A photovoltaic power station distribution network fault detection method and device based on digital twinning and a storage medium
By generating geometric, physical, behavioral, and knowledge models of photovoltaic power plant distribution networks using digital twin technology, and combining graph convolutional neural networks and a three-layer memory system, the problems of data silos and insufficient model accuracy in traditional photovoltaic power plant distribution network monitoring systems are solved, enabling efficient fault detection and equipment management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID JIBEI ELECTRIC POWER COMPANY LIMITED CHENGDE POWER SUPPLY
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-21
AI Technical Summary
Traditional photovoltaic power plant distribution network monitoring systems suffer from low data granularity and severe data silos, making it impossible to achieve panoramic status perception. The simulation models lack accuracy and are difficult to simulate dynamic operating condition changes caused by new energy fluctuations and flexible loads. Fault early warning and scheduling optimization lack accurate data support, and faults caused by equipment aging and meteorological reasons are difficult to identify in advance.
A digital twin-based method for detecting distribution network faults in photovoltaic power plants is adopted. By collecting hardware equipment parameters and operating data, geometric, physical, behavioral, and knowledge models are generated. A graph convolutional neural network is used to construct the knowledge model, detect faults in real time, and render and display them. A three-layer memory system is combined to optimize task scheduling.
It improves the real-time performance and accuracy of fault detection, enhances the efficiency and performance of equipment management, and realizes panoramic status perception and dynamic operating condition simulation of photovoltaic power station distribution network.
Smart Images
Figure CN122436985A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence, digital twins and new energy fault detection integration technology, specifically to a method, device and storage medium for detecting distribution network faults in photovoltaic power plants based on digital twins. Background Technology
[0002] The current distributed photovoltaic (PV) power station distribution network exhibits typical characteristics such as strong source-load randomness, complex grid structure, diverse equipment types, and variable operating conditions. The limitations of traditional management models relying on manual inspections, fixed threshold adjustments, and post-fault emergency repairs are becoming increasingly apparent. On the one hand, traditional monitoring systems suffer from low data granularity and severe data silos, failing to achieve a comprehensive understanding of the overall status of distribution network equipment, lines, and power flow. On the other hand, the accuracy of distributed PV power station distribution network simulation models is insufficient, making it difficult to simulate dynamic changes in operating conditions caused by new energy fluctuations and flexible load switching. Furthermore, the real-time performance of twin models is inadequate, resulting in a lack of accurate data support for fault warnings and scheduling optimization. Simultaneously, the full lifecycle management system for distributed PV power station distribution network equipment is incomplete, making it difficult to identify problems caused by equipment aging and weather conditions in advance. Summary of the Invention
[0003] In view of one or more technical defects in the prior art, the present invention proposes the following technical solution.
[0004] A method for detecting distribution network faults in photovoltaic power plants based on digital twins, the method comprising: The data acquisition steps involve obtaining hardware equipment parameters from the photovoltaic power station's distribution network and using sensors to collect operational data from the distribution network. The hardware equipment parameters include the distribution network topology, electrical connections, equipment parameters, and geographic information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operational data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather conditions, and load fluctuations. Twin step: Generate a twin model of the photovoltaic power station distribution network based on the hardware device parameters and operating data. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The detection step involves using the knowledge model to detect faults in the photovoltaic power station's distribution network in real time based on operational data, and mapping the equipment and location corresponding to the faults to the corresponding locations in the geometric and physical models. The rendering step involves rendering the corresponding device and location of the fault with the appropriate color based on the fault level, and then displaying it in the twin model.
[0005] Furthermore, in the acquisition step, the hardware device parameters and operating data are preprocessed, including deduplication and outlier repair.
