propagation algorithm for improvement. The improved message passing rules include:

[0032] An attenuation factor α is introduced into the message passing rule from the variable node to the factor node to improve the message passing rule:

[0033] ;

[0034] An attenuation factor is introduced into the message passing rule from the factor node to the variable node to improve the message passing rule:

[0035] ;

[0036] The attenuation factor is defined as: , d is the shortest path hop count between the two nodes in the factor graph model G, and β is an adjustable attenuation coefficient.

[0037] In an alternative approach, the improvement to the belief propagation algorithm further includes: introducing an anomaly feature weight function into the belief propagation algorithm. :

[0038] Based on the photovoltaic fault type e corresponding to the evaluation test sample, obtain the corresponding abnormal feature weights. ;

[0039] When calculating the messages passed from the factor node to the parameter node, the weights are determined according to the fault type e. Weight the messages:

[0040] .

[0041] In an alternative way, the improvement of the belief propagation algorithm further comprises: employing a hierarchical iteration strategy to execute the improved belief propagation algorithm to obtain the influence probability and the influence path, comprising:

[0042] inputting the device parameters in the evaluation test sample, performing belief propagation between nodes inside a single photovoltaic device, and outputting device layer messages;

[0043] inputting the device layer messages into the system layer, performing belief propagation between cross-device nodes, and outputting system layer messages;

[0044] inputting the system layer messages, globally passing to business indicator nodes, and outputting the influence path and the influence probability.

[0045] According to a second aspect of the embodiment of the present application, a photovoltaic monitoring system abnormal test data generation device is provided, comprising:

[0046] An adversarial generation network module is configured to build a GAN model, including building a generator network and a discriminator network, wherein the generator network comprises a physical constraint module, a multi-modal encoding-decoding structure, and a time sequence generation module; and the discriminator network comprises a physical rule filtering layer, a multi-modal discrimination branch, and a time sequence analysis module.

[0047] An evaluation framework building module is configured to build a fusion framework of a knowledge graph and a probabilistic graph model, including building an ontology model in the photovoltaic field to form a knowledge graph in the photovoltaic field, and establishing a probabilistic graph model according to the constraint relationship in the knowledge graph.

[0048] A test data generation module is configured to iteratively train the generator network and the discriminator network according to a photovoltaic fault type, feature parameters corresponding to the photovoltaic fault type, and key features corresponding to historical operation data of a photovoltaic monitoring system until convergence is achieved; and the trained generator network outputs a preset number of evaluation test samples of predetermined photovoltaic fault types.

[0049] A business influence evaluation module is configured to input a preset number of evaluation test samples into the probabilistic graph model, output a multi-dimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtain a business influence degree corresponding to the evaluation test sample according to the multi-dimensional influence probability.

[0050] A test data acquisition module is configured to filter the evaluation test samples according to the business influence degree to obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

[0051] According to a third aspect of the embodiments of the present application, a computer device is provided, comprising a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface complete communication with each other through the communication bus;

[0052] The memory is configured to store at least one executable instruction, and the executable instruction causes the processor to perform the operations of the photovoltaic monitoring system abnormal test data generation method according to any one of the first aspect.

[0053] According to another aspect of the embodiments of the present application, a storage medium is provided, comprising at least one executable instruction stored in the storage medium, and the executable instruction causes the computer device to perform the photovoltaic monitoring system abnormal test data generation method according to any one of the first aspect when running on the computer device.

[0054] The embodiments of the present application can quickly and automatically generate test data covering more than several types of photovoltaic fault types by constructing a generative adversarial network (GAN) model fused with physical constraints, multi-modal encoding-decoding structure and time sequence generation module, and combining a deep fusion framework of knowledge graph and probabilistic graph model. The test data is used for input into the photovoltaic monitoring system to detect the data processing capability of the photovoltaic monitoring system on the photovoltaic system, and solves the problems of incomplete coverage, low authenticity and poor consistency caused by relying on limited real abnormal scenes or artificial construction in traditional test data. By introducing a physical constraint mechanism, a time sequence correlation preservation strategy and multi-modal data fusion into a standard generative adversarial network (GAN) model, it is ensured that the evaluation test samples generated by the generator not only conform to the operation rules of the photovoltaic system, but also have time continuity and cross-modal consistency. Through joint reasoning of the knowledge graph and the probabilistic graph model, quantitative evaluation and path tracking of the business impact of the evaluation test samples are realized, which significantly improves the pertinence, practicality and evaluation efficiency of the test data. The present application can be widely applied to the testing and verification of data cleaning, fault tolerance processing, fault detection and other functions of the photovoltaic monitoring system, and has important engineering application value and promotion prospect.

[0055] The above description is only a summary of the technical solutions of the embodiments of the present application, in order to more clearly understand the technical means of the embodiments of the present application, the embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the embodiments of the present application more obvious and easy to understand, the specific embodiments of the present application are described below. BRIEF DESCRIPTION OF DRAWINGS

[0056] The accompanying drawings are only used to illustrate the embodiments, and are not considered as limiting the present application. Moreover, the same reference signs are used to represent the same parts throughout the drawings. In the drawings:

[0057] Figure 1 A flowchart of the photovoltaic monitoring system abnormal test data generation method provided by the embodiment 1 of the present application is shown;

[0058] Figure 2 A flowchart of the photovoltaic monitoring system abnormal test data generation method provided by the embodiment 1 of the present application is shown;

[0059] Figure 3 A structural schematic diagram of the photovoltaic monitoring system abnormal test data generation device provided by the embodiment 2 of the present application is shown;

[0060] Figure 4 A structural schematic diagram of the computer device provided by the embodiment 3 of the present application is shown. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Embodiment 1

[0062] Please refer to Figure 1 and Figure 2 , the embodiment 1 provides a photovoltaic monitoring system abnormal test data generation method, the method comprises:

[0063] S1: constructing a GAN model, including constructing a generator network and a discriminator network, the generator network comprises a physical constraint module, a multi-modal encoding-decoding structure and a time sequence generation module; the discriminator network comprises a physical rule filtering layer, a multi-modal discrimination branch and a time sequence analysis module.

[0064] The standard GAN model is improved through a physical constraint condition mechanism, including:

[0065] The generator network comprises a physical constraint module, the physical law of a photovoltaic system (such as the law of conservation of energy, the photoelectric conversion equation) is introduced into the generator of the standard GAN as a hard constraint, so as to ensure that the data generated by the generator conforms to the basic physical rules, specifically:

[0066] By defining a constraint function , S is a data sample vector (i.e. the generator output ), including multiple parameters of the photovoltaic system, such as irradiance, temperature, power and the like, for example: , is the irradiance , is the ambient temperature , is the component temperature , is the direct current power (W),

[0067] is the direct current voltage (V), is the direct current current (A), is the inverter efficiency. represents the ith physical constraint, for example, S is the power-related abnormal data, ,

[0068] is the photovoltaic array area (m²), is the component nominal efficiency, is the allowed error threshold (W), P is the total number of physical constraints. When , it means that the data sample S complies with the physical rule, and when , it means that the data sample S violates the physical rule.

[0069] It is not a fixed value, and needs to be calculated based on the historical normal data of the photovoltaic system, and the calculation method is: , wherein and are the mean and standard deviation of in the historical data, respectively.

[0070] A penalty term is introduced in the standard generator loss function, which ensures that the data generated by the generator complies with the abnormal characteristics while not violating the basic physical law. When the data sample S violates the physical law, , a positive penalty value will be generated; when the data sample S does not violate the physical law, the penalty value is 0. The penalty term is:

[0071] ;

[0072] is the penalty coefficient, The value of can be adjusted according to the scale and constraint strength of the specific photovoltaic system, and the value range is 0.1 to 10; p is the total number of physical constraint functions; is the generated data output by the generator; Z is the random noise vector of the input layer structure of the generator, with dimension k, C is the condition vector, with dimension m, containing the feature parameters of the fault type; is the generator network parameter.

[0073] The total loss function of the generator becomes: is the original generator loss function. According to the total loss function, the weight of the generator is updated, and after the weight is updated, the generator network The internal computing logic of the discriminator is slightly changed, so that the next time it is used to output a data sample S, for the same input noise Z and condition C, it will output a better data sample S. "Better" means that this new data sample S is more likely to deceive the discriminator than the old data sample S, and less likely to violate physical rules.

[0074] The data sample S output by the generator is input to the input layer of the discriminator, the intermediate layer of the discriminator processes the data sample S, and finally the output layer of the discriminator outputs a scalar [0,1] representing the probability score of the input layer inputting the data sample S. The scalar output by the discriminator is where

[0075] A physical rule filtering layer is added to the input layer of the discriminator to filter out sample data S that does not comply with physical laws. Specifically, a physical rule function is defined, and a filtering mechanism is set up:

[0076] ; The original discriminator output, i.e. through the physical rule filtering layer, data samples S that obviously violate physical rules are directly marked as "false" (output 0), while compliant data samples are allowed to enter the subsequent neural network discriminator .

[0077] The data processing logic of the discriminator is changed, represented as where g(S) is the rule function, and the physical rule filtering layer indirectly affects the gradient calculation and weight update of the subsequent discriminator backbone network by changing the distribution of valid input data, the weights of the discriminator network are updated, and the internal computing logic of the discriminator network is slightly changed, so that for input data S that can pass the physical filter, the discrimination ability of the discriminator is more accurate. "Accurate" means that this updated discriminator D is better than the old discriminator at distinguishing between generated data that "looks real but actually violates physical laws" and real data, because it receives more training signals from compliant data that need to be finely distinguished, rather than a large amount of obviously invalid data.

