A photovoltaic power station fault detection method based on knowledge graph

Abstract

The present invention discloses a photovoltaic power station fault detection method based on a knowledge graph, which relates to the field of photovoltaic power station fault detection, including collecting multi-dimensional time series data and preprocessing it, constructing a knowledge graph and a graph neural network model, and learning node features on the knowledge graph, generating an embedded representation of the node, establishing a fault detection model, and realizing the identification of photovoltaic power station faults through link prediction technology; optimizing and evaluating the confidence of the fault detection results of the fault detection model through an energy propagation mechanism, and completing the photovoltaic power station fault detection. The present invention realizes the conversion of heterogeneous time series data into structured graph data by constructing a knowledge graph; constructing a graph neural network model according to the knowledge graph and learning node features, generating an embedded representation of the node, establishing a fault detection model, and realizing fault identification through link prediction technology, so that the system identifies potential faults at an early stage, reduces the downtime of the photovoltaic power station, and improves the operating efficiency.

Description

Technical Field

[0001] The present invention relates to the field of fault detection using knowledge graph technology, and in particular to a photovoltaic power station fault detection method based on knowledge graph. Background Art

[0002] In the operation of modern photovoltaic power stations, due to the large number of equipment, complex systems and diverse operating environments, photovoltaic power stations face various types of fault risks. These faults may reduce power generation efficiency and even cause safety hazards. Therefore, fault detection and diagnosis technology for photovoltaic power stations is particularly important.

[0003] Traditional photovoltaic power plant fault detection methods mainly rely on expert experience or rule-based methods, which usually identify faults by monitoring specific parameters or indicators. However, with the expansion of photovoltaic power plant system scale and the diversification of operating environments, these experience-based and rule-based methods are unable to cope with complex and new faults. Especially in large-scale data environments, traditional methods are difficult to effectively process multi-dimensional time series data, and cannot fully utilize the correlation information between different devices for accurate fault detection.

[0004] In recent years, with the rapid development of big data technology and artificial intelligence, data-driven methods have gradually been applied to fault detection in photovoltaic power plants. These methods can mine potential patterns and abnormal behaviors in equipment operation by analyzing large amounts of historical data. However, such methods usually focus on statistical or machine learning techniques, which are difficult to fully capture the complex relationships between devices, especially when dealing with complex faults with spatiotemporal dependencies, and have certain limitations.

[0005] As a technology that can effectively represent entities and their relationships, knowledge graphs have been widely used in many fields. They can intuitively present multidimensional data and complex relationships in the form of graphs, and through advanced machine learning technologies such as graph neural networks, they can learn deeper features and related information. Therefore, the photovoltaic power station fault detection method based on knowledge graphs is expected to break through the limitations of traditional methods and provide a photovoltaic power station fault detection method based on knowledge graphs. Summary of the invention

[0006] In view of the fact that the existing photovoltaic power station fault detection methods have poor detection accuracy and real-time performance in complex environments and are difficult to effectively identify new faults, the present invention is proposed.

[0007] Therefore, the problem to be solved by the present invention is how to use knowledge graph and graph neural network technology to improve the accuracy and efficiency of photovoltaic power station fault detection and identify unknown fault types.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In the first aspect, an embodiment of the present invention provides a photovoltaic power station fault detection method based on a knowledge graph, which includes collecting multidimensional time series data during the operation of the photovoltaic power station, and preprocessing the multidimensional time series data to construct a knowledge graph; based on the spatiotemporal structure and semantic information of the knowledge graph, constructing a graph neural network model, and learning the node features on the knowledge graph to generate an embedded representation of the node; according to the embedded representation of the node, establishing a fault detection model, and realizing the identification of photovoltaic power station faults through link prediction technology; optimizing and confidence evaluating the fault detection results of the fault detection model through an energy propagation mechanism to complete the photovoltaic power station fault detection.

[0010] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, wherein: the multidimensional time series data includes photovoltaic power station operation data, photovoltaic component status data and inverter status data; the embedded representation includes a first embedded representation and a second embedded representation; the energy propagation mechanism includes the following steps: based on the second embedded representation of the node, constructing an energy function to calculate the energy score of the node; using the node connection relationship of the knowledge graph to construct an adjacency matrix to define the energy propagation path between nodes; at the same time, the energy vector of the node is initialized, and energy propagation is performed based on the adjacency matrix; the energy score of the node is updated through several iterative propagations to determine the fault state of the node; according to the energy score of the node, the fault detection result of the fault detection model is confidence evaluated, the fault node is identified, and the fault detection result is output.