[0006] Furthermore, the twinning steps involve: using a first twinning task to construct a geometric and physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twinning task to construct a behavioral model of the photovoltaic power station distribution network based on the operational data; and using a third twinning task to construct a knowledge model based on the hardware device parameters, historical operational data of the photovoltaic power station distribution network, distribution network operation and maintenance rules, fault cases, and expert experience. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage magnitudes in transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
[0007] Furthermore, the first twin task, the second twin task, and the third twin task are matched with the optimal memory in the memory system for processing based on the type of the task.
[0008] Furthermore, the memory system comprises three layers: local memory, edge memory, and cloud memory, with the parameters for each layer defined as: R k =(C k B k L k P k U k ), k=0, 1, 2, representing local memory, edge memory, and cloud memory respectively, C k B k L k P k U k These represent capacity, bandwidth, access latency, power consumption, and current utilization, respectively. The first twin task, the second twin task, and the third twin task are represented as: T i =(m i b i t i dl ), i = 0, 1, 2, representing the first twin task, the second twin task, and the third twin task, respectively, m i b i t i dl These represent the memory requirements, bandwidth requirements, and task deadlines for the respective tasks. Calculate the matching values between the first twin task, the second twin task, and the third twin task and the three layers of memory: ; The corresponding task is scheduled to the memory layer with the largest matching value. If the Uk of the memory layer with the largest matching value is greater than the first threshold, the task is scheduled to the memory layer with the second largest matching value. in, 0 represents the attenuation coefficient, typically taken as 0.5, L max L min These represent the system's maximum and minimum delays, respectively. Specifically, when calculating the matching value of the first twin task, W1, W2, and W3 are 0.7, 0.2, and 0.1, respectively; when calculating the matching value of the second twin task, W1, W2, and W3 are 0.2, 0.3, and 0.5, respectively; and when calculating the matching value of the third twin task, W1, W2, and W3 are 0.1, 0.2, and 0.7, respectively.
[0009] Furthermore, the knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are defined as follows: if two hardware devices are connected, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
[0010] Furthermore, the feature vector of a node in the graph is a vector composed of the parameters and operating data of the hardware device corresponding to that node.
[0011] Furthermore, the initial weights of each edge are calculated as follows: ; Then initialize the weights of each side. Normalization is performed to obtain the weight values of each edge. Where i and j represent the i-th and j-th nodes, respectively. Let i and j represent the feature vectors of the i-th and j-th nodes, respectively. () indicates the calculation of cosine similarity. Let i represent the set of neighboring nodes. This represents the computation of converting a vector into a scalar.
[0012] This invention also proposes a photovoltaic power plant distribution network fault detection device based on digital twins, the device comprising: The data acquisition unit obtains hardware equipment parameters in the photovoltaic power station distribution network and uses sensors to collect operational data of the photovoltaic power station distribution network. The hardware equipment parameters include distribution network topology, electrical connection relationships, equipment parameters, and geographical information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operational data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather, and load fluctuation. Twin unit: Based on the hardware device parameters and operating data, a twin model of the photovoltaic power station distribution network is generated. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The detection unit uses the knowledge model to detect faults in the distribution network of the photovoltaic power station in real time based on the operating data, and maps the equipment and location corresponding to the fault to the corresponding locations in the geometric model and physical model. The rendering unit renders the corresponding device and location based on the fault level and displays it in the twin model.
[0013] Furthermore, in the acquisition unit, the hardware device parameters and operating data are preprocessed, including deduplication and outlier repair.
[0014] Furthermore, the operation of the twin unit is as follows: using a first twin task to construct a geometric model and a physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twin task to construct a behavioral model of the photovoltaic power station distribution network based on the operational data; and using a third twin task to construct a knowledge model based on the hardware device parameters and the historical operational data, distribution network operation and maintenance rules, fault cases, and expert experience of the photovoltaic power station distribution network. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage magnitudes in the transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
[0015] Furthermore, the first twin task, the second twin task, and the third twin task are matched with the optimal memory in the memory system for processing based on the type of the task.