[0078] The standard GAN model is improved by a timing correlation maintenance mechanism, including:

[0079] The generator network includes a temporal generation module. A Temporal Generative Adversarial Network (TCGAN) is constructed using a Long Short-Term Memory (LSTM) network and the aforementioned GAN model. By introducing a time state transfer mechanism, the high-fidelity temporal correlation of the data samples S output by the generator is ensured. The time state transfer function is implemented through the forward computation process of the LSTM unit, and is as follows: , This represents the state at the current time step, which is determined by the hidden state. and cell state constitute, It is the state of the previous time step. It is the input data for the current time step, generated from the data of the previous time step. It is formed by concatenating the random noise vector Z and the conditional vector C, i.e. .

[0080] The specific calculation of the LSTM state transfer function H includes the input gate, forget gate, output gate, and cell state update:

[0081] Input Gate: ;

[0082] Forgotten Gate: ;

[0083] Output gate: ;

[0084] New cell state: ;

[0085] Final cell state: ;

[0086] Final hidden state: ;

[0087] It is the sigmoid activation function. This represents element-wise multiplication. and These are trainable parameter matrices and vectors. The forward computation process of the LSTM unit is existing technology and will not be described in detail here.

[0088] Input the state of the previous time step into the LSTM unit , Current input vector The new state at the current time step is generated through a series of calculations, including the input gate, forget gate, and output gate. Final hidden state It is sent to an output layer (such as a fully connected layer) of the generator, which will... Mapping as generator final output data point at time step t , where O is the output layer function, are output layer parameters; the output of the current step is part of the input of the next time step, while the state is passed to the LSTM unit of the next time step, and so on, until a complete time series of the required length {S_1, S_2,..., S_T} is generated:

[0089] The generator receives a random noise vector Z and a conditional vector C describing the anomaly, initializes the hidden state h_0 and the cell state c_0 as zero vectors; from t = 1, the initial input S_in_1 = [0; Z; C] (where 0 is the initialized data vector) is input into the LSTM network, the state Q_1 is calculated and the first data point S_1 is generated. Subsequently, S_in_2 = [S_1; Z; C] is input as the input of the next time step, and the LSTM unit is recursively called to generate the data sequence {S_1, S_2,..., S_T} in turn. This process ensures the continuity and long-range correlation of the output sequence in the time dimension through the successive transmission of the state Q_t.

[0090] On the basis of standard discriminator feature extraction, a time series analysis module is introduced, so that the discriminator can analyze the rationality of the sample data S from the perspective of the overall time series. The feature extraction logic of the discriminator changes, and the final output feature representation is . According to the loss L D calculated by the feature H, the weight of the discriminator is updated, After the weight is updated, the internal feature extraction logic of the discriminator network D changes slightly, so that the next time for the same input sequence S, it can extract a "better" new feature representation that can more effectively reveal the irrationality in the time dimension of the sequence, such as the duration of the anomaly, the high mutation frequency, or the recovery pattern that does not conform to the normal, so that the final discrimination result D(S) is more accurate.

[0091] The standard GAN model is improved through multi-modal data fusion, including:

[0092] On the basis of standard generator input processing, a multi-modal encoding-decoding structure is introduced, which maps numerical, categorical, and spatio-temporal heterogeneous data to a unified hidden space for fusion through an independent encoder, and then reconstructs it into coordinated multi-modal data through a decoder. The data fusion logic of the generator changes, and is represented as: where F is the multi-modal feature extracted from the historical operation data of the photovoltaic system.

[0093] Specifically, the numerical data encoder takes in a numerical feature vector , with dimension (e.g. irradiance, temperature, voltage, current, etc. continuous parameters), and outputs a hidden representation , where hhidden is the unified dimension of the hidden space; the categorical data encoder takes in a categorical feature vector , with dimension dcat (e.g. weather state, device status flag, communication status, etc.), and outputs a hidden representation , with the same dimension as the numerical hidden representation; the spatio-temporal data encoder takes in a spatio-temporal feature tensor (e.g. GPS trajectory, time-series image, spatial distribution map, etc.), and outputs a hidden representation ; the three types of hidden representations are summed up with weights according to their importance: ;

[0094] where , and the initial suggestion is .

[0095] The decoder in the generator takes in the fused hidden representation , and reconstructs it into the multi-modal data output S through deconvolution (Deconv) or FC layers.

[0096] The total loss function is calculated according to the fused features, and is represented as: , where L_num is the mean square error (MSE) loss for numerical data; L_cat is the cross-entropy (Cross-Entropy) loss for categorical data, and L_spatio is the combined loss of Huber loss and cosine similarity for spatio-temporal data. are the corresponding weight coefficients, which can be dynamically adjusted according to the specific fault type and test target: if the test focuses more on spatial anomalies (such as GPS drift), the weight can be appropriately increased; if the test focuses on state jumps (such as communication interruption), the weight can be increased, etc.

[0097] The weights of the generator are updated according to the total loss function , and the updated weights cause a slight change in the internal encoding-decoding logic of the generator network G, so that in the next forward propagation, for the same input noise Z and condition C, it will output a better data sample S, which is more logically consistent within and between modalities than the old data, for example, when generating a "GPS coordinate drift" anomaly, the azimuth and inclination data will change in a geometrically consistent manner, and these spatial changes are logically consistent with the numerical power decline and categorical communication interruption flag on the timeline.

[0098] A multimodal discriminant branch is introduced on top of the standard discriminator architecture: three parallel subnetworks. , The discriminator independently evaluates the authenticity of the input data S across each modality through parallel dedicated subnetworks and performs weighted fusion decisions. The discriminator architecture and loss calculation logic are altered, as shown below:

[0099] , Keep them consistent with the weight coefficients in the generator to ensure training consistency.

[0100] Obtain the multimodal fusion loss to update the discriminator weights. Weight After the update, the internal feature extraction and decision logic of the discriminator network D undergoes a slight change, resulting in a more "accurate" discriminant output for the same input data S in the next iteration. The new discriminant comprehensively and meticulously considers the authenticity of the data across all modal dimensions. For example, the updated discriminator can keenly identify "numerical sequence patterns that seem reasonable but are seriously inconsistent with the labeled weather category 'rainstorm'" or "GPS drift trajectories that are spatially continuous but whose associated string current data do not show the expected shadow occlusion changes," thereby significantly improving the ability to detect cross-modal inconsistencies.

[0101] Furthermore, regarding the setting of the weighting coefficients, the initial recommended value is... If the test focuses on numerical anomalies (such as sudden power changes), then increase the... If the test focuses on state transitions (such as communication interruptions), then improve... If the test focuses on spatial anomalies (such as GPS drift), then improve... .

[0102] The improvement to the construction of GAN models can be made by selecting one or more of the following mechanisms from the data characteristics of the evaluation test samples for photovoltaic fault types as needed: physical constraint mechanism, temporal correlation preservation mechanism, and multimodal data fusion.

[0103] S2: Constructing a fusion framework for knowledge graphs and probabilistic graphical models, including building an ontology model in the photovoltaic field to form a knowledge graph for the photovoltaic field, and establishing the probabilistic graphical model based on the constraint relationships in the ontology model, including:

[0104] S21: Construct an ontology model Q=(E, R, A) in a photovoltaic system, E is an entity set, R is a relation set, and A is an attribute set, the constraint relationship of the knowledge graph includes an entity set, a relation set, and an attribute set. The entity set E includes various physical entities in the photovoltaic system, such as E={inverter, photovoltaic module, irradiance sensor, temperature sensor, GPS module, string, combiner box...}; R is a relation set, which describes the structure, function and causal relationship between entities, including input-output relationship, control relationship, physical connection relationship, data dependency relationship, influence relationship, etc.; A is an attribute set, which is used to define the observable or calculable attributes of each entity, such as A={efficiency, temperature, voltage, current, power, coordinates, timestamp,...}. Through the ontology model, a structured photovoltaic field knowledge graph is formed, realizing the standardized representation of the composition, operation logic and abnormal propagation path of the photovoltaic system.

[0105] S22: Establishing the probabilistic graph model according to the constraint relationship in the knowledge graph includes:

[0106] S221: According to the relation set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein the nodes of the Bayesian structure represent system parameters X or business indicators I, and the edges represent conditional probability distributions derived from the relation set R .

[0107] The nodes in the Bayesian structure are divided into two categories: system parameter nodes X and business indicator nodes I. The system parameter nodes X mainly include environmental parameters and equipment operation parameters The business indicator nodes I are defined from the perspective of photovoltaic power plant operation, maintenance and business decision, and are used to quantify the overall health status, performance and economic value of the photovoltaic system. They are high-level performance indicators, mainly including fault detection accuracy, performance evaluation indicators, power generation prediction values, economic benefit indicators, etc.

[0108] The edges of the Bayesian structure represent conditional probability distributions derived from the relation set R The conditional probability distribution is learned by maximum likelihood estimation on historical normal operation data in the photovoltaic field, which ensures that it reflects the statistical dependence relationship between parameters and indicators in the actual system.

[0109] Further, the S221 includes:

[0110] S2211: Map the attributes in the attribute set A to the system parameter nodes X in the Bayesian structure.

