A photovoltaic power station remote monitoring and fault diagnosis system based on Internet of Things

Abstract

The application discloses a photovoltaic power station remote monitoring and fault diagnosis system based on Internet of Things, and relates to the photovoltaic monitoring field, comprising: a power generation preliminary judgment module, which is used for collecting meteorological data of a target power station area, real-time power generation data of the target power station, and calling historical power generation data, comparing the fluctuation trend of real-time data and historical data, and preliminarily judging whether the current power generation state is abnormal; an image analysis module, which is used for starting a preset image acquisition device when the power generation preliminary judgment module determines that there is an abnormality, acquiring the monitoring image of the photovoltaic panel array of the target power station, and performing visual analysis on each group of photovoltaic panels in the image to extract the cover state features of the surface; the application accurately associates the abnormal fluctuation of power generation data with the shielding of a specific type and specific area of cover, quantifies the influence degree, improves the accuracy and operability of diagnosis, and provides a direct basis for subsequent accurate maintenance.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic monitoring technology, specifically to a remote monitoring and fault diagnosis system for photovoltaic power plants based on the Internet of Things. Background Technology

[0002] As one of the mainstays of renewable energy, photovoltaic (PV) power generation has seen continuous and rapid growth in installed capacity, with power plants becoming increasingly widespread and decentralized on a large scale. With the maturation and widespread adoption of next-generation information technologies such as the Internet of Things (IoT), big data, and artificial intelligence (AI), smart energy management has become an inevitable trend in the industry. In this context, achieving real-time status perception, assessment, and rapid fault location for massive amounts of PV assets has become a key issue in improving the economic benefits of power plants throughout their entire lifecycle and ensuring the stable operation of the power grid.

[0003] Existing photovoltaic power plant monitoring systems primarily focus on remotely collecting electrical parameters such as power generation, voltage, and current, and providing simple over-limit alarms. They lack the ability to conduct in-depth root cause analysis of power generation performance fluctuations. When abnormal power generation data occurs, maintenance personnel struggle to quickly distinguish whether the cause is due to changes in the meteorological environment, natural aging of equipment, or manageable soft faults such as dust accumulation, snow buildup, or shading on the photovoltaic panel surface. This often necessitates dispatching personnel to the site for visual inspection, resulting in delayed response and high costs. Traditional fault diagnosis methods cannot quantify and assess the differentiated impacts of various coverings, such as dust, bird droppings, and shadows, on different photovoltaic panel properties and in different microenvironments. This leads to low diagnostic accuracy and a high false alarm rate, failing to provide effective decision support for precise operation and maintenance. Summary of the Invention

[0004] (a) Technical problems to be solved In view of the above-mentioned shortcomings of the existing technology, the present invention provides a remote monitoring and fault diagnosis system for photovoltaic power plants based on the Internet of Things, which can effectively solve the problems of the existing technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a remote monitoring and fault diagnosis system for photovoltaic power plants based on the Internet of Things, comprising: The power generation preliminary judgment module is used to collect meteorological data of the area where the target power station is located, real-time power generation data of the target power station, and call historical power generation data. By comparing the fluctuation trend of real-time data and historical data, it makes a preliminary judgment on whether there is any abnormality in the current power generation status. The image analysis module is used to activate a preset image acquisition device to acquire monitoring images of the photovoltaic panel array of the target power station when the power generation preliminary judgment module determines that there is an anomaly, and to perform visual analysis on each group of photovoltaic panels in the image to extract the surface covering status features. A data acquisition module is used to synchronously acquire the attribute data of the photovoltaic panel; The coverage analysis module constructs a coverage impact model based on deep learning algorithms. The model takes the characteristics of the cover state and the attribute data of the photovoltaic panel as joint inputs. By analyzing the optical shading and thermal effects of the cover under specific attribute conditions, it outputs the impact coefficient of the current cover state on the photovoltaic panel power generation efficiency. The fault matching module is used to receive the preliminary judgment of abnormal fluctuation trend characteristics and influence coefficients, and pre-store the fault knowledge base. The fault knowledge base contains a variety of typical fault modes. Each fault mode is associated with a specific fluctuation trend characteristic range and influence coefficient threshold. The current abnormal fluctuation trend characteristics and influence coefficients are combined and matched with the modes in the fault knowledge base to diagnose the fault. The alarm triggering module is used to trigger alarm behaviors of the corresponding level based on the fault items matched by the fault matching module.