[0016] Furthermore, the memory system comprises three layers: local memory, edge memory, and cloud memory, with the parameters for each layer defined as: R k =(C k B k L k P k Uk ), k=0, 1, 2, representing local memory, edge memory, and cloud memory respectively, C k B k L k P k U k These represent capacity, bandwidth, access latency, power consumption, and current utilization, respectively. The first twin task, the second twin task, and the third twin task are represented as: T i =(m i b i t i dl ), i = 0, 1, 2, representing the first twin task, the second twin task, and the third twin task, respectively, m i b i t i dl These represent the memory requirements, bandwidth requirements, and task deadlines for the respective tasks. Calculate the matching values between the first twin task, the second twin task, and the third twin task and the three layers of memory: ; The corresponding task is scheduled to the memory layer with the largest matching value. If the Uk of the memory layer with the largest matching value is greater than the first threshold, the task is scheduled to the memory layer with the second largest matching value. in, 0 represents the attenuation coefficient, L max L min These represent the system's maximum and minimum delays, respectively. Specifically, when calculating the matching value of the first twin task, W1, W2, and W3 are 0.7, 0.2, and 0.1, respectively; when calculating the matching value of the second twin task, W1, W2, and W3 are 0.2, 0.3, and 0.5, respectively; and when calculating the matching value of the third twin task, W1, W2, and W3 are 0.1, 0.2, and 0.7, respectively.
[0017] Furthermore, the knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are defined as follows: if two hardware devices are connected, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
[0018] Furthermore, the feature vector of a node in the graph is a vector composed of the parameters and operating data of the hardware device corresponding to that node.
[0019] Furthermore, the initial weights of each edge are calculated as follows: ; Then initialize the weights of each side. Normalization is performed to obtain the weight values of each edge. Where i and j represent the i-th and j-th nodes, respectively. Let i and j represent the feature vectors of the i-th and j-th nodes, respectively. () indicates the calculation of cosine similarity. Let i represent the set of neighboring nodes. This represents the computation of converting a vector into a scalar.
[0020] The present invention also proposes a computer-readable storage medium storing computer program code, which, when executed by a computer, performs any of the methods described above.
[0021] The technical advantages of this invention are as follows: This invention provides a method, device, and storage medium for detecting faults in a photovoltaic power station distribution network based on digital twins. The method includes: a data acquisition step S101, acquiring hardware device parameters in the photovoltaic power station distribution network and using sensors to collect operational data of the photovoltaic power station distribution network; the hardware device parameters include distribution network topology, electrical connection relationships, equipment parameters, and geographical information; the hardware devices include at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment; the operational data includes at least: voltage, current, power, and hardware device parameters. Temperature, hardware vibration, environmental weather, load fluctuation; Twin step S102: Generate a twin model of the photovoltaic power station distribution network based on the hardware parameters and operating data. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model; Detection step S103: Use the knowledge model to detect faults in the photovoltaic power station distribution network in real time based on the operating data, and map the equipment and location corresponding to the fault to the corresponding positions in the geometric model and physical model; Rendering step S104: Render the equipment and location corresponding to the fault with the corresponding color according to the fault level and display it in the twin model. This invention proposes to first acquire the hardware equipment parameters of the photovoltaic power station distribution network and use sensors to collect the operational data of the photovoltaic power station distribution network. Then, based on the hardware equipment parameters and operational data, a twin model of the photovoltaic power station distribution network is generated. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The knowledge model is used to detect faults in the photovoltaic power station distribution network in real time based on the operational data, and the equipment and location corresponding to the faults are mapped to the corresponding locations in the geometric and physical models, and then rendered and displayed. When generating the twin model, this invention uses different tasks to generate different models based on the changing patterns of the photovoltaic power station distribution network data. This is because some models do not need to be generated in real time, while the knowledge model used for fault identification generates the original model and requires real-time fault detection. That is, the knowledge model is iterated in real time. Therefore, this invention creatively uses different task types and performs twinning at different memory levels according to the changing patterns of the photovoltaic power station distribution network data, which improves the real-time performance of the twin model and improves the efficiency and performance of fault detection. This is one of the important inventive concepts of this invention. Attached Figure Description
[0022] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.