[0111] S2212: Map the pre-defined business indicators to the business indicator nodes I in the Bayesian structure.

[0112] S2213: determining the existence and direction of the directed edges between the system parameter nodes and between the system parameter nodes and the business indicator nodes according to the relationship set R, obtaining the Bayesian structure.

[0113] S222: jointly mapping the Bayesian structure and the ontology model through the shared variable nodes to form a unified factor graph structure , wherein represents a set of factor nodes composed of entity-relation in the knowledge graph, is a set of variable nodes composed of nodes in the Bayesian structure; the factor graph realizes the deep fusion of domain knowledge and statistical model, and supports bidirectional information transmission and dynamic knowledge update.

[0114] S223: calculating the factor function parameters corresponding to the factor nodes in the factor graph model according to the historical operation data of the photovoltaic system through the maximum likelihood estimation method, and obtaining the trained probabilistic graph model.

[0115] S3: iteratively training the generator network and the discriminator network according to the photovoltaic fault type, the feature parameters corresponding to the photovoltaic fault type, and the key features corresponding to the historical operation data of the photovoltaic monitoring system until convergence; the trained generator network outputs a preset number of evaluation test samples of a predetermined photovoltaic fault type.

[0116] Further, S3 includes:

[0117] S31: collecting the historical operation data of the photovoltaic system, preprocessing the historical operation data and extracting the key features of the photovoltaic system, the key features including time sequence features, parameter correlation features and spatial distribution features.

[0118] The historical operation data includes: environmental parameters (such as irradiance, ambient temperature, component temperature, etc.), equipment operation parameters (such as direct current voltage / current, inverter output power, inverter efficiency, string current, etc.), geographic location information (such as GPS coordinates, azimuth, inclination, etc.), time characteristics (such as sampling time stamp, daily variation curve, seasonal variation characteristics, etc.). The historical operation data is collected from the historical database of the target photovoltaic monitoring system through a standardized data interface (such as ODBC, OPCUA, MQTT) for a continuous time period (such as 12 months). The collected historical operation data needs to be preprocessed, including processing missing values (using time series interpolation or regression filling based on associated parameters), smoothing noise (using moving average or wavelet transform denoising), and standardization / normalization to ensure the stability of subsequent feature extraction and model training. The cleaned historical data is analyzed to extract the key features corresponding to the historical data. The technical scheme for preprocessing the historical data of the photovoltaic system and extracting the key features is the prior art, which will not be described in detail.

[0119] The key features include: time series features, parameter correlation features, and spatial distribution features, etc.

[0120] The time series features can be used as a time dimension reference when the generator outputs the data sample S, ensuring that the generated data sample S has a reasonable variation rule in time. Time series feature extraction includes: extracting the daily cycle, seasonal cycle, etc. frequency components of irradiance and power output through Fourier transform (FFT); analyzing the inertia of the time series through autocorrelation function (ACF); calculating the statistics (such as mean, variance, skewness, kurtosis) within the sliding window to describe the local dynamics, etc. Time series features are the basis for simulating abnormal time patterns (such as sudden changes, drifts, oscillations).

[0121] The parameter correlation features can ensure the consistency of the physical laws between parameters when the generator outputs the data sample S, making the output data sample S more realistic. Parameter correlation feature extraction includes: calculating the correlation coefficient or fitting regression model of irradiance (Irr) and direct current power under different weather conditions; analyzing the heat transfer lag relationship model between ambient temperature and component temperature , etc. Parameter correlation features ensure that the generated evaluation data has logical consistency and avoids producing contradictory data that violates physical or operational logic.

[0122] The spatial distribution feature can enable the generator to consider the differences of devices at different positions when outputting the data sample S, so that the output data sample S is more representative. The spatial distribution feature extraction includes: calculating the physical distance between different groups of strings and inverters through geographic information system (GIS) coordinates; statistically analyzing the distribution (such as Gaussian distribution, skewness distribution) of current values of different regional groups of strings at the same time; and establishing a location-based irradiance difference model. These features are necessary for generating anomalies with spatial characteristics (such as local shadows, group string faults, and GPS drift).

[0123] S32: Collecting fault data of photovoltaics, dividing the fault data according to fault types, and defining abnormal feature parameters corresponding to the fault types according to parameters of key features in the fault data, the fault types including sensor abnormalities, device operation abnormalities, communication abnormalities, location information abnormalities, data consistency abnormalities, etc.

[0124] Continuously collecting fault data in the field of photovoltaics, dividing the fault data according to fault types, and defining abnormal feature parameters corresponding to the fault types according to parameters of key features in the fault data, creating a fault database according to fault types and corresponding abnormal feature parameters, the fault database being continuously updated, including more than 200 fault types and corresponding abnormal feature parameters, and directly obtaining a predetermined photovoltaic fault type and corresponding abnormal feature parameters from the fault database when needed.

[0125] For each divided fault type, let the fault type set be: , analyze the key features in the fault data corresponding to the fault type, and each fault type has a corresponding feature parameter set: The generator extracts quantifiable abnormal feature parameters from the feature parameter set that define the nature of the fault, forming a condition vector C corresponding to the fault type and inputting it into the generator. For example, for the fault type of irradiance mutation (sensor anomaly): the extracted abnormal feature parameters include the mutation amplitude (e.g., when the irradiance baseline is approximately 800 W / m² based on historical data and environmental conditions (e.g., weather, solar altitude angle) at the current moment, and the peak value after mutation is approximately 2400 W / m² ± 50 W / m², the mutation amplitude is 300% ± 5%), mutation duration (e.g., 15 min ± 2 min), rise time (e.g., 5 min), and occurrence time preference (e.g., UTC time 10:00-15:00); for the fault type of inverter efficiency... For equipment malfunctions (abnormal equipment operation), the extracted abnormal feature parameters include the direction of efficiency deviation (positive / negative), the degree of deviation (e.g., negative values ​​(e.g., -5% to -20%), or extremely high values ​​(e.g., 150% ± 5%)), and the frequency of abnormal fluctuations. For fault types such as GPS coordinate drift (abnormal location information), the extracted abnormal feature parameters include the drift speed (e.g., 0.05° / h to 0.5° / h), the drift direction (e.g., azimuth), and the maximum drift distance. For fault types such as communication interruption (abnormal communication), the extracted abnormal feature parameters include the interruption duration, data loss rate, and reconnection mode.

[0126] S33: The GAN model is iteratively trained until convergence based on the photovoltaic fault type, the abnormal feature parameters corresponding to the photovoltaic fault type, and the key features corresponding to the historical operating data.

[0127] First, the conditional vector C corresponding to the fault type and random noise Z are input into the generator, which outputs sample data S. Then, the key features corresponding to the sample data S and historical running data are input into the discriminator, which outputs corresponding scalars. The discriminator aims to maximize its ability to distinguish between real and generated data. The discriminator's loss function... as follows:

[0128] L_D(θ d )=E[log(D(S_real))]+E[log(1-D(S_fake))]=E_{S~p_data}[log(D(S;θ d ))]+E_{Z, C~p_z, p_c}[log(1-D(G(Z,C;θ g );θ d E_{S~p_data} represents the expectation of a sample S sampled from the true data distribution (p_data); Indicates the prior distribution from noise and conditional distribution Calculate the expectation of Z and C from the joint sampling. Represents a real historical data sample. Represents the data samples generated by the generator;

[0129] By using the backpropagation algorithm, only the discriminator's parameter θ is updated. d This increases D(S_real) (approaching 1) and decreases D(S_fake) (approaching 0), forcing the discriminator D to better learn the characteristics of normal historical data S_real and accurately identify abnormal data S_fake generated by the generator that does not conform to the normal pattern.

[0130] Another batch of Z and C is sampled from the random noise and fault condition distribution. The generator's goal is to minimize the probability that it will be detected by the discriminator, i.e., to generate anomalous data that can "fool" the current discriminator. The generator's loss function L_G is as follows:

[0131] L_G(θ g )=E_{Z, C~p_z, p_c}[log(1-D(G(Z, C; θ g );θ d E_{Z, C~p_z, p_c} represents the expectation of Z and C jointly sampled from the noise prior distribution (p_z) and the conditional distribution (p_c).

[0132] The backpropagation algorithm is used to update only the generator's parameters. This allows the generated abnormal data S_fake to increase the value of D(S_fake) (approaching 1); it forces the generator G to adjust its parameters so that the generated fault data S_fake retains the specified fault characteristics (controlled by C) while being as close as possible in statistical characteristics to the distribution of the "real" data (i.e. normal data) as the discriminator considers it, thereby generating highly realistic abnormal data that conforms to physical laws.

[0133] Repeat the above steps, alternating between updating the discriminator D and the generator G. Through this adversarial training, the performance of the generator and discriminator continuously competes to improve: the discriminator continuously improves its ability to distinguish between real and fake data, while the generator continuously optimizes the realism of the generated anomalous data. The training process converges when the model reaches Nash equilibrium, i.e., when the discriminator can no longer distinguish between real and generated data with a higher probability than random guessing (D(S_real)≈0.5, D(S_fake)≈0.5). At this point, the generator G is able to output highly realistic anomalous data corresponding to the fault type to be evaluated based on the given fault condition vector C.