[0011] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, the specific formula of the energy function is as follows:

[0012] ;

[0013] in, is the energy score of the second embedding representation of node v, is the number of fault categories, is the weight vector corresponding to category a, is the second embedding representation of node v at time t.

[0014] The specific formula for energy propagation is as follows:

[0015] ;

[0016] in, is the energy score of node v at layer k, is the adjacency matrix value between node u and node v, , and To control the parameters of energy propagation, is the neighbor set of node v.

[0017] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, it also includes: when the embedded representation of the node is within the preset range, the monitoring continues and the state of this node is recorded as normal; when the embedded representation of the node deviates from the preset range, the node is marked as a potential fault point and fault detection is performed; the fault detection includes, if a potential fault point is detected, the link prediction technology is used for analysis, and the confidence assessment is performed through the energy propagation mechanism; if the probability of the fault link exceeds the preset threshold, the link is confirmed as a fault link; if several adjacent links are marked as fault links at the same time, the device group connected to the link is marked as a fault area, and the fault type is identified; if the fault feature is a known fault type, it is classified into the corresponding fault type; if the fault feature pattern has not been seen, it will be marked as an unknown fault type and a manual review process will be started; the confidence assessment includes, when the energy score of the node When the energy score of the node is greater than the confidence threshold, it is reported as a high confidence fault. When it is less than the confidence threshold, the fault is not reported temporarily and the node is added to the monitoring list.

[0018] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, the construction process of the knowledge graph includes the following steps: feature extraction and time-varying entity extraction of the multi-dimensional time series data, identification of equipment status and operating conditions; using the extracted feature mapping as nodes and relationships in the knowledge graph, calculating the time-varying relationship strength between entities, and selecting key entity pairs. The specific formula is as follows:

[0019] ;

[0020] in, is the coefficient of the time-varying relationship strength between entity i and entity j at time t, is the mutual relationship coefficient between entity i and entity j at time t, is the transfer entropy between entity i and entity j at time t, and is a balance parameter.

[0021] The coefficient of the time-varying relationship strength is used as the weight of the edge to construct a knowledge graph, and the prior knowledge is used to supplement the spatiotemporal structure and semantic information of the knowledge graph to obtain a tensor G with a dimension of M×M×L as the input of the graph neural network model; where M is the total number of entities and L is the time step.

[0022] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, wherein: the construction step of the graph neural network model includes the following steps: designing a graph convolution operation suitable for the knowledge graph structure, aggregating the node and neighbor node features in the knowledge graph structure, and generating a first embedded representation of the node; based on the first embedded representation, introducing an attention mechanism to enhance the weight of the node feature in the embedded representation; through multi-layer graph convolution, combining spatiotemporal features and semantic information, aggregating the information of neighbor nodes, and processing the first embedded representation of the node through a nonlinear activation function to generate a second embedded representation of the node, the specific formula is as follows:

[0023] ;

[0024] in, is the second embedding representation of node v at layer k, is the second embedding representation of node u at layer k-1, is the neighbor set of node v, is the attention weight of node u to node v, is the weight matrix of the kth layer, is the bias vector of the kth layer, is a non-linear activation function.

[0025] As a preferred solution of the photovoltaic power station fault detection method based on knowledge graph described in the present invention, wherein: the link prediction technology is a generative adversarial network method based on deep learning; the construction process of the fault detection model includes the following steps: according to the second embedding representation of the node, the link prediction technology is used to identify the fault link in the photovoltaic power station, and the fault detection model is constructed and trained; the cross entropy loss function is used to optimize the fault detection model parameters through back propagation to minimize the error between the prediction result and the actual annotation. The specific formula is as follows:

[0026] ;

[0027] in, The second embedding representation for node v is and the second embedding representation of node u The probability of a failed link, is the label value of the fault link between node u and node v, is the weight matrix of the fault detection model.

[0028] In the second aspect, an embodiment of the present invention provides a photovoltaic power station fault detection system based on a knowledge graph, which includes: an acquisition module, which is used to collect multi-dimensional time series data during the operation of the photovoltaic power station, and pre-process the multi-dimensional time series data to construct a knowledge graph; a construction module, which constructs a graph neural network model based on the spatiotemporal structure and semantic information of the knowledge graph, and learns the node features on the knowledge graph to generate an embedded representation of the node; an identification module, which establishes a fault detection model according to the embedded representation of the node, and realizes the identification of photovoltaic power station faults through link prediction technology; an optimization module, which is used to optimize and confidence evaluate the fault detection results of the fault detection model through an energy propagation mechanism to complete the photovoltaic power station fault detection.