[0023] Figure 1 This is a flowchart of a photovoltaic power plant distribution network fault detection method based on digital twin according to an embodiment of the present invention.
[0024] Figure 2This is a structural diagram of a photovoltaic power station distribution network fault detection device based on digital twin according to an embodiment of the present invention. Detailed Implementation
[0025] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0027] The theoretical basis of this invention is as follows: Grid line loss is the active power loss during power transmission. The amount of line loss must be compensated for by generating electricity from the generating units. Fossil fuel power generation directly generates carbon emissions, while renewable energy units achieve zero-carbon power supply. The intensity of carbon emissions from line loss is strongly correlated with the grid's power structure, line loss rate, and equipment operating parameters: grids in areas with a high proportion of coal-fired power have higher carbon emission intensity from line loss; older lines and heavily loaded equipment have greater line losses, corresponding to higher carbon emissions. This invention will achieve comprehensive monitoring of carbon emissions from line loss through precise calculation of line loss, dynamic selection of carbon emission factors, and carbon flow allocation. This is one of the key inventive concepts of this invention.
[0028] Figure 1 This invention illustrates a method for detecting distribution network faults in a photovoltaic power plant based on digital twins. The method includes: In step S101, the hardware equipment parameters in the photovoltaic power station distribution network are obtained, and the operation data of the photovoltaic power station distribution network is collected using sensors. The hardware equipment parameters include the distribution network topology, electrical connection relationships, equipment parameters, and geographical information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operation data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather, and load fluctuation. Twin step S102: Generate a twin model of the photovoltaic power station distribution network based on the hardware device parameters and operating data. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. In detection step S103, the knowledge model is used to detect faults in the distribution network of the photovoltaic power station in real time based on the operating data, and the equipment and location corresponding to the faults are mapped to the corresponding locations in the geometric model and physical model. In rendering step S104, the device and location corresponding to the fault are rendered with the corresponding color according to the fault level and then displayed in the twin model.
[0029] To address the shortcomings of existing twin models in terms of insufficient real-time performance and difficulty in accurately identifying faults, this invention proposes a method to first acquire the hardware equipment parameters of the photovoltaic power station's distribution network and then use sensors to collect operational data from the distribution network. Based on these hardware parameters and operational data, a twin model of the photovoltaic power station's distribution network is generated. This twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The knowledge model is used to detect faults in the photovoltaic power station's distribution network in real time based on the operational data, and the corresponding equipment and location of the fault are mapped to the corresponding locations in the geometric and physical models before rendering and display. This invention, in generating the twin model, [further details about the invention are needed]. Based on the changing patterns of photovoltaic power plant distribution network data, different tasks are used to generate different models. This is because some models do not need to be generated in real time. For example, physical models can be generated at certain intervals, such as 10 minutes, while behavioral models require shorter intervals, such as 2 minutes. However, the knowledge model used for fault identification generates the original model, which requires real-time fault detection. That is, the knowledge model is iterated in real time. Therefore, this invention creatively adopts different task types based on the changing patterns of photovoltaic power plant distribution network data and performs twinning at different memory levels, thereby improving the real-time performance of the twinning model and enhancing the efficiency and performance of fault detection. This is one of the important inventive concepts of this invention.
[0030] In one embodiment, in the data acquisition step S101, the hardware device parameters and operating data are preprocessed, including deduplication and outlier repair. Data preprocessing can improve the accuracy of subsequent model generation and fault detection.