[0134] The GAN model in S1 is trained based on the fault type and its corresponding abnormal feature parameters and key features. The generator training objective is to minimize the function. The generated data output by the generator can maximize the deception of the discriminator, and the training target of the discriminator is to maximize the function The ability of the discriminator to distinguish between real data and generated data is improved, and by alternately optimizing the two objective functions, the GAN model gradually reaches Nash Equilibrium, i.e., the GAN model satisfies the convergence condition, so that the generator can finally generate highly realistic evaluation test samples, which are highly realistic abnormal feature parameters corresponding to the fault type. Iterative training of the GAN model until convergence is a prior art, which will not be described here.

[0135] S34: The generator network that has completed training outputs a predetermined number of evaluation test samples of a predetermined photovoltaic fault type.

[0136] According to the test requirements, the tester can select one or more predetermined photovoltaic fault types from the fault type library, and set the condition vector C according to the abnormal feature parameters corresponding to the fault type, or directly set the condition vector C, input the random noise Z and the condition vector C into the generator, and the generator G outputs a predetermined number of evaluation test samples S that meet all parameter settings of the predetermined photovoltaic fault type.

[0137] This embodiment can obtain a predetermined number of evaluation test samples S corresponding to a predetermined photovoltaic fault type, avoiding the problem of poor test results caused by the lack of data for testing the existing photovoltaic monitoring system, and realizing the automatic and industrial production of data for testing. The pain points of traditional methods completely relying on scarce real fault data or inefficient artificial construction are solved. The generated data has strong diversity, high authenticity, and can be accurately controlled, greatly improving the coverage and depth of photovoltaic monitoring system testing.

[0138] S4: inputting a predetermined number of evaluation test samples into the probabilistic graph model, outputting a multi-dimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtaining a business impact degree corresponding to the evaluation test sample according to the multi-dimensional influence probability.

[0139] The evaluation test sample S is input into the factor graph model, and the abnormal feature parameters contained in the evaluation test sample are identified (for example, it is identified that the timestamp The irradiance parameter has a mutation), then the system parameter node corresponding to the abnormal parameter is found in the factor graph model (such as finding the node representing the irradiance ), and the observation value of the node is set to the value in the abnormal data (for example, at time, the is forced), which is equivalent to "injecting" an evidence on this specific node of the probabilistic graph model.

[0140] An improved belief propagation algorithm is started, which involves two types of core message calculations:

[0141] Messages from variable nodes to factor nodes: ;

[0142] Messages from factor nodes to variable nodes: .

[0143] The belief propagation algorithm starts from the node (Xirr) injected with evidence, propagates the uncertainty represented by this change along the conditional probability relationships defined by the edges in the factor graph to the factor nodes (F) and business indicator nodes (I) connected to it, and after the propagation calculation is completed, the belief propagation algorithm calculates the posterior marginal probability of each business indicator node state in the factor graph model with respect to the abnormal parameter. The posterior marginal probability directly quantifies the possibility of the abnormality causing the business indicator to deviate from the normal range, obtains the multi-dimensional influence probability, and then calculates the business impact degree corresponding to the evaluation test sample S according to the multi-dimensional influence probability.

[0144] S5: filtering the evaluation test sample according to the business impact degree to obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

[0145] The evaluation test sample S is filtered by the business impact degree, and the evaluation test sample S with high business impact degree is selected to form test data for testing the abnormal detection capability of the photovoltaic monitoring system, so that the test data has a significant impact on the key business indicators of the photovoltaic monitoring system, thereby effectively improving the pertinence and efficiency of the photovoltaic monitoring system test, ensuring that the test data not only has high authenticity and diversity, but also has clear business orientation and fault coverage depth, and the photovoltaic monitoring system can be systematically verified in terms of fault detection, performance evaluation, power generation prediction, and economic operation. The abnormal response and fault tolerance processing capability in multiple dimensions.

[0146] Further, the S222 includes:

[0147] S2221: mapping each conditional probability distribution in the Bayesian structure to a factor node in the factor graph model, and connecting the factor node to all variable nodes involved in the conditional probability distribution;

[0148] For example, the nodes of the Bayesian structure are irradiance , component temperature , direct current power , and the corresponding edges are: (irradiance affects component temperature), (Irradiance affects DC power) (Component temperature also affects DC power), so the conditional probability distribution is: (Probability distribution of component temperature given irradiance) (Probability distribution of DC power given irradiance and component temperature) (Prior probability distribution of irradiance), representing Simultaneously and The decision was made.

[0149] for Create a factor node: (Representing temperature function), for Create a factor node: (Representing power function), for Create a factor node: (Represents prior irradiance).

[0150] Connect each newly created factor node to all variable nodes involved in the conditional probability distribution: for each factor node... (from) ), involving and Therefore, With variable nodes and Connect them with edges respectively.

[0151] S2222: Map the relations in the relation set R to the factor nodes in the factor graph model, and connect the factor nodes to the variable nodes corresponding to the entities participating in the relations; wherein the variable nodes are composed of the system parameter nodes and the business indicator nodes, thus completing the construction of the factor graph model.

[0152] In this embodiment, knowledge graphs and probabilistic graphical models are combined. Knowledge graphs provide deterministic constraints, resolving the potential absurd reasoning that pure probabilistic models may exhibit in sparse data regions. Probabilistic graphical models offer the ability to quantify uncertainty and learn complex relationships from data, compensating for the often lack of numerical strength in knowledge graphs. This combination makes the evaluation model both reliable and flexible.

[0153] The fusion framework of the trained knowledge graph and the probabilistic graphical model becomes a "digital twin" evaluation system. When the fault data (evaluation test sample S) corresponding to the photovoltaic fault is injected, the reasoning algorithm can calculate how the fault data will affect the final business indicators (I), thereby realizing the quantitative mapping from "data anomaly" to "business impact" and obtaining the multi-dimensional impact probability corresponding to the fault data (evaluation test sample S).

[0154] Further, in S4, an improved belief propagation algorithm is performed in the factor graph model to obtain the multi-dimensional influence probability and influence path of the evaluation test sample on the business indicator node; the improvement of the belief propagation algorithm includes:

[0155] An attenuation factor is introduced in the message passing rule of the belief propagation algorithm for improvement, and the improved message passing rule includes:

[0156] An attenuation factor a is introduced in the message passing rule from the variable node to the factor node to improve the message passing rule:

[0157] ;

[0158] An attenuation factor is introduced in the message passing rule from the factor node to the variable node to improve the message passing rule:

[0159] ;

[0160] The attenuation factor is defined as: , d is the shortest path hop count between the two nodes in the factor graph model G, and β is an adjustable attenuation coefficient, with a recommended value interval of [0.1, 5.0] and an initial recommended value of β = 1.0.

[0161] By introducing the attenuation factor in the message passing rule of the belief propagation algorithm, the estimation deviation problem of the influence probability caused by unlimited message propagation or excessive diffusion in the traditional algorithm in the large-scale factor graph of the photovoltaic system is effectively solved. The embodiment can accurately quantify the attenuation characteristics of abnormal influence with network topology distance, significantly improve the accuracy and reliability of multi-dimensional influence probability and influence path calculation, and provide a solid theoretical basis and calculation foundation for subsequent objective evaluation of business influence degree.

[0162] Further, the improvement of the belief propagation algorithm further includes introducing an abnormal feature weight function ;

[0163] According to the photovoltaic fault type e corresponding to the evaluation test sample S, the corresponding abnormal feature weight w(e) is obtained, for example, sensor abnormality: when the irradiance suddenly changes to 300%, w(e) = 1.2, equipment operation abnormality (such as inverter efficiency abnormality): w(e) = 1.5, communication abnormality (such as communication interruption): w(e) = 0.8, position information abnormality (such as GPS drift): w(e) = 1.0, data consistency abnormality: w(e) = 1.1, etc.

[0164] In calculating the message passed by the factor node to the variable node, the message is weighted by the weight w(e) of the failure type e:

[0165] ;

[0166] e: photovoltaic failure type; is the photovoltaic failure type associated with the factor node f; is the photovoltaic failure type corresponding weight function.

[0167] By introducing the abnormal feature weight function w(e) closely related to the photovoltaic failure type in the belief propagation algorithm, the precise quantification and dynamic modulation of the differentiated influence intensity of different categories of abnormalities are realized. The embodiment effectively overcomes the limitations of traditional algorithms on various failures, significantly improves the discrimination and business fitting of multi-dimensional influence probability calculation, and makes the final evaluation result more objectively reflect the real damage level of a specific failure, providing a more reliable basis for precise screening of test data and system vulnerability analysis.

[0168] Further, the improvement of the belief propagation algorithm also includes: adopting a hierarchical iteration strategy to execute the improved belief propagation algorithm to obtain the influence probability and the influence path, including:

[0169] Inputting the device parameters in the evaluation test sample S, performing belief propagation between nodes within a single photovoltaic device, and outputting device layer messages;

[0170] Inputting the device layer messages into the system layer, performing belief propagation between cross-device nodes, and outputting system layer messages;

[0171] Inputting the system layer messages, globally delivering to business indicator nodes, and outputting the influence path and the influence probability.

[0172] By adopting a hierarchical iteration strategy to execute the improved belief propagation algorithm, the complex global influence propagation calculation is decomposed into multi-level fine processing of the device layer, the system layer, and the global business layer, effectively solving the problems of high computational complexity, slow convergence speed, and large path noise faced by direct global propagation in large-scale photovoltaic monitoring systems. The hierarchical mechanism significantly improves the calculation efficiency and accuracy of the influence probability and the influence path, and can clearly reveal the complete propagation chain of the abnormality from the local device parameters to the system-level performance indicators and finally to the global business impact, providing efficient and reliable algorithm support for precise evaluation of failure influence range and test data screening.