[0029] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the photovoltaic power station fault detection method based on the knowledge graph as described in the first aspect of the present invention are implemented.

[0030] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the photovoltaic power station fault detection method based on the knowledge graph as described in the first aspect of the present invention are implemented.

[0031] The beneficial effects of the present invention are as follows: the present invention realizes the conversion of heterogeneous time series data into structured graph data by constructing a knowledge graph; based on the spatiotemporal structure and semantic information of the knowledge graph, a graph neural network model is constructed and node features are learned to generate an embedded representation of the node, so that the system can effectively capture the multi-dimensional features of each component and its association in the photovoltaic power station and improve the model's perception of fault modes; according to the embedded representation of the node, a fault detection model is established and fault identification is realized through link prediction technology, so that the system can identify potential faults at an early stage, reduce the downtime of the photovoltaic power station, and thus improve the operating efficiency; the fault detection results are optimized and the confidence is evaluated through the energy propagation mechanism. The accuracy and reliability of the fault detection model are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0033] Figure 1 This is a flow chart of a photovoltaic power station fault detection method based on knowledge graph in Example 1. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0037] Example 1

[0038] Reference Figure 1 , which is the first embodiment of the present invention, and provides a photovoltaic power station fault detection method based on knowledge graph, including:

[0039] S1: Collect multi-dimensional time series data during the operation of photovoltaic power stations, pre-process the multi-dimensional time series data, and construct a knowledge graph.

[0040] Specifically, the multi-dimensional time series data includes photovoltaic power station operation data, photovoltaic module status data and inverter status data.

[0041] Furthermore, the construction process of the knowledge graph includes the following steps: feature extraction and time-varying entity extraction of multi-dimensional time series data to identify equipment status and operating conditions; using the extracted feature mapping as nodes and relationships in the knowledge graph, calculating the time-varying relationship strength between entities, and selecting key entity pairs. The specific formula is as follows:

[0042] ;

[0043] in, is the coefficient of the time-varying relationship strength between entity i and entity j at time t, is the mutual relationship coefficient between entity i and entity j at time t, is the transfer entropy between entity i and entity j at time t, and is a balance parameter.

[0044] Furthermore, the coefficient of time-varying relationship strength is used as the weight of the edge to construct a knowledge graph, and prior knowledge is used to supplement the knowledge graph with spatiotemporal structure and semantic information to obtain a tensor G with a dimension of M×M×L as the input of the graph neural network model; where M is the total number of entities and L is the time step.

[0045] It should be noted that multi-dimensional time series data refers to data of multiple dimensions collected at different time points, including photovoltaic power station operation data, photovoltaic module status data and inverter status data in the present invention. These data reflect the overall operation status of photovoltaic power stations at different time points. Knowledge graph is a structured knowledge representation method that describes the relationship between entities through nodes and edges. In the present invention, knowledge graph is used to represent the complex relationship and operation status between the components of photovoltaic power stations.

[0046] Specifically, the time-varying relationship strength describes the degree to which the relationship between entities in the knowledge graph changes over time. It is calculated by the mutual correlation coefficient and transfer entropy, reflecting the mutual influence of dynamic changes between PV power station components.

[0047] Furthermore, feature extraction and time-varying entity extraction can extract key information from massive multi-dimensional time series data and identify equipment status and operating conditions; time-varying relationship strength calculation accurately quantifies the dynamic relationship between entities by introducing mutual correlation coefficients and transfer entropy.

[0048] S2: Based on the spatiotemporal structure and semantic information of the knowledge graph, a graph neural network model is constructed, and the node features on the knowledge graph are learned to generate an embedded representation of the node.

[0049] Specifically, the embedding representation includes a first embedding representation and a second embedding representation. The steps for constructing the graph neural network model include the following steps: designing a graph convolution operation suitable for the knowledge graph structure, aggregating the features of nodes and neighboring nodes in the knowledge graph structure, and generating the first embedding representation of the node; based on the first embedding representation, introducing an attention mechanism to enhance the weight of node features in the embedding representation; through multi-layer graph convolution, combining spatiotemporal features and semantic information, aggregating the information of neighboring nodes, and processing the first embedding representation of the node through a nonlinear activation function to generate the second embedding representation of the node. The specific formula is as follows:

[0050] ;

[0051] in, is the second embedding representation of node v at layer k, is the second embedding representation of node u at layer k-1, is the neighbor set of node v, is the attention weight of node u to node v, is the weight matrix of the kth layer, is the bias vector of the kth layer, is a non-linear activation function.