[0031] In one embodiment, the operation of the twin step S102 is as follows: using a first twin task to construct a geometric model and a physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twin task to construct a behavioral model of the photovoltaic power station distribution network based on the operating data; and using a third twin task to construct a knowledge model based on the hardware device parameters and the historical operating data, distribution network operation and maintenance rules, fault cases, and expert experience of the photovoltaic power station distribution network. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage magnitudes in the transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
[0032] In one embodiment, the first twin task, the second twin task, and the third twin task are processed by matching the optimal memory in the memory system according to the type of the task.
[0033] A key inventive concept of this invention is to generate different models using different tasks based on the changing patterns of photovoltaic power plant distribution network data. This is because some models do not require real-time generation; for example, physical models can be generated at intervals of a certain time, such as 10 minutes, while behavioral models require shorter intervals, such as 2 minutes. However, knowledge models used for fault identification generate original models that require real-time fault detection, meaning that knowledge models are iterated in real time. Therefore, this invention creatively employs different task types based on the changing patterns of photovoltaic power plant distribution network data.
[0034] However, physical models generate a large amount of data and consume a lot of memory, while knowledge models are iterative in real time and have the highest time limit requirements. Behavioral models have lower time limit requirements than knowledge models and lower memory requirements than physical models. Therefore, this invention proposes a memory-task matching method, as follows: First, the memory system comprises three layers: local memory, edge memory, and cloud memory. The parameters for each layer are defined as: R k =(C k B k L k U k ), k=0, 1, 2, representing local memory, edge memory, and cloud memory respectively, C k B k L k U k These represent capacity, bandwidth, access latency, and current utilization, respectively. Secondly, the first twin task, the second twin task, and the third twin task are represented as: T i =(m i b i t i dl ), i = 0, 1, 2, representing the first twin task, the second twin task, and the third twin task, respectively, m i b i t i dl These represent the memory requirements, bandwidth requirements, and task deadlines for the respective tasks. Calculate the matching values between the first twin task, the second twin task, and the third twin task and the three layers of memory: ; The corresponding task is scheduled to the memory layer with the largest matching value. If the memory layer with the largest matching value has a U value... k If the value is greater than the first threshold (e.g., 0.85), the task will be scheduled to the memory layer with the second largest matching value. in, 0 represents the attenuation coefficient, typically taken as 0.5, L max L min These represent the system's maximum and minimum delays, respectively.
[0035] The matching value calculation method of this invention can be used to process three different tasks in the appropriate memory layer, which improves the speed of twins. Furthermore, by setting a threshold, it can avoid the lack of memory resources available for real-time tasks, which is one of the important inventive points of this invention.
[0036] In this invention, based on the changing patterns of photovoltaic power plant distribution network data, different twin tasks (i.e., the first, second, and third tasks) are employed. Since each task has different requirements for memory, time limits, and bandwidth, the weights of each task differ when calculating the matching value. A recommended set of values is as follows: when calculating the matching value of the first twin task, W1, W2, and W3 are 0.7, 0.2, and 0.1, respectively; when calculating the matching value of the second twin task, W1, W2, and W3 are 0.2, 0.3, and 0.5, respectively; and when calculating the matching value of the third twin task, W1, W2, and W3 are 0.1, 0.2, and 0.7, respectively. Furthermore, those skilled in the art can adjust the corresponding weight values based on historical data.
[0037] In one embodiment, the knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are defined as follows: if two hardware devices are connected, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
[0038] In one embodiment, the feature vector of a node in the graph is a vector composed of the parameters and operating data of the hardware device corresponding to that node.
[0039] Furthermore, the initial weights of each edge are calculated as follows: ; Then initialize the weights of each side. Normalization is performed to obtain the weight values of each edge. Where i and j represent the i-th and j-th nodes, respectively. Let i and j represent the feature vectors of the i-th and j-th nodes, respectively. () indicates the calculation of cosine similarity. Let i represent the set of neighboring nodes. This represents the computation of converting a vector into a scalar.
