Photovoltaic module insulation test system
Through the photovoltaic module insulation testing system integrating intelligent sensors, physical models, graph neural networks and reinforcement learning modules, the problem of low detection accuracy is solved, efficient and accurate fault diagnosis and positioning is achieved, adapting to complex environments, and the safety and stability of the photovoltaic system are improved.
Patent Information
- Application Number
- CN202510682492.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-26
- Publication Date
- 2025-08-08
AI Technical Summary
The existing photovoltaic module insulation testing methods have problems such as low detection accuracy and great influence on human factors, and it is difficult to capture potential insulation performance hidden dangers in complex and changeable outdoor environments in a timely manner.
The central control module is adopted to combine intelligent sensor modules, physical model modules, data enhancement modules, graph neural network modules and reinforcement learning modules to achieve high-precision fault diagnosis and positioning through real-time data acquisition, theoretical model construction, diversified data generation and dynamic testing strategy optimization.
It significantly improves the accuracy and fault tolerance of photovoltaic module insulation tests, reduces artificial errors, adapts to complex environment changes, improves operation and maintenance efficiency and fault response speed, and reduces the risk of false alarms and missed reports.
Smart Images
Figure CN120446692A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of insulation testing, and in particular to a photovoltaic component insulation testing system. Background Art
[0002] Photovoltaic (PV) modules are the core component of a solar photovoltaic (PV) power generation system. They consist of several solar cells connected in series or parallel and encapsulated within a frame. Their function is to convert solar energy into electricity. PV modules typically consist of glass, EVA film, cells, a backsheet, and a frame. Glass has high light transmittance, allowing maximum sunlight to pass through. EVA film encapsulates and secures the cells, which are crucial for photoelectric conversion. The backsheet provides insulation and waterproofing, while the frame enhances the module's mechanical strength and facilitates installation. Insulation testing of PV modules is essential because during operation, PV systems are exposed to a variety of factors, including light, temperature, humidity, and wind. This can degrade the module's insulation performance. Poor insulation can lead to leakage, which not only causes power loss and reduced power generation efficiency, but can also pose a risk of electric shock, endangering personnel, and even causing serious accidents such as fires. Therefore, regular insulation testing of PV modules can promptly identify changes in insulation performance and potential problems, ensuring the safe, stable, and efficient operation of PV systems.
[0003] In the current photovoltaic industry, testing of PV module insulation performance remains largely traditional and rudimentary. Currently, there are two common testing methods. One relies on manual labor using various testing tools. Workers must manually operate these tools to test the insulation performance of each PV module at the PV power plant site. This process is not only labor-intensive and time-consuming, but also subject to significant human influence. Differences in operator technique and experience can easily lead to biased test results, making it difficult to ensure accurate and reliable results. Another method involves the use of simple sensors. These sensors have relatively limited functionality, providing limited data on PV module insulation performance. Furthermore, their accuracy and stability often fall short of the requirements for accurate insulation performance assessment. The limitations of this simple sensor approach become even more pronounced in the complex and changing outdoor environment, as well as the subtle variations in insulation performance that may occur in PV modules. This makes it difficult to detect potential insulation performance issues promptly and accurately. Summary of the Invention
[0004] In view of the shortcomings of the existing technology, the present invention provides a photovoltaic module insulation testing system, which solves the problem of low detection accuracy in the existing technology.
[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a photovoltaic module insulation testing system, including a central control module, wherein the central control module is respectively connected to an intelligent sensor module, a physical model module, a data enhancement module, a graph neural network module, a reinforcement learning module and a communication module;
[0006] The intelligent sensor module is deployed near the photovoltaic module to collect the voltage V, current I, and temperature T data of the photovoltaic module in real time, and has a built-in edge computing unit to perform local insulation resistance R ins and leakage current I leak The central control module is connected to the intelligent sensor module for coordinating the test process, storing data and performing global analysis; the physical model module is integrated into the central control module, and an electrical characteristic model is constructed based on the material properties, structural parameters and environmental conditions of the photovoltaic module to output the theoretical insulation resistance value R model And the theoretical value of leakage current I model The data enhancement module uses a generative adversarial network (GAN) to generate diversified test data for training the insulation test algorithm. The graph neural network module models photovoltaic components and their electrical connection relationships as a graph structure, and realizes fault diagnosis through feature analysis of nodes and edges. The reinforcement learning module dynamically adjusts the test strategy and optimizes the test parameters based on the Markov decision process. The communication module supports low-power wide area network (LPWAN) or 5G data transmission between the intelligent sensor module and the central control module.