[0173] Further, based on the factor graph structure, an improved belief propagation algorithm is used to calculate the multi-dimensional influence probability and influence path of the test sample S on the business index I. The improved belief propagation algorithm introduces a decay factor to control the decay of the influence with the network distance on the basis of the standard belief propagation, and introduces an abnormal feature weight function w(e) to dynamically adjust the propagation weight according to different fault types e, so as to more accurately reflect the influence characteristics of different types of abnormalities. The propagation process adopts a hierarchical iteration strategy, and message passing is performed in the device layer, system layer and business layer in turn, and finally the influence probability value of each business index node and its propagation path atlas are output, providing a quantitative basis for subsequent business impact calculation and test report generation.

[0174] In an implementation manner, the possibility that the abnormality causes the business index to deviate from the normal range is directly quantified according to the posterior marginal probability, and a multi-dimensional influence probability is obtained, including:

[0175] The posterior marginal probability is obtained, and is respectively denoted as: fault detection abnormality probability , performance evaluation abnormality probability , power prediction abnormality probability , and economic benefit abnormality probability ;

[0176] An evaluation matrix is formed according to the fault sensitivity function, the performance deviation function, the prediction error accumulation function and the economic loss estimation model, including:

[0177] The fault sensitivity function is constructed , if the fault detection related node probability is too high, a fault detection influence function is called, inputting the evaluation test sample S and the fault detection algorithm set A, and outputting a fault detection influence degree quantization value FDI, including:

[0178] ;

[0179] represents the fault detection time, represents normal data, represents the weight coefficient of the algorithm , and satisfies , is the time required for the algorithm to detect a fault under the input evaluation test sample S, and is the time required for the algorithm to detect a fault under normal data ;

[0180] According to the key performance index set of the photovoltaic power station (Including performance ratio, capacity utilization, inverter efficiency, etc.), define the performance deviation function.

[0181] ,in This represents the calculation of the first [number] under the data S condition. Each performance metric value, This represents the weighting coefficient of the indicator, satisfying... ; Under the evaluation test sample S, the value of the i-th performance index, ; in normal data The value of the i-th performance metric;

[0182] Constructing the prediction error accumulation function If the power prediction is abnormal If the error exceeds a preset threshold, the prediction error accumulation function will be invoked. Perform the calculation:

[0183] Quantitative evaluation of the test sample S on the power generation prediction function F over the time interval The cumulative effect within; To enable the power generation prediction model to perform under abnormal data S input. The predicted value at any given time; : In normal data Under the input, the power generation prediction model is in Predicted value at time, The time frame for conducting the assessment.

[0184] An economic loss estimation model L_e(X) is established. If the probability of abnormal economic benefit P_eb exceeds a preset threshold, the economic loss estimation model L_e(X) is called for calculation. The model L_e(X) integrates the economic consequences brought about by the performance evaluation impact (PEI, calculated by D_p), the power prediction impact (EFI, calculated by E_f), and the fault detection impact (FDI, calculated by S_f), and maps them to specific economic loss values.

[0185] : No. Class due to performance deviation and prediction error The loss mapping function that is commonly caused, for example, This can indicate a decrease in system performance. Inaccurate power generation forecasts The combined result of the loss of power generation revenue; : No. Classified by fault detection delay or misjudgment The resulting loss mapping function, for example, Can represent the additional maintenance costs, fines or power loss caused by failure to be detected in time and accurately; : economic coefficient corresponding to the loss type respectively, k is the total number of economic loss types considered.

[0186] Barrier sensitivity function, performance deviation function, prediction error accumulation function and economic loss estimation model are constructed to form an evaluation matrix. The posterior marginal probability is input into the evaluation matrix, the abnormal probability of each dimension is calculated, if the value exceeds the preset threshold, the abnormal probability is taken as a trigger signal, and the original evaluation test sample S and normal benchmark data As input, the specific impact degree value of this dimension is calculated:

[0187] If Exceeds the threshold, the fault detection impact function is called, S and S0 are input, and the fault detection impact degree FDI is calculated and output.

[0188] If P_pe exceeds the threshold, the performance evaluation impact function is called, S and S0 are input, and the performance evaluation impact degree PEI is calculated and output.

[0189] If P_ef exceeds the threshold, the power prediction impact function is called, S and S0 are input, and the power prediction impact degree EFI is calculated and output.

[0190] If P_eb exceeds the threshold, the economic benefit impact function is called, S and S0 are input, and the economic benefit impact degree EBI is calculated and output, which depends on the calculation results of PEI and EFI.

[0191] Output fault detection impact probability P(FDI), performance evaluation impact probability P(PEI), power prediction impact probability P(EFI) and economic benefit impact probability P(EBI).

[0192] Further, the business impact degree of the evaluation test sample is calculated according to the multi-dimensional impact probability, including:

[0193] The multi-dimensional impact probability of the evaluation test sample is obtained, including: fault detection impact possibility P(FDI), performance evaluation impact possibility P(PEI), power prediction impact possibility P(EFI), and economic benefit impact possibility P(EBI);

[0194] The corresponding weight coefficient is set for the impact possibility corresponding to the impact possibility, and the business impact degree of the evaluation test sample is obtained according to the weight coefficient and the impact possibility:

[0195] , wherein ​are weight coefficients and satisfy .

[0196] are weight coefficients set in advance according to the operation strategy and test preference of the photovoltaic system, for example, if the current test pays more attention to the fault detection capability, then can be set to be higher. By affecting the probability value, it can be intuitively reflected that the abnormal parameters contained in the evaluation test sample have a "harm level" or "test value" on the photovoltaic monitoring as a whole.

[0197] By constructing an evaluation matrix that integrates multi-dimensional quantitative functions, and realizing accurate mapping and trigger calculation from fault data to specific business impact based on posterior marginal probability, and finally obtaining a comprehensive business impact degree through weighted fusion, a complete, objective and configurable abnormal data business value evaluation system is established. The embodiment effectively solves the problems of strong subjectivity, single dimension and difficulty in quantification in the traditional method of abnormal data impact evaluation, can accurately quantify the differentiated impact degree and probability of specific abnormalities on four key dimensions of fault detection, performance evaluation, power prediction and economic benefit, and supports dynamic adjustment of evaluation weights according to actual operation strategy and test target, so as to ensure that the generated test data not only has true diversity, but also has clear business pertinence and high test value, greatly improving the coverage depth and evaluation efficiency of photovoltaic monitoring system test.

[0198] In an implementation manner of the embodiment, the S4 further includes:

[0199] inputting the evaluation test sample into the probabilistic graph model, and outputting an impact path corresponding to the evaluation test sample.

[0200] mapping the impact path into a domain entity relationship according to the constraint relationship of the ontology model, and outputting an impact path graph corresponding to the evaluation test sample according to the domain entity relationship.

[0201] outputting a test report of the evaluation test sample according to the impact path graph, the multi-dimensional impact probability and the business impact degree, wherein the test report includes a fault type, an impact path graph, a multi-dimensional impact probability and a business impact degree.

[0202] Specifically, for example, when the probability graph model output shows that the impact path of the evaluation test sample is: irradiance sensor → string current → inverter DC input power → inverter efficiency → system total output power, according to the constraint relationship defined in the pre-constructed ontology model Q=(E, R, A) in the photovoltaic field, such as the inverter and string in the entity set E, the direct impact and the belonging in the relationship set R, and the current and efficiency in the attribute set A), the abstract impact path output by the above probability model is mapped to a specific and richly semantic entity relationship chain, and the system automatically generates a visual impact path diagram according to the relationship chain. The diagram clearly shows the complete chain of the anomaly starting from the source device (such as the irradiance sensor), passing through the intermediate components and parameters, and finally affecting the high-level business indicators (such as the power generation). The system automatically generates a structured test report based on the impact path diagram, the calculated multi-dimensional impact probability (i.e., P(FDI), P(PEI), P(EFI), P(EBI)), and the quantified business impact degree. The core content of the test report includes: fault type, impact path diagram, multi-dimensional impact probability, and business impact degree.

[0203] The embodiment of the present application realizes the transparent and explainable precise tracing from the bottom layer data anomaly to the top layer business impact by deeply fusing the output of the probability graph model with the field knowledge graph and automatically generating a detailed test report containing a visual impact path diagram and multi-dimensional quantitative indicators.

[0204] Meanwhile, the test report can be used as a feedback signal for the GAN model training process, providing a novel, objective and quantifiable external verification mechanism for evaluating the convergence degree of the generative adversarial network. This scheme effectively overcomes the limitations of traditional GAN training, which only relies on the fluctuations of the generator and discriminator loss functions for subjective judgment. By continuously analyzing the stability and rationality of the business impact degree, multi-dimensional impact probability and impact path in the test report, it can accurately judge from the business logic level whether the model has learned the real, effective and physically consistent anomaly pattern, thereby significantly improving the controllability of model training, the reliability of convergence results, and the quality and practical value of the finally generated data.