[0040] In this invention, in order to improve the accuracy of fault detection, a knowledge model based on graph neural convolutional networks is proposed. In order to improve the correctness of reasoning, the initial values of edge weights are modified based on the fault nodes, that is, the edges where the fault nodes are located are given larger weight values. This is in line with the distribution network rules of photovoltaic power plants, thereby improving the accuracy and efficiency of fault detection. This is one of the important inventive concepts of this invention.
[0041] Figure 2 This invention illustrates a photovoltaic power plant distribution network fault detection device based on digital twins, the device comprising: The acquisition unit 201 acquires hardware equipment parameters in the photovoltaic power station distribution network and uses sensors to collect operational data of the photovoltaic power station distribution network. The hardware equipment parameters include distribution network topology, electrical connection relationships, equipment parameters, and geographical information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operational data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather, and load fluctuation. Twin Unit 202: Generates a twin model of the photovoltaic power station distribution network based on the hardware device parameters and operating data. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The detection unit 203 uses the knowledge model to detect faults in the photovoltaic power station distribution network in real time based on the operating data, and maps the equipment and location corresponding to the fault to the corresponding locations in the geometric model and physical model; The rendering unit 204 renders the device and location corresponding to the fault with the corresponding color according to the fault level and then displays it in the twin model.
[0042] To address the shortcomings of existing twin models in terms of insufficient real-time performance and difficulty in accurately identifying faults, this invention proposes a method to first acquire the hardware equipment parameters of the photovoltaic power station's distribution network and then use sensors to collect operational data from the distribution network. Based on these hardware parameters and operational data, a twin model of the photovoltaic power station's distribution network is generated. This twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The knowledge model is used to detect faults in the photovoltaic power station's distribution network in real time based on the operational data, and the corresponding equipment and location of the fault are mapped to the corresponding locations in the geometric and physical models before rendering and display. This invention, in generating the twin model, [further details about the invention are needed]. Based on the changing patterns of photovoltaic power plant distribution network data, different tasks are used to generate different models. This is because some models do not need to be generated in real time. For example, physical models can be generated at certain intervals, such as 10 minutes, while behavioral models require shorter intervals, such as 2 minutes. However, the knowledge model used for fault identification generates the original model, which requires real-time fault detection. That is, the knowledge model is iterated in real time. Therefore, this invention creatively adopts different task types based on the changing patterns of photovoltaic power plant distribution network data and performs twinning at different memory levels, thereby improving the real-time performance of the twinning model and enhancing the efficiency and performance of fault detection. This is one of the important inventive concepts of this invention.
[0043] In one embodiment, the acquisition unit 201 performs preprocessing on hardware device parameters and operational data, including deduplication and outlier repair. Data preprocessing can improve the accuracy of subsequent model generation and fault detection.
[0044] In one embodiment, the operation of the twin unit 202 is as follows: using a first twin task to construct a geometric model and a physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twin task to construct a behavioral model of the photovoltaic power station distribution network based on the operational data; and using a third twin task to construct a knowledge model based on the hardware device parameters and the historical operational data, distribution network operation and maintenance rules, fault cases, and expert experience of the photovoltaic power station distribution network. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage magnitudes in the transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
[0045] In one embodiment, the first twin task, the second twin task, and the third twin task are processed by matching the optimal memory in the memory system according to the type of the task.
[0046] A key inventive concept of this invention is to generate different models using different tasks based on the changing patterns of photovoltaic power plant distribution network data. This is because some models do not require real-time generation; for example, physical models can be generated at intervals of a certain time, such as 10 minutes, while behavioral models require shorter intervals, such as 2 minutes. However, knowledge models used for fault identification generate original models that require real-time fault detection, meaning that knowledge models are iterated in real time. Therefore, this invention creatively employs different task types based on the changing patterns of photovoltaic power plant distribution network data.