[0007] Preferably, the insulation resistance theoretical value R of the physical model module model Calculated by the following formula:
[0008]
[0009] Where ρ(T) is the temperature-dependent resistivity of the material, L is the component structure length, A is the cross-sectional area, α is the temperature coefficient, and T0 is the reference temperature.
[0010] Preferably, the generative adversarial network GAN of the data enhancement module includes a generator G and a discriminator D, and their loss functions are:
[0011]
[0012] Among them, z is the noise input, x is the real test data, and the generated virtual data covers insulation aging, local short circuit and leakage current abnormality scenarios.
[0013] Preferably, the graph convolution operation of the graph neural network module is defined as:
[0014]
[0015] in, To add a self-connected adjacency matrix, is the degree matrix, H (l) is the l-th layer node feature, W (l) is the trainable weight matrix and σ is the activation function.
[0016] Preferably, the Q value update formula of the reinforcement learning module is:
[0017]
[0018] Among them, s and a are state and action respectively, r is reward, γ is discount factor, α is learning rate, and the goal is to minimize the test time t and maximize the accuracy η, that is, the reward function
[0019] Preferably, the local insulation resistance of the intelligent sensor module is calculated using the improved Ohm's law:
[0020]
[0021] Where k is the temperature compensation coefficient, and the noise is eliminated by sliding average filtering:
[0022]
[0023] Preferably, the user interface module supports remote generation of a test report, the report content including the insulation resistance error ΔR=|R ins -R model | and leakage current deviation ΔI=|I leak -I model |.
[0024] A photovoltaic module insulation testing method specifically comprises the following steps:
[0025] S1. Collect data through the intelligent sensor module and calculate R ins and I leak ;
[0026] S2. Comparison with R ins With R model 、、I leak with I model , if ΔR>R th or ΔI>I th , triggering fault cut-off;
[0027] S3. Locate the faulty component or connection line using a graph neural network module;
[0028] S4. Dynamically adjust the test voltage V based on the reinforcement learning module test and sampling frequency f s, optimize subsequent testing strategies.
[0029] The present invention provides a photovoltaic module insulation testing system. It has the following beneficial effects:
[0030] The present invention provides a photovoltaic module insulation testing system. The system of the present invention integrates physical models, data enhancement, graph neural networks and reinforcement learning technologies to form complementary advantages. The physical model provides theoretical support, data enhancement expands training samples, graph neural networks capture complex associations, and reinforcement learning dynamic optimization strategies significantly improve the system's comprehensive judgment ability and fault tolerance. Local edge computing and cloud-based global management work together, and the intelligent sensor module has built-in data processing capabilities to complete preliminary calculations locally, reduce data transmission volume, and reduce communication delays. The central control module integrates multi-node data and combines physical models with reinforcement learning for global optimization, supporting centralized monitoring and strategy updates for large-scale power stations. The system is new and forms an efficient architecture suitable for large-scale operation and maintenance of distributed photovoltaic systems. The system has the ability to adapt to environmental changes. Environmental variables such as temperature and humidity are introduced into the physical model to correct theoretical values. The intelligent sensor adopts a temperature compensation algorithm to ensure that the test results are not affected by environmental fluctuations. Reinforcement learning self-adjusts the test strategy and dynamically optimizes according to the real-time environmental status to avoid the failure of fixed thresholds under complex working conditions, enhance system robustness, hierarchical and visualized fault diagnosis, local sensors quickly screen anomalies, trigger in-depth analysis of the global graph neural network, clarify the fault type and location, and the user interface integrates real-time data display, historical trend analysis and fault location map to improve operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION
[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0033] like Figure 1 As shown, an embodiment of the present invention provides a photovoltaic module insulation testing system, including a central control module, which is respectively connected to an intelligent sensor module, a physical model module, a data enhancement module, a graph neural network module, a reinforcement learning module and a communication module.