[0052] It should be noted that the first embedding representation is generated through graph convolution operations, which mainly aggregates the features of nodes and their neighboring nodes, captures the local structure and relationship of nodes in the graph, and reflects the basic properties of nodes and direct connections with neighbors. Since the generation process depends on neighborhood information, the first embedding representation provides a basis for subsequent feature learning. The second embedding representation is further enhanced on the basis of the first embedding representation by introducing attention mechanisms and multi-layer graph convolutions. It combines spatiotemporal features with richer semantic information to reflect the role and importance of nodes in the entire graph. By dynamically adjusting the weights of features, the second embedding representation can accurately capture key information, enabling the model to demonstrate stronger reasoning capabilities when processing complex data.

[0053] S3: Based on the embedded representation of the nodes, a fault detection model is established, and the fault identification of the photovoltaic power station is realized through the link prediction technology.

[0054] Specifically, link prediction technology is a generative adversarial network method based on deep learning; the generative adversarial network (GAN) method based on deep learning has shown significant advantages in the field of link prediction. Link prediction aims to identify edges that do not yet exist in the graph, and GAN effectively improves the accuracy of prediction through the adversarial training mechanism of the generator and the discriminator. The generator uses the structural information and node features of the graph to generate potential connections, while the discriminator optimizes the model's judgment ability by analyzing the differences between the generated edges and the real edges. This adversarial process prompts the generator to continuously adjust the generation strategy to make it more consistent with the real data distribution, thereby improving the model's ability to learn complex graph structures.

[0055] Furthermore, the construction process of the fault detection model includes the following steps: according to the second embedding representation of the node, the link prediction technology is used to identify the faulty links in the photovoltaic power station, and the fault detection model is constructed and trained; the cross entropy loss function is used to optimize the fault detection model parameters through back propagation to minimize the error between the prediction result and the actual annotation. The specific formula is as follows:

[0056] ;

[0057] in, The second embedding representation for node v is and the second embedding representation of node u The probability of a failed link, is the label value of the fault link between node u and node v, is the weight matrix of the fault detection model.

[0058] S4: The fault detection results of the fault detection model are optimized and confidence evaluated through the energy propagation mechanism to complete the fault detection of the photovoltaic power station.

[0059] Specifically, the energy propagation mechanism includes the following steps: based on the second embedding representation of the node, construct an energy function to calculate the energy score of the node; use the node connection relationship of the knowledge graph to construct an adjacency matrix to define the energy propagation path between nodes; initialize the energy vector of the node at the same time, and perform energy propagation based on the adjacency matrix; update the energy score of the node through several iterative propagation to determine the fault state of the node; according to the energy score of the node, perform confidence evaluation on the fault detection result of the fault detection model, identify the faulty node, and output the fault detection result.

[0060] Furthermore, the specific formula of the energy function is as follows:

[0061] ;

[0062] in, is the energy score of the second embedding representation of node v, is the number of fault categories, is the weight vector corresponding to category a, is the second embedding representation of node v at time t.

[0063] Furthermore, the specific formula for energy propagation is as follows:

[0064] ;

[0065] in, is the energy score of node v at layer k, is the adjacency matrix value between node u and node v, , and To control the parameters of energy propagation, is the neighbor set of node v.

[0066] Specifically, when the embedding representation of a node is within a preset range, monitoring continues and the state of the node is recorded as normal; when the embedding representation of a node deviates from the preset range, the node is marked as a potential fault point and fault detection is performed.

[0067] Furthermore, fault detection includes, if a potential fault point is detected, using link prediction technology for analysis and performing confidence assessment through the energy propagation mechanism; if the probability of a faulty link exceeds a preset threshold, the link is confirmed as a faulty link; if several adjacent links are marked as faulty links at the same time, the device group connected to the link is marked as a faulty area to identify the fault type; if the fault feature is a known fault type, it is classified into the corresponding fault type; if the fault feature pattern is unseen, it is marked as an unknown fault type and the manual review process is initiated.