[0047] However, physical models generate a large amount of data and consume a lot of memory, while knowledge models are iterative in real time and have the highest time limit requirements. Behavioral models have lower time limit requirements than knowledge models and lower memory requirements than physical models. Therefore, this invention proposes a memory-task matching method, as follows: First, the memory system comprises three layers: local memory, edge memory, and cloud memory. The parameters for each layer are defined as: R k =(C k B k L k U k ), k=0, 1, 2, representing local memory, edge memory, and cloud memory respectively, C k B k L k U k These represent capacity, bandwidth, access latency, and current utilization, respectively. Secondly, the first twin task, the second twin task, and the third twin task are represented as: T i =(m i b i t i dl ), i = 0, 1, 2, representing the first twin task, the second twin task, and the third twin task, respectively, m i b i t i dl These represent the memory requirements, bandwidth requirements, and task deadlines for the respective tasks. Calculate the matching values between the first twin task, the second twin task, and the third twin task and the three layers of memory: ; The corresponding task is scheduled to the memory layer with the largest matching value. If the memory layer with the largest matching value has a U value... k If the value is greater than the first threshold (e.g., 0.85), the task will be scheduled to the memory layer with the second largest matching value. in, 0 represents the attenuation coefficient, typically taken as 0.5, Lmax L min These represent the system's maximum and minimum delays, respectively.
[0048] The matching value calculation method of this invention can be used to process three different tasks in the appropriate memory layer, which improves the speed of twins. Furthermore, by setting a threshold, it can avoid the lack of memory resources available for real-time tasks, which is one of the important inventive points of this invention.
[0049] In this invention, based on the changing patterns of photovoltaic power plant distribution network data, different twin tasks (i.e., the first, second, and third tasks) are employed. Since each task has different requirements for memory, time limits, and bandwidth, the weights of each task differ when calculating the matching value. A recommended set of values is as follows: when calculating the matching value of the first twin task, W1, W2, and W3 are 0.7, 0.2, and 0.1, respectively; when calculating the matching value of the second twin task, W1, W2, and W3 are 0.2, 0.3, and 0.5, respectively; and when calculating the matching value of the third twin task, W1, W2, and W3 are 0.1, 0.2, and 0.7, respectively. Furthermore, those skilled in the art can adjust the corresponding weight values based on historical data.
[0050] In one embodiment, the knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are defined as follows: if two hardware devices are connected, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
[0051] In one embodiment, the feature vector of a node in the graph is a vector composed of the parameters and operating data of the hardware device corresponding to that node.
[0052] Furthermore, the initial weights of each edge are calculated as follows: ; Then initialize the weights of each side. Normalization is performed to obtain the weight values of each edge. Where i and j represent the i-th and j-th nodes, respectively. Let i and j represent the feature vectors of the i-th and j-th nodes, respectively. () indicates the calculation of cosine similarity. Let i represent the set of neighboring nodes. This represents the computation of converting a vector into a scalar.
[0053] In this invention, in order to improve the accuracy of fault detection, a knowledge model based on graph neural convolutional networks is proposed. In order to improve the correctness of reasoning, the initial values of edge weights are modified based on the fault nodes, that is, the edges where the fault nodes are located are given larger weight values. This is in line with the distribution network rules of photovoltaic power plants, thereby improving the accuracy and efficiency of fault detection. This is one of the important inventive concepts of this invention.
[0054] One embodiment of the present invention provides a computer storage medium storing a computer program. When the computer program on the computer storage medium is executed by a processor, the above-described method is implemented. The computer storage medium may be a hard disk, DVD, CD, flash memory, or other storage device.
[0055] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.
[0056] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the apparatus described in various embodiments or some parts of the embodiments of this application.