[0034] The intelligent sensor module is deployed near the photovoltaic module to collect the voltage V, current I, and temperature T data of the photovoltaic module in real time, and has a built-in edge computing unit to perform local insulation resistance Rins and leakage current I leak The central control module communicates with the intelligent sensor module to coordinate the test process, store data and perform global analysis; the physical model module is integrated into the central control module, and the electrical characteristic model is constructed based on the material properties, structural parameters and environmental conditions of the photovoltaic module, and the theoretical insulation resistance value R is output. model And the theoretical value of leakage current I model The data enhancement module uses a generative adversarial network (GAN) to generate diverse test data for training the insulation test algorithm. The graph neural network module models photovoltaic modules and their electrical connection relationships as a graph structure, and performs fault diagnosis through feature analysis of nodes and edges. The reinforcement learning module dynamically adjusts the test strategy and optimizes the test parameters based on the Markov decision process. The communication module supports low-power wide area network (LPWAN) or 5G data transmission between the intelligent sensor module and the central control module.
[0035] Theoretical value of insulation resistance R of the physical model module model Calculated by the following formula:
[0036]
[0037] Where ρ(T) is the temperature-dependent resistivity of the material, L is the component structure length, A is the cross-sectional area, α is the temperature coefficient, and T0 is the reference temperature.
[0038] The generative adversarial network GAN of the data enhancement module includes a generator G and a discriminator D, and their loss functions are:
[0039]
[0040] Among them, z is the noise input, x is the real test data, and the generated virtual data covers insulation aging, local short circuit and leakage current abnormality scenarios.
[0041] The graph convolution operation of the graph neural network module is defined as:
[0042]
[0043] in, To add a self-connected adjacency matrix, is the degree matrix, H (l) is the l-th layer node feature, W (l) is the trainable weight matrix and σ is the activation function.
[0044] The Q-value update formula of the reinforcement learning module is:
[0045]
[0046] Among them, s and a are state and action respectively, r is reward, γ is discount factor, α is learning rate, and the goal is to minimize the test time t and maximize the accuracy η, that is, the reward function
[0047] The local insulation resistance of the smart sensor module is calculated using the modified Ohm's law:
[0048]
[0049] Where k is the temperature compensation coefficient, and the noise is eliminated by sliding average filtering:
[0050]
[0051] The user interface module supports remote generation of test reports, which include insulation resistance error ΔR=|R ins -R model | and leakage current deviation ΔI=|I leak -I model |.
[0052] A photovoltaic module insulation testing method specifically comprises the following steps:
[0053] S1. Collect data through the intelligent sensor module and calculate R ins and I leak ;
[0054] S2. Comparison with R ins With R model 、、I leak with I model , if ΔR>R th or ΔI>I th , triggering fault cut-off;
[0055] S3. Locate the faulty component or connection line using a graph neural network module;
[0056] S4. Dynamically adjust the test voltage V based on the reinforcement learning module test and sampling frequency f s , optimize subsequent testing strategies.
[0057] Specifically, the system of the present invention integrates physical models, data augmentation, graph neural networks, and reinforcement learning technologies to form complementary advantages:
[0058] Physical models drive the construction of theoretical models based on material properties and environmental parameters, providing a benchmark for measured data and avoiding the risk of misjudgment caused by reliance on a single data set. Data augmentation expands the boundaries by simulating extreme operating conditions and rare faults through generative adversarial networks (GANs), effectively covering scenarios difficult to reach with traditional testing and enhancing the algorithm's ability to identify complex anomalies. Graph neural networks accurately locate the electrical connections between components as a graph structure. Combining feature analysis of nodes and edges, they break through the limitations of traditional single-point detection and accurately distinguish between faulty components and connecting lines. Reinforcement learning dynamic optimization adjusts test parameters in real time based on environmental conditions, balancing test speed and accuracy and avoiding the performance degradation of fixed strategies in changing scenarios. The synergistic effect of physical models provides theoretical support, data augmentation expands training samples, graph neural networks capture complex relationships, and reinforcement learning dynamically optimizes strategies. The combination of these four significantly improves the system's comprehensive judgment capabilities and fault tolerance.
[0059] This invention leverages the built-in data processing capabilities of local edge computing intelligent sensor modules to perform preliminary calculations of insulation resistance and leakage current locally, reducing the amount of raw data transmitted, lowering communication latency, and ensuring real-time feedback on key indicators. The cloud-based global management central control module integrates multi-node data, combines physical models with reinforcement learning for global optimization, and supports centralized monitoring and policy updates for large-scale power stations. The edge quickly responds to anomalies, while the cloud coordinates resource allocation, forming an efficient "edge real-time processing + cloud-based in-depth analysis" architecture suitable for large-scale operation and maintenance of distributed photovoltaic systems.