[0205] Embodiment 2

[0206] Based on the photovoltaic monitoring system anomaly test data generation method provided in the above embodiment 1, the present embodiment generates test data corresponding to the "irradiance mutation" specific fault type according to the photovoltaic monitoring system anomaly test data generation method.

[0207] The GAN model is constructed, the generator of the GAN model adopts a 5-layer LSTM-CNN hybrid architecture, the number of LSTM neurons in each layer is 128, and the number of CNN convolution kernels is 64, which is used to learn and generate data with high-fidelity time sequence characteristics. The discriminator adopts a 4-layer residual network (ResNet) structure as a time sequence analysis module, which is used to judge the rationality of the data from the overall sequence perspective.

[0208] Define the data sample vector For this embodiment, the constraint function is set as , wherein is the area of the photovoltaic array, is the nominal efficiency of the component, is set to 50W, and the penalty coefficient is 0.5.

[0209] The photovoltaic system ontology model Q=(E, R, A) is constructed, wherein the entity set E includes {irradiance sensor, photovoltaic string, inverter,...}; the relationship set R includes {directly affects, belongs to, input and output}; the attribute set A includes {irradiance, component temperature, direct current power,...}; the probability graph model is established according to the constraint relationship in the ontology model, specifically including mapping the attributes to system parameter nodes, determining the directed edges between the nodes according to the relationship set, and finally fusing into a factor graph model The factor function parameters are calculated by the maximum likelihood estimation method using the historical operation data, and the training of the probability graph model is completed.

[0210] The historical operation data of a 500KW photovoltaic power station from January to December (a total of 12 months) in a year is collected through a standardized interface, the sampling granularity is 5 minutes, the total amount of data is about 105120, and the irradiance time sequence data is extracted, including time domain analysis (calculating the intra-day change feature function ), frequency domain analysis (identifying periodic patterns through Fourier transform F[Irr(t)]), and conditional probability modeling (constructing P(Irr|W, t), W is a weather state vector), analysis shows that the average peak value of the irradiance of this data set is 842W / m², the standard deviation , and the short-time change rate is usually not more than 10% / 5min.

[0211] The fault type "irradiance mutation" and the abnormal feature parameters corresponding to this fault type are selected from the preset fault library, and the conditional vector C is constructed according to the feature parameters of the anomaly: mutation amplitude (Amplitude): 300%±5%, mutation duration (Duration): 15min±2min, mutation rise time (RiseTime): 5min.

[0212] 80% of the historical operation data as a training set, 20% as a validation set to train the GAN model: using the condition vector C and the key features of the historical data, the GAN model constructed in S1 is iteratively trained. The training parameters are set as: learning rate , batch size b=64, training period e=500. The loss function is the sum of the Wasserstein distance and the physical constraint penalty term. After 500 rounds of training, the discriminator accuracy is 92.7%, the generator loss is reduced to 0.068, and the model converges.

[0213] After training, input the abnormal feature parameters corresponding to the "irradiance mutation" specific fault type into the generator as the condition vector C to generate evaluation test samples containing irradiance mutation features; input the evaluation test samples into the probabilistic graph model to output the corresponding multi-dimensional impact probability and impact path graph. The multi-dimensional impact probability includes: fault detection impact (FDI): 0.876 (high impact), performance evaluation impact (PEI): 0.724 (high-medium impact), electric quantity prediction impact (EFI): 0.913 (high impact), economic benefit impact (EBI): estimated daily power generation loss about 3.2%, EBI=0.032; impact path graph: anomaly from irradiance sensor → string current → inverter DC power → system total output power → power generation revenue.

[0214] According to the multi-dimensional impact probability, the business impact degree corresponding to the evaluation test sample is calculated, including: setting the corresponding weight coefficient, , according to the weight coefficient and the impact probability, the business impact degree of the evaluation test sample is obtained:

[0215] P(CI(S))=t1·P(FDI)+t2·P(PEI)+t3·P(EFI)+t4·P(EBI)= (0.4×0.876)+(0.2×0.724)+(0.1×0.913)+(0.3×0.032)=0.5961.

[0216] The comprehensive business impact degree is 0.596, indicating that the fault type is "irradiance mutation", the feature parameter of the abnormal event is "irradiance mutation 300% for 15 minutes, mutation rise time 5min", and the overall business impact degree of the photovoltaic monitoring system is 59.6%. The data shows that this abnormal event is a very valuable test case, which can effectively test the core data processing and fault tolerance capability of the photovoltaic monitoring system.

[0217] According to the number of test data for testing the abnormal detection capability of the photovoltaic monitoring system, the control generator outputs a corresponding number of evaluation test samples, inputs the evaluation test samples into the probabilistic graph model, and outputs a corresponding multi-dimensional influence probability and a corresponding influence path graph. According to the calculation of the corresponding comprehensive business influence degree, the evaluation test samples are screened according to the comprehensive business influence degree, and the screened evaluation test samples are the test data for testing the abnormal detection capability of the photovoltaic monitoring system.

[0218] According to the screened evaluation test samples, the corresponding influence path graph, multi-dimensional influence probability and business influence degree are outputted, and the corresponding test report corresponding to the test data is formed. The test report includes: abnormal type: irradiance mutation, feature parameter: "irradiance mutation 300% for 15 minutes, mutation rise time 5min", multi-dimensional influence probability includes: FDI=0.876, PEI=0.724, EFI=0.913, EBI=0.032, comprehensive business influence degree P(CI(S))=0.596, main influence path: irradiance sensor→string current→inverter DC power→system total output power→power generation capacity benefit, test value rating: high.

[0219] Embodiment 3

[0220] As shown in Figure 3 , based on the photovoltaic monitoring system abnormal test data generation method provided in embodiment 1, the structure schematic diagram of the photovoltaic monitoring system abnormal test data generation device provided in embodiment 3 is shown. As shown in Figure 3 , the device includes: an adversarial generation network module 100, an evaluation framework construction module 200, a test data generation module 300, a business influence evaluation module 400 and a test data acquisition module 500.

[0221] The adversarial generation network module 100 is used to construct a GAN model, including constructing a generator network and a discriminator network. The generator network includes a physical constraint module, a multi-modal encoding-decoding structure and a time sequence generation module. The discriminator network includes a physical rule filtering layer, a multi-modal discrimination branch and a time sequence analysis module.

[0222] The evaluation framework construction module 200 is used to construct a fusion framework of knowledge graph and probabilistic graph model, including constructing an ontology model in the photovoltaic field to form a knowledge graph in the photovoltaic field, and establishing the probabilistic graph model according to the constraint relationship in the ontology model.

[0223] The test data generation module 300 is configured to iteratively train the generator network and the discriminator network according to a photovoltaic fault type, a feature parameter corresponding to the photovoltaic fault type, and a key feature corresponding to historical operation data of the photovoltaic monitoring system until convergence; and the trained generator network outputs a preset number of evaluation test samples of a predetermined photovoltaic fault type.

[0224] The business impact assessment module 400 is configured to input the preset number of evaluation test samples into the probabilistic graph model, output a multidimensional impact probability corresponding to each evaluation test sample according to mapping of the constraint relationship of the knowledge graph, and obtain a business impact degree corresponding to the evaluation test sample according to the multidimensional impact probability.

[0225] The test data acquisition module 500 is configured to filter the evaluation test samples according to the business impact degree, and obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

[0226] In an optional manner, the evaluation framework construction module 200 includes a constructed ontology model unit and a constructed probabilistic graph model unit.

[0227] The constructed ontology model unit is configured to construct an ontology model Q=(E, R, A) in a photovoltaic system, E is an entity set, R is a relationship set, and A is an attribute set, the constraint relationship of the knowledge graph includes the entity set, the relationship set, and the attribute set. The entity set E includes various physical entities in the photovoltaic system, such as E={inverter, photovoltaic module, irradiance sensor, temperature sensor, GPS module, string, combiner box...}. R is a relationship set, which describes the structure, function and causal relationship between entities, including input-output relationship, control relationship, physical connection relationship, data dependency relationship, influence relationship, etc. A is an attribute set, which is used to define the observable or calculable attributes of each entity, such as A={efficiency, temperature, voltage, current, power, coordinate, timestamp,...}. Through the ontology model, a structured photovoltaic field knowledge graph is formed, and a standardized representation of the composition, operation logic and abnormal propagation path of the photovoltaic system is realized.

[0228] The constructed probabilistic graph model unit is configured to establish the probabilistic graph model according to the constraint relationship in the knowledge graph, including:

[0229] According to the relationship set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein a node of the Bayesian structure represents a system parameter X or a business index I, and an edge represents a conditional probability distribution derived from the relationship set R. , comprising: mapping the attributes in the attribute set A as system parameter nodes X in the Bayesian structure, and mapping the pre-defined business indicators as business indicator nodes I in the Bayesian structure; determining the existence and direction of the directed edges between the system parameter nodes and between the system parameter nodes and the business indicator nodes according to the relationship set R, to obtain the Bayesian structure; and mapping the Bayesian structure and the ontology model by sharing variable nodes to form a factor graph model , a set of factor nodes constituted by the relationship-attribute in the knowledge graph, a set of variable nodes constituted by the nodes in the Bayesian structure, wherein the factor graph model is the probabilistic graph model, comprising: mapping each conditional probability distribution in the Bayesian structure as a factor node in the factor graph model, and connecting the factor node with all the variable nodes involved in the conditional probability distribution; mapping the relationships in the relationship set R as the factor nodes in the factor graph model, and connecting the factor nodes with the variable nodes corresponding to the entities participating in the relationships; wherein the variable nodes are constituted by the system parameter nodes and the business indicator nodes, and the construction of the factor graph model is completed; and calculating the factor function parameters of the factor nodes in the factor graph model by a maximum likelihood estimation method according to the historical operation data of the photovoltaic system, to obtain the trained probabilistic graph model.