[0068] Furthermore, the confidence evaluation includes when the energy score of the node When the energy score of the node is greater than the confidence threshold, it is reported as a high confidence fault. When it is less than the confidence threshold, the fault is not reported temporarily and the node is added to the monitoring list.

[0069] It should be noted that the confidence threshold is customized according to the characteristics and operational requirements of the PV power plant. It is used to determine the fault status of the node in the energy propagation mechanism and help the system distinguish between high-confidence faults that need to be reported immediately and low-confidence anomalies that need to be continuously monitored.

[0070] Example 2

[0071] On the basis of the first embodiment, this embodiment further provides a photovoltaic power station fault detection system based on a knowledge graph, including: an acquisition module, which is used to collect multi-dimensional time series data during the operation of the photovoltaic power station, and pre-process the multi-dimensional time series data to construct a knowledge graph; a construction module, which constructs a graph neural network model based on the spatiotemporal structure and semantic information of the knowledge graph, and learns the node features on the knowledge graph to generate an embedded representation of the node; an identification module, which establishes a fault detection model according to the embedded representation of the node, and realizes the identification of photovoltaic power station faults through link prediction technology; an optimization module, which is used to optimize and confidence evaluate the fault detection results of the fault detection model through an energy propagation mechanism to complete the photovoltaic power station fault detection.

[0072] This embodiment also provides a computer device, which is suitable for the photovoltaic power station fault detection method based on the knowledge graph, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the photovoltaic power station fault detection method based on the knowledge graph proposed in the above embodiment.

[0073] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0074] This embodiment also provides a storage medium having a computer program stored thereon, which implements the following steps when executed by a processor: collecting multidimensional time series data during the operation of a photovoltaic power station, and preprocessing the multidimensional time series data to construct a knowledge graph; constructing a graph neural network model based on the spatiotemporal structure and semantic information of the knowledge graph, and learning the node features on the knowledge graph to generate an embedded representation of the node; establishing a fault detection model based on the embedded representation of the node, and identifying photovoltaic power station faults through link prediction technology; optimizing and confidence evaluating the fault detection results of the fault detection model through an energy propagation mechanism to complete photovoltaic power station fault detection.

[0075] In summary, the present invention realizes the conversion of heterogeneous time series data into structured graph data by constructing a knowledge graph; based on the spatiotemporal structure and semantic information of this knowledge graph, a graph neural network model is constructed and node features are learned to generate an embedded representation of the node, so that the system can effectively capture the multi-dimensional features of each component and its association in the photovoltaic power station and improve the model's perception of fault modes; according to the embedded representation of the node, a fault detection model is established and fault identification is realized through link prediction technology, so that the system can identify potential faults at an early stage, reduce the downtime of the photovoltaic power station, and thus improve the operating efficiency; the fault detection results are optimized and the confidence is evaluated through the energy propagation mechanism. Improve the accuracy and reliability of the fault detection model.

[0076] Example 3

[0077] Referring to Table 1, which is the second embodiment of the present invention, this embodiment provides a photovoltaic power station fault detection method based on a knowledge graph. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0078] Specifically, the experiment selected the actual operating data of a photovoltaic power station, including multi-dimensional time series data of photovoltaic modules, inverters, combiner boxes and transformers, including information on voltage, current, power, temperature, humidity and irradiance; the data collection process uses high-precision sensors to record the operating status of the photovoltaic power station in real time, generate a multi-dimensional time series data set, and perform pre-processing operations on the collected data, including denoising, removing outliers, filling missing values ​​and data normalization. After data pre-processing, feature extraction technology is used to identify the key operating status and working conditions of the equipment from the time series data, such as the input and output parameters of the inverter and the operating voltage of the photovoltaic module.

[0079] Furthermore, the preprocessed data is used to construct a knowledge graph of the photovoltaic power station. The nodes of the knowledge graph represent various types of equipment in the power station, and the edges represent the associations between the equipment, such as electrical connections, power transmission, and state dependencies. By calculating the time-varying relationship strength between nodes and combining the semantic information of the equipment, a knowledge graph with a spatiotemporal structure is formed. Based on the constructed knowledge graph, the graph neural network model is used to learn the node features and generate an embedded representation of each node. The graph neural network model aggregates information by combining the spatiotemporal features of the nodes and the features of the neighboring nodes through multi-layer graph convolution operations. The generated embedded representation is used as the input of the fault detection model. The attention mechanism is introduced in the graph neural network to distinguish the degree of influence of different neighboring nodes on the target node.