[0060] Environmental variables such as temperature and humidity are incorporated into the physical model to modify theoretical values. The intelligent sensor employs a temperature compensation algorithm to ensure that test results are not affected by environmental fluctuations. The test strategy is dynamically optimized based on real-time environmental conditions (such as sudden temperature changes and voltage fluctuations), preventing the failure of fixed thresholds under complex operating conditions. The system maintains stable output even in harsh environments such as high temperature, high humidity, and large diurnal temperature swings, reducing the risk of false alarms or missed alarms due to insufficient environmental adaptability.
[0061] Local sensors quickly screen for anomalies, triggering in-depth analysis using a global graph neural network to pinpoint the fault type (e.g., component aging, connection failure) and location, reducing manual troubleshooting costs. The user interface integrates real-time data display, historical trend analysis, and a fault location map, allowing operations and maintenance personnel to quickly locate problematic components and develop targeted maintenance plans. The complete automation process, from anomaly detection to fault location, significantly shortens fault response time and minimizes power plant downtime losses.
[0062] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.
Claims
1. A photovoltaic module insulation test system, including a central control module, characterized in that: The central control module is respectively connected to the intelligent sensor module, the physical model module, the data enhancement module, the graph neural network module, the reinforcement learning module and the communication module; The intelligent sensor module is deployed near the photovoltaic module to collect the voltage V, current I, and temperature T data of the photovoltaic module in real time, and has a built-in edge computing unit to perform local insulation resistance R ins and leakage current I leak The central control module is connected to the intelligent sensor module to coordinate the test process, store data and perform global analysis. The physical model module is integrated into the central control module to build an electrical characteristic model based on the material properties, structural parameters and environmental conditions of the photovoltaic module, and output the theoretical insulation resistance value R model And the theoretical value of leakage current I model The data enhancement module uses a generative adversarial network (GAN) to generate diverse test data for training the insulation test algorithm. The graph neural network module models photovoltaic components and their electrical connection relationships as a graph structure, and implements fault diagnosis through feature analysis of nodes and edges. The reinforcement learning module dynamically adjusts the test strategy based on the Markov decision process and optimizes the test parameters. The communication module supports low-power wide area network (LPWAN) or 5G data transmission between the smart sensor module and the central control module.
2. A photovoltaic module insulation testing system according to claim 1, characterized in that: The theoretical value of insulation resistance R of the physical model module model Calculated by the following formula: Where ρ(T) is the temperature-dependent resistivity of the material, L is the component structure length, A is the cross-sectional area, α is the temperature coefficient, and T0 is the reference temperature.
3. The photovoltaic module insulation testing system according to claim 1, characterized in that: The generative adversarial network GAN of the data enhancement module includes a generator G and a discriminator D, and their loss functions are: Among them, z is the noise input, x is the real test data, and the generated virtual data covers insulation aging, local short circuit and leakage current abnormality scenarios.
4. The photovoltaic module insulation testing system according to claim 1, characterized in that: The graph convolution operation of the graph neural network module is defined as: in, To add a self-connected adjacency matrix, is the degree matrix, H (l) is the l-th layer node feature, W (l) is the trainable weight matrix and σ is the activation function.
5. The photovoltaic module insulation testing system according to claim 1, characterized in that: The Q value update formula of the reinforcement learning module is: Among them, s and a are state and action respectively, r is reward, γ is discount factor, α is learning rate, and the goal is to minimize the test time t and maximize the accuracy η, that is, the reward function 6. The photovoltaic module insulation testing system according to claim 1, characterized in that: The local insulation resistance of the smart sensor module is calculated using the improved Ohm's law: Where k is the temperature compensation coefficient, and the noise is eliminated by sliding average filtering:
7. The photovoltaic module insulation testing system according to claim 1, characterized in that: The user interface module supports remote generation of test reports, and the report content includes insulation resistance error ΔR=|R ins -R model | and leakage current deviation ΔI=|I leak -I model |.
8. A photovoltaic module insulation testing method according to any one of claims 1 to 7, characterized in that: The specific steps include: S1. Collect data through the intelligent sensor module and calculate R ins and I leak ; S2. Comparison with R ins With R model 、、I leak with I model , if ΔR>R th or ΔI>I th , triggering fault cut-off; S3. Locate the faulty component or connection line using a graph neural network module; S4. Dynamically adjust the test voltage V based on the reinforcement learning module test and sampling frequency f s , optimize subsequent testing strategies.