[0230] In an optional manner, the test data generation module 300 comprises:

[0231] collecting historical operation data of a photovoltaic system, pre-processing the historical operation data and extracting key features of the photovoltaic system, wherein the key features include time series features, parameter correlation features and spatial distribution features; collecting fault data of the photovoltaic system, dividing the fault data according to fault types, and defining abnormal feature parameters corresponding to the fault types according to the key features in the fault data, wherein the fault types include sensor abnormality, equipment operation abnormality, communication abnormality, location information abnormality, data consistency abnormality, etc.; iteratively training the GAN model according to the photovoltaic fault types, the abnormal feature parameters corresponding to the photovoltaic fault types and the key features corresponding to the historical operation data until convergence; and controlling the generator network outputting a predetermined number of evaluation test samples corresponding to a predetermined photovoltaic fault type to evaluate the evaluation test samples.

[0232] In an optional manner, the business impact evaluation module 400 comprises: a belief propagation algorithm for performing improvement in the factor graph model, used to obtain multi-dimensional influence probability and influence path of the evaluation test samples on the business indicator nodes; the improvement of the belief propagation algorithm comprises:

[0233] The message passing rule of the belief propagation algorithm is improved by introducing a damping factor, and the improved message passing rule comprises:

[0234] The message passing rule from the variable node to the factor node is improved by introducing a damping factor a:

[0235] ;

[0236] The message passing rule from the factor node to the variable node is improved by introducing a damping factor:

[0237] ;

[0238] The damping factor is defined as: , d is the shortest path hop count between the two nodes in the factor graph model G, and β is an adjustable damping coefficient.

[0239] The improvement of the belief propagation algorithm further comprises: introducing an abnormal feature weight function w(e) in the belief propagation algorithm:

[0240] According to the photovoltaic fault type e corresponding to the evaluation test sample, the corresponding abnormal feature weight w(e) is obtained;

[0241] When calculating the message passed by the factor node to the variable node, the message is weighted according to the weight w(e) of the fault type e:

[0242] ;

[0243] e: photovoltaic fault type; is the photovoltaic fault type associated with the factor node ; is the weight function corresponding to the photovoltaic fault type .

[0244] The improvement of the belief propagation algorithm further comprises: adopting a hierarchical iteration strategy to execute the improved belief propagation algorithm to obtain the influence probability and the influence path, comprising:

[0245] Input the device parameters in the evaluation test sample, perform belief propagation between nodes within a single photovoltaic device, and output device layer messages;

[0246] Input the device layer messages into the system layer, perform belief propagation between cross-device nodes, and output system layer messages;

[0247] Input the system layer messages, globally pass to the business index node, and output the influence path and the influence probability.

[0248] In an implementation, the business impact assessment module 400 further comprises: directly quantifying the possibility of the abnormality causing the business index to deviate from the normal range according to the posterior marginal probability, to obtain a multi-dimensional impact probability, comprising: obtaining the posterior marginal probability, respectively denoted as: fault detection abnormality probability P fd, performance evaluation abnormality probability P pe, power prediction abnormality probability P ef, economic benefit abnormality probability P eb; constructing an evaluation matrix according to the fault sensitivity function, the performance deviation function, the prediction error accumulation function and the economic loss estimation model; inputting the posterior marginal probability into the evaluation matrix, calculating the abnormality probability of each dimension, if the value exceeds the preset threshold, taking the abnormality probability as a trigger signal, and taking the original evaluation test sample S and the normal benchmark data S0 as input, calculating the specific impact degree value of this dimension: if P fd exceeds the threshold, calling the fault detection impact function, inputting S and S0, calculating and outputting the fault detection impact degree FDI; if P pe exceeds the threshold, calling the performance evaluation impact function, inputting S and S0, calculating and outputting the performance evaluation impact degree PEI; if P ef exceeds the threshold, calling the power prediction impact function, inputting S and S0, calculating and outputting the power prediction impact degree EFI; if P eb exceeds the threshold, calling the economic benefit impact function, inputting S and S0, and comprehensively depending on the calculation results of PEI and EFI, calculating and outputting the economic benefit impact degree EBI. Output fault detection impact probability P(FDI), performance evaluation impact probability P(PEI), power prediction impact probability P(EFI) and economic benefit impact probability P(EBI).

[0249] The posterior is then calculated according to the multi-dimensional impact probability to calculate the business impact degree corresponding to the evaluation test sample, comprising: obtaining the multi-dimensional impact probability corresponding to the evaluation test sample, the multi-dimensional impact probability comprising: fault detection impact possibility P(FDI), performance evaluation impact possibility P(PEI), power prediction impact possibility P(EFI), economic benefit impact possibility P(EBI); setting a corresponding weight coefficient for the impact possibility corresponding to the impact possibility, and obtaining the business impact degree of the evaluation test sample according to the weight coefficient and the impact possibility:

[0250] , wherein is a weight coefficient and satisfies .

[0251] In an implementation manner, the business influence evaluation module 400 further comprises: inputting the evaluation test sample into the probabilistic graph model, outputting an influence path corresponding to the evaluation test sample; mapping the influence path into a domain entity relationship according to the constraint relationship of the ontology model, outputting an influence path graph corresponding to the evaluation test sample according to the domain entity relationship; outputting a test report of the evaluation test sample according to the influence path graph, the multi-dimensional influence probability and the business influence degree, wherein the test report comprises: a fault type, an influence path graph, a multi-dimensional influence probability and a business influence degree.

[0252] The device provided by the embodiment of the present application can quickly and automatically generate test data corresponding to more than several photovoltaic fault types by constructing a generative adversarial network (GAN) model of the adversarial generation network module, which integrates physical constraints, multi-modal encoding-decoding structure and time sequence generation module, and combining a deep fusion framework of a knowledge graph and a probabilistic graph model. The test data is used for input into a photovoltaic monitoring system to detect the data processing capability of the photovoltaic monitoring system on the photovoltaic system, and solves the problems of incomplete coverage, low authenticity and poor consistency caused by the dependence of traditional test data on limited real abnormal scenes or artificial construction. The standard generative adversarial network (GAN) model is improved by introducing a physical constraint mechanism, a time sequence correlation maintenance strategy and multi-modal data fusion, so as to ensure that the evaluation test sample generated by the generator conforms to the operation law of the photovoltaic system and has time continuity and cross-modal consistency. The joint reasoning of the knowledge graph and the probabilistic graph model realizes the quantitative evaluation and path tracking of the business influence of the evaluation test sample, and significantly improves the pertinence, practicality and evaluation efficiency of the test data. The present application can be widely applied to the testing and verification of data cleaning, fault tolerance processing, fault detection and other functions of the photovoltaic monitoring system, and has important engineering application value and promotion prospect.

[0253] Embodiment 4

[0254] Figure 4 The structural schematic diagram of the computer device provided by the embodiment 4 of the present application is shown, and the specific implementation of the computer device is not limited in the specific embodiment of the present application.

[0255] As shown in Figure 3 , the computer device can include a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0256] The processor 402, the communication interface 404, and the memory 406 communicate with each other through the communication bus 408. The communication interface 404 is configured to communicate with network elements such as clients or other servers. The processor 402 is configured to execute the program 410, and particularly can execute the related steps in any of the embodiments of the method for generating abnormal test data of a photovoltaic monitoring system provided in Embodiment 1.

[0257] Specifically, the program 410 can include program codes including computer-executable instructions.

[0258] The processor 402 can be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the device can be of the same type, such as one or more CPUs, or can be of different types, such as one or more CPUs and one or more ASICs.

[0259] The memory 406 is configured to store the program 410. The memory 406 can include a high-speed RAM memory, and can also include a non-volatile memory such as at least one disk memory.

[0260] The program 410 can particularly be invoked by the processor 402 to cause the computer device to perform the following operations:

[0261] The embodiments of the present application provide a computer-readable storage medium, which stores at least one executable instruction, and the executable instruction causes a computer device to perform the method for generating abnormal test data of a photovoltaic monitoring system in any of the method embodiments when the computer device runs the executable instruction.

[0262] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general purpose systems can be used with these teachings, or with modifications that take into account the teachings herein. In view of the above description, those of ordinary skill in the art will be able to devise various arrangements that are considered within the scope of the embodiments of the present application. Further, the embodiments of the present application are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages can be used to implement the teachings of the present application as described herein, and any references below to specific languages are provided for disclosure of enablement of the best mode of the present application.

[0263] In the specification provided herein, a large number of specific details are described. However, it can be understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order not to obscure the understanding of this specification.

[0264] Similarly, it is to be understood that the features of the embodiments of the present application described above, while initially chosen to facilitate the present application and to aid in the understanding of one or more aspects of the present application, can be provided in any combination of one or more aspects of the present application. It is to be understood that the disclosed method is not to be interpreted as reflecting an intention that the claimed application require a greater scope that the features expressly recited in each claim.