[0057] Finally, it should be noted that the above embodiments are for illustration only and not for limiting the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention without departing from the spirit and scope of the present invention. Any modifications or partial substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for detecting distribution network faults in photovoltaic power plants based on digital twins, characterized in that, The method includes: The data acquisition steps involve obtaining hardware equipment parameters from the photovoltaic power station's distribution network and using sensors to collect operational data from the distribution network. The hardware equipment parameters include the distribution network topology, electrical connections, equipment parameters, and geographic information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operational data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather conditions, and load fluctuations. Twin step: Generate a twin model of the photovoltaic power station distribution network based on the hardware device parameters and operating data. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The detection step involves using the knowledge model to detect faults in the photovoltaic power station's distribution network in real time based on operational data, and mapping the equipment and location corresponding to the faults to the corresponding locations in the geometric and physical models. The rendering step involves rendering the corresponding device and location of the fault with the appropriate color based on the fault level, and then displaying it in the twin model.
2. The method according to claim 1, characterized in that, In the acquisition step, the hardware device parameters and operating data are preprocessed, including deduplication and outlier repair.
3. The method according to claim 2, characterized in that, The twinning process involves: using a first twinning task to construct a geometric and physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twinning task to construct a behavioral model of the photovoltaic power station distribution network based on the operational data; and using a third twinning task to construct a knowledge model based on the hardware device parameters, historical operational data of the photovoltaic power station distribution network, distribution network operation and maintenance rules, fault cases, and expert experience. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage in transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
4. The method according to claim 3, characterized in that, The knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are the hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are as follows: if there is a connection between two hardware devices, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
5. The method according to claim 4, characterized in that, The feature vector of a node in the diagram is a vector composed of the parameters and operating data of the hardware device corresponding to that node.
6. A photovoltaic power station distribution network fault detection device based on digital twin, characterized in that, The device includes: The data acquisition unit obtains hardware equipment parameters in the photovoltaic power station distribution network and uses sensors to collect operational data of the photovoltaic power station distribution network. The hardware equipment parameters include distribution network topology, electrical connection relationships, equipment parameters, and geographical information. The hardware equipment includes at least: distribution transformers, switching stations, ring main units, transmission lines, distributed photovoltaic power stations, energy storage devices, charging piles, and terminal electrical equipment. The operational data includes at least: voltage, current, power, hardware equipment temperature, hardware equipment vibration, environmental weather, and load fluctuation. Twin unit: Based on the hardware device parameters and operating data, a twin model of the photovoltaic power station distribution network is generated. The twin model includes a geometric model, a physical model, a behavioral model, and a knowledge model. The detection unit uses the knowledge model to detect faults in the distribution network of the photovoltaic power station in real time based on the operating data, and maps the equipment and location corresponding to the fault to the corresponding locations in the geometric model and physical model. The rendering unit renders the corresponding device and location based on the fault level and displays it in the twin model.
7. The apparatus according to claim 6, characterized in that, In the acquisition unit, hardware device parameters and operating data are preprocessed, including deduplication and outlier repair.
8. The apparatus according to claim 7, characterized in that, The operation of the twin unit is as follows: using a first twin task to construct a geometric model and a physical model of the photovoltaic power station distribution network based on the hardware device parameters; using a second twin task to construct a behavioral model of the photovoltaic power station distribution network based on the operational data; and using a third twin task to construct a knowledge model based on the hardware device parameters, historical operational data of the photovoltaic power station distribution network, distribution network operation and maintenance rules, fault cases, and expert experience. The geometric model is used to display the appearance, geographical location, and network topology of the hardware device; the physical model is used to simulate the electrical characteristics, power flow transmission patterns, and equipment loss characteristics of the hardware device; and the behavioral model is used to describe load switching, the output power of each photovoltaic power station, the current and voltage magnitudes in transmission lines, and the real-time status of energy storage devices, charging piles, and terminal electrical equipment.
9. The apparatus according to claim 8, characterized in that, The knowledge model is constructed based on a graph convolutional neural network. The graph of the graph convolutional neural network is composed of the following: the nodes in the graph are the hardware devices in the distribution network of the photovoltaic power station, and the edges in the graph are as follows: if there is a connection between two hardware devices, then there is an edge between the two hardware devices; otherwise, there is no edge between the two hardware devices.
10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1-5.