[0080] Furthermore, the fault detection model uses link prediction technology to identify potential fault links in photovoltaic power plants based on the embedded representation of nodes; link prediction technology identifies possible faulty nodes by calculating the similarity or potential connection possibility between nodes; to further optimize the fault detection results, the model introduces an energy propagation mechanism; the energy propagation mechanism iteratively optimizes the energy score of the node by constructing an energy function, and gradually converges to a stable state by combining the state information of neighboring nodes, and finally outputs a high-confidence fault detection result. In order to verify the effectiveness of the proposed method, the fault detection technology of the method of the present invention is compared with the fault detection technology of the traditional machine learning method. The test sets multiple indicators, including fault detection accuracy, sensitivity, specificity, and computational overhead, to comprehensively evaluate the performance of the method.

[0081] Specifically, as shown in Table 1, it can be clearly seen that the fault detection method based on the knowledge graph is superior to the existing traditional methods in all indicators. First, in terms of fault detection accuracy, the fault detection method based on the knowledge graph reached 97.8%, which is significantly higher than the 85.3% of the traditional machine learning method and the 78.9% of the traditional statistical analysis method. This significant improvement can be attributed to the fact that the knowledge graph can effectively capture the complex relationships between the various devices in the photovoltaic power station and deeply learn these relationships through the graph neural network model. In contrast, the traditional machine learning method cannot fully utilize the correlation information between devices and can only rely on a single feature, which limits its performance in complex fault detection scenarios.

[0082] Table 1 Comparison between the present invention and the traditional method

[0083]

[0084] Furthermore, in terms of fault detection sensitivity, the method based on knowledge graph also showed obvious advantages, with a sensitivity of 95.6%. This means that this method has extremely high detection capabilities when detecting real faults, and can effectively reduce missed reports. The sensitivity of traditional machine learning methods and statistical analysis methods is only 82.7% and 75.4%, respectively, which makes it difficult to effectively detect all faults. In addition, although the sensitivity of the non-optimized graph neural network method has certain performance, its sensitivity is still lower than that of the method of the present invention due to the lack of optimization of the energy propagation mechanism. In terms of fault detection specificity, the method based on knowledge graph also showed superiority, with a specificity of 98.3%. This result shows that this method has a strong ability to distinguish between normal and fault states, and the false alarm rate is extremely low. The specificity of traditional machine learning methods and statistical analysis methods is 87.2% and 80.1%, respectively, which is prone to false alarm problems in practical applications, reducing the reliability of the system.