[0265] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and can be further split into more sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the specification (including the accompanying claims, abstract and drawings), and any method or process or steps of an apparatus so disclosed can be made unless it is explicitly stated otherwise. Each feature disclosed in the specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features providing the same, equivalent or similar functionality unless it is explicitly stated otherwise.

[0266] It is noted that the foregoing examples have been provided merely for the purposes of illustration and are not intended to limit the application of the application, which is defined by the claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word comprising does not exclude the presence of elements or steps other than those listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of both hardware and software, and any combinations of these. In a unit claim, any reference signs placed between parentheses in the claim do not limit the claim. The use of the word "at least" followed by a list of elements does not exclude the presence of one or more additional elements of the list. The use of the word "one" or "said" to refer to an element does not exclude the presence of a plurality of such elements. The implementation of the method steps, processes, or procedures disclosed in the specification as an engine, software, or code can not require the physical operating of the internal mechanism of a physical engine, software, or code, in cases in which the engine, software, or code produces a physical result or discloses a physical transformation.

Claims

1. A photovoltaic monitoring system anomaly test data generation method, characterized by, The method comprises: constructing a GAN model, including constructing a generator network and a discriminator network, the generator network comprising a physical constraint module, a multi-modal encoding-decoding structure and a time sequence generation module; the discriminator network comprising a physical rule filtering layer, a multi-modal discrimination branch and a time sequence analysis module; The fusion framework of the knowledge graph and the probability graph model is constructed, including constructing an ontology model in a photovoltaic field to form a knowledge graph in the photovoltaic field, and establishing the probability graph model according to constraint relationships in the knowledge graph, including: constructing an ontology model Q=(E, R, A) in a photovoltaic system, E is an entity set, R is a relationship set, and A is an attribute set, the constraint relationships of the knowledge graph include the entity set, the relationship set, and the attribute set; the probability graph model is established according to the constraint relationships in the knowledge graph, including: according to the relationship set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein a node of the Bayesian structure represents a system parameter X or a business index I, and an edge represents a conditional probability distribution P(I|X) derived from the relationship set R; the Bayesian structure and the ontology model are jointly mapped through a shared variable node to form a factor graph model , is a set of factor nodes composed of relationships-attributes in the knowledge graph, is a set of variable nodes composed of nodes in the Bayesian structure, and the factor graph model is the probability graph model; according to historical operation data of the photovoltaic system, the factor function parameters corresponding to the factor nodes in the factor graph model are calculated through a maximum likelihood estimation method, and a trained probability graph model is obtained. iteratively training the generator network and the discriminator network according to the photovoltaic fault type, the feature parameters corresponding to the photovoltaic fault type and the key features corresponding to the historical operation data of the photovoltaic monitoring system until convergence; the trained generator network outputs a preset number of evaluation test samples of predetermined photovoltaic fault types; inputting the preset number of evaluation test samples into the probabilistic graph model, outputting the multi-dimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtaining the business influence degree corresponding to the evaluation test sample according to the multi-dimensional influence probability; screening the evaluation test samples according to the business influence degree to obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

2. The method of claim 1, wherein, The method further comprises: inputting the evaluation test sample into the probabilistic graph model to output the influence path corresponding to the evaluation test sample; mapping the influence path to a domain entity relationship according to the constraint relationship of the ontology model, and outputting the influence path graph corresponding to the evaluation test sample according to the domain entity relationship; outputting a test report of the evaluation test sample according to the influence path graph, the multi-dimensional influence probability and the business influence degree, wherein the test report comprises: fault type, influence path graph, multi-dimensional influence probability and business influence degree.

3. The method according to claim 1 or 2, characterized in that, the parameter-index causal network is established according to the relationship set and the attribute set in the ontology model, wherein a node of the Bayesian structure represents a system parameter X or a business index I, and an edge represents a conditional probability distribution derived from the relationship set R comprising: mapping the attributes in the attribute set A to the system parameter nodes in the Bayesian structure; mapping the pre-defined business indicators to the business indicator nodes in the Bayesian structure; determining the existence and direction of the directed edges between the system parameter nodes and between the system parameter nodes and the business indicator nodes according to the relationship set R, and obtaining the Bayesian structure.

4. The method of claim 3, wherein, mapping the Bayesian structure and the ontology model through a shared variable node to form a factor graph model , a set of factor nodes composed of relations-attributes in the knowledge graph, a set of variable nodes composed of nodes in the Bayesian structure, the factor graph model being the probabilistic graph model, comprising: mapping each conditional probability distribution in the Bayesian structure to a factor node in the factor graph model, and connecting the factor node to all variable nodes involved in the conditional probability distribution; mapping the relationships in the relationship set R to the factor nodes in the factor graph model, and connecting the factor nodes to the variable nodes corresponding to the entities participating in the relationships; wherein the variable nodes are composed of the system parameter nodes and the business indicator nodes, and the construction of the factor graph model is completed.

5. The photovoltaic monitoring system anomaly test data generation method of claim 4, wherein, The method comprises the following steps: The improved belief propagation algorithm is executed in the factor graph model to obtain the multi-dimensional influence probability and influence path of the evaluation test sample on the business index node; The improvement of the belief propagation algorithm comprises: An attenuation factor is introduced in the message passing rule of the belief propagation algorithm for improvement, and the improved message passing rule comprises: Introducing a decay factor in the message passing rules from the variable nodes to the factor nodes Improving the message passing rules: ; An attenuation factor is introduced in the message passing rule from the factor node to the variable node for improvement of the message passing rule: ; The attenuation factor is defined as: , β > 0, d is the shortest path hop count between two nodes in the factor graph model G, is an adjustable attenuation coefficient.

6. The photovoltaic monitoring system anomaly test data generation method of claim 5, wherein, The improvement of the belief propagation algorithm further comprises: introducing an abnormal feature weight function w(e) in the belief propagation algorithm: According to the photovoltaic fault type e corresponding to the evaluation test sample, the corresponding abnormal feature weight w(e) is obtained; When calculating the message passed from the factor node to the variable node, the message is weighted according to the weight w(e) of the fault type e: ; e: photovoltaic fault type; for the photovoltaic fault type associated with the factor node for the photovoltaic fault type associated with the factor node f; for the photovoltaic fault type associated with the factor node f.

7. The photovoltaic monitoring system anomaly test data generation method according to claim 5 or 6, characterized in that, The improvement of the belief propagation algorithm further comprises: adopting a hierarchical iteration strategy to execute the improved belief propagation algorithm to obtain the influence probability and influence path, comprising: Input the equipment parameters in the evaluation test sample, execute belief propagation between nodes in a single photovoltaic equipment, and output equipment layer messages; Input the system layer messages, execute belief propagation between cross-device nodes, and output system layer messages; Input the system layer messages, globally pass to the business index node, and output the influence path and the influence probability.

8. A photovoltaic monitoring system anomaly test data generation apparatus characterized by comprising: The device comprises: An adversarial generative network module for constructing a GAN model, including constructing a generator network and a discriminator network, wherein the generator network comprises a physical constraint module, a multi-modal encoding-decoding structure, and a time sequence generation module; the discriminator network comprises a physical rule filtering layer, a multi-modal discrimination branch, and a time sequence analysis module; The evaluation framework modeling module is configured to build a fusion framework of a knowledge graph and a probabilistic graph model, including building an ontology model in a photovoltaic field to form a knowledge graph of the photovoltaic field, and establishing the probabilistic graph model according to constraint relationships in the knowledge graph, including: building an ontology model Q=(E, R, A) in a photovoltaic system, E is an entity set, R is a relationship set, and A is an attribute set, the constraint relationships of the knowledge graph include an entity set, a relationship set, and an attribute set; the probabilistic graph model is established according to the constraint relationships in the knowledge graph, including: according to the relationship set and the attribute set in the ontology model, a Bayesian structure of a parameter-index causal network is established, wherein the nodes of the Bayesian structure represent system parameters X or business indicators I, and the edges represent conditional probability distributions derived from the relationship set R ; the Bayesian structure and the ontology model are jointly mapped through shared variable nodes to form a factor graph model , is a set of factor nodes composed of relationship-attribute in the knowledge graph, is a set of variable nodes composed of nodes in the Bayesian structure, and the factor graph model is the probabilistic graph model; according to historical operation data of the photovoltaic system, the factor function parameters corresponding to the factor nodes in the factor graph model are calculated through a maximum likelihood estimation method to obtain a trained probabilistic graph model; A test data generation module for iteratively training the generator network and the discriminator network according to the photovoltaic fault type, the feature parameters corresponding to the photovoltaic fault type, and the key features corresponding to the historical operation data of the photovoltaic monitoring system until convergence; the trained generator network outputs a preset number of evaluation test samples of predetermined photovoltaic fault types; A business impact evaluation module for inputting a preset number of evaluation test samples into the probabilistic graph model, outputting the multi-dimensional influence probability corresponding to each evaluation test sample according to the mapping of the constraint relationship of the knowledge graph, and obtaining the business impact degree corresponding to the evaluation test sample according to the multi-dimensional influence probability; A test data acquisition module for screening the evaluation test samples according to the business impact degree to obtain test data for testing the abnormal detection capability of the photovoltaic monitoring system.

9. A computer device, comprising: It comprises: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface complete communication with each other through the communication bus. The memory is configured to store at least one executable instruction, and the executable instruction is configured to enable the processor to perform the operation of the photovoltaic monitoring system abnormal test data generation method according to any one of claims 1-7.

Images