[0085] Furthermore, the photovoltaic power station fault detection method based on knowledge graph of the present invention significantly outperforms the existing technology in terms of accuracy, sensitivity and specificity of fault detection by effectively utilizing the spatiotemporal relationship between devices, combining graph neural network and energy propagation mechanism. The innovation of this method lies in constructing a knowledge graph to capture the complex relationship between devices, and further improving the detection performance through optimized graph neural network, showing its great potential in practical applications.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A photovoltaic power station fault detection method based on knowledge graph, characterized in that: include, Collect multi-dimensional time series data during the operation of the photovoltaic power station, pre-process the multi-dimensional time series data, and construct a knowledge graph; Based on the spatiotemporal structure and semantic information of the knowledge graph, a graph neural network model is constructed, and node features on the knowledge graph are learned to generate an embedded representation of the node; According to the embedded representation of the node, a fault detection model is established, and the photovoltaic power station fault is identified by link prediction technology; The fault detection results of the fault detection model are optimized and confidence evaluated through an energy propagation mechanism to complete the fault detection of the photovoltaic power station; The multi-dimensional time series data includes photovoltaic power station operation data, photovoltaic module status data and inverter status data; the embedded representation includes a first embedded representation and a second embedded representation; the energy propagation mechanism includes the following steps: Based on the second embedding representation of the node, construct an energy function to calculate the energy score of the node; Use the node connection relationship of the knowledge graph to build an adjacency matrix and define the energy propagation path between nodes; At the same time, the energy vector of the node is initialized, and energy propagation is performed based on the adjacency matrix; The energy score of the node is updated through several iterative propagations to determine the fault status of the node; According to the energy score of the node, a confidence evaluation is performed on the fault detection result of the fault detection model, the faulty node is identified, and the fault detection result is output; The specific formula of the energy function is as follows: ； in, is the energy score of the second embedding representation of node v, is the number of fault categories, is the weight vector corresponding to category a, is the second embedding representation of node v at time t; The specific formula for energy propagation is as follows: ； in, is the energy score of node v at layer k, is the adjacency matrix value between node u and node v, , and The parameters that control the energy propagation, is the neighbor set of node v; Also includes, When the embedding representation of the node is within the preset range, monitoring continues and the state of this node is recorded as normal; When the embedding representation of a node deviates from the preset range, the node is marked as a potential fault point for fault detection; The fault detection includes: if a potential fault point is detected, link prediction technology is used for analysis, and confidence assessment is performed through an energy propagation mechanism; if the probability of a faulty link exceeds a preset threshold, the link is confirmed as a faulty link; if several adjacent links are marked as faulty links at the same time, the device group connected to the link is marked as a faulty area to identify the fault type; if the fault feature is a known fault type, it is classified into the corresponding fault type; if the fault feature pattern is an unseen fault, it is marked as an unknown fault type and a manual review process is initiated; The confidence evaluation includes: when the energy score of the node When the energy score of the node is greater than the confidence threshold, it is reported as a high confidence fault. When it is less than the confidence threshold, the fault will not be reported temporarily and the node will be added to the monitoring list; The construction process of the knowledge graph includes the following steps: Performing feature extraction and time-varying entity extraction on the multi-dimensional time series data to identify equipment status and operating conditions; The extracted feature maps are used as nodes and relationships in the knowledge graph to calculate the time-varying relationship strength between entities and select key entity pairs. The specific formula is as follows: ； in, is the coefficient of the time-varying relationship strength between entity i and entity j at time t, is the mutual relationship coefficient between entity i and entity j at time t, is the transfer entropy between entity i and entity j at time t, and is the balance parameter; The coefficient of the time-varying relationship strength is used as the weight of the edge to construct a knowledge graph, and the prior knowledge is used to supplement the knowledge graph with spatiotemporal structure and semantic information to obtain a tensor G with a dimension of M×M×L as the input of the graph neural network model; Among them, M is the total number of entities, and L is the time step; The steps of constructing the graph neural network model include the following steps: Design graph convolution operations suitable for knowledge graph structures, aggregate node and neighbor node features in the knowledge graph structure, and generate the first embedding representation of the node; Based on the first embedding representation, an attention mechanism is introduced to enhance the weight of node features in the embedding representation; Through multi-layer graph convolution, combined with spatiotemporal features and semantic information, the information of neighboring nodes is aggregated, and the first embedding representation of the node is processed through a nonlinear activation function to generate the second embedding representation of the node. The specific formula is as follows: ； in, is the second embedding representation of node v at layer k, is the second embedding representation of node u at layer k-1, is the neighbor set of node v, is the attention weight of node u to node v, is the weight matrix of the kth layer, is the bias vector of the kth layer, is a nonlinear activation function; The link prediction technology is a generative adversarial network method based on deep learning; the construction process of the fault detection model includes the following steps: According to the second embedding representation of the nodes, the link prediction technology is used to identify the faulty links in the photovoltaic power station, and the fault detection model is constructed and trained; The cross entropy loss function is used to optimize the fault detection model parameters through back propagation to minimize the error between the predicted results and the actual annotations. The specific formula is as follows: ； in, The second embedding representation for node v is and the second embedding representation of node u The probability of a failed link, is the label value of the fault link between node u and node v, is the weight matrix of the fault detection model.

2. A photovoltaic power station fault detection system based on knowledge graph, based on the photovoltaic power station fault detection method based on knowledge graph according to claim 1, characterized in that: include, A collection module is used to collect multi-dimensional time series data during the operation of the photovoltaic power station, and pre-process the multi-dimensional time series data to construct a knowledge graph; A construction module, which constructs a graph neural network model based on the spatiotemporal structure and semantic information of the knowledge graph, learns the node features on the knowledge graph, and generates an embedded representation of the node; An identification module, which establishes a fault detection model according to the embedded representation of the node and realizes the identification of photovoltaic power station faults through link prediction technology; The optimization module is used to optimize and conduct confidence evaluation on the fault detection results of the fault detection model through an energy propagation mechanism to complete the fault detection of the photovoltaic power station.

3. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the photovoltaic power station fault detection method based on knowledge graph described in claim 1 are implemented.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the photovoltaic power station fault detection method based on knowledge graph described in claim 1 are implemented.

Images