[0006] Furthermore, the process for determining the power generation status in the initial power generation judgment module is as follows: The collected real-time power generation sequence is filtered to eliminate instantaneous noise, and historical power generation data is classified according to the same meteorological condition type, which includes at least sunny, cloudy, rainy, and snowy days. Based on the current real-time meteorological data, determine the meteorological condition type, and extract the power generation data of the same season and time period from the historical data of the corresponding category. Generate the power generation benchmark curve and its confidence interval under the condition through statistical analysis methods. The preprocessed real-time power generation sequence is synchronously compared with the reference curve to calculate the deviation of the real-time sequence from the reference curve; when the deviation continuously exceeds the upper or lower limit of the preset confidence interval and the duration reaches the preset time window threshold, it is determined that the current power generation state has an abnormal fluctuation trend. After determining that an anomaly exists, the spatial distribution characteristics of the deviation are obtained. If the anomaly only appears in a local combiner box or string circuit of the power station, it is initially located as a local fault. If the anomaly is widespread throughout the entire station, it is initially determined whether it is a global meteorological factor or a power station-level equipment failure by combining the current meteorological data.

[0007] Furthermore, the overlay status features in the image analysis module include overlay type, distribution density, and coverage area ratio, and their extraction process is as follows: The acquired photovoltaic panel monitoring images are preprocessed, including image denoising, geometric correction, and illumination equalization. The pre-trained convolutional neural network model is used to perform semantic segmentation on the pre-processed image, classifying the image pixels into background, clean surface of photovoltaic panels, and different types of coverings, including at least dust, snow, bird droppings, and plant shadows. Based on the semantic segmentation results, for each group of photovoltaic panels, the proportion of the projected area of ​​various coverings in its effective light-receiving area is obtained as the coverage density feature. Extract the uniformity of the distribution of coverings on the surface of photovoltaic panels, and the spatial characteristics of their mixed distribution when multiple coverings are present; The covering type, covering density characteristics, distribution uniformity characteristics, and mixed distribution spatial characteristics are fused and vectorized to form a comprehensive feature vector characterizing the current state of the photovoltaic panel surface covering.

[0008] Furthermore, the attribute data in the data acquisition module includes photovoltaic panel model, installation tilt angle, azimuth angle, current theoretical peak power, and environmental data of the area. The environmental data is acquired through miniature sensors deployed on the back panel or support of the photovoltaic panel, including panel surface temperature, back panel irradiance, and local wind speed.

[0009] Furthermore, the construction process of the cover impact model in the cover analysis module is as follows: Historical and real-time acquired photovoltaic panel monitoring images labeled with actual power generation efficiency loss rates and their corresponding photovoltaic panel attribute data are collected to form a training sample set; the power generation efficiency loss rate is calculated by comparing the power generation data of the same type of photovoltaic panel without cover during the same period with the actual power generation data. A deep neural network model is constructed, which includes a multi-layer convolutional feature extraction sub-network for processing monitoring images, a fully connected encoding sub-network for processing photovoltaic panel attribute data, and a feature fusion and regression output layer; the feature fusion layer concatenates or weights the high-dimensional features extracted by the two sub-networks using an attention mechanism. The deep neural network model is trained in a supervised manner using a training sample set. The predicted impact coefficient is calculated by forward propagation, and the loss function between the predicted impact coefficient and the labeled power generation efficiency loss rate is minimized by backpropagation and optimization algorithm until the model converges. The prediction accuracy and generalization ability of the trained model are evaluated using an independent validation dataset. Once the preset performance metrics are met, the model parameters are fixed and deployed to the coverage analysis module.

[0010] Furthermore, the fault knowledge base in the fault matching module performs pattern verification and correction based on statistical analysis of historical fault cases. It collects and labels actual fault events, corresponding power generation data fluctuation curves, on-site images, and final processing measures in historical operation and maintenance records. By extracting statistical patterns, it calibrates the basic fault modes generated by simulation and supplements the composite faults and edge cases that occur in actual operation and maintenance.

[0011] Furthermore, during the operation phase, the alarm triggering module classifies alarm behaviors into different levels based on the magnitude of the impact coefficient of the fault on power generation efficiency, the risk of fault propagation, and the expected repair time. Different levels correspond to different notification channels, response time limits, and reporting processes. The alarm behaviors include generating alarm information on the monitoring interface, sending SMS or email notifications to maintenance personnel, and generating work orders containing fault location and preliminary diagnosis reports.

[0012] Furthermore, the alarm triggering module is connected to a data interaction module via a wireless network. The data interaction module is used to store all historical data, real-time data, model parameters and fault records required for system operation, and provides a data interface. It also synchronizes data and receives instructions from external systems through a standardized API interface.

[0013] Furthermore, the coverage analysis module and the fault matching module are interconnected via a wireless network with a model optimization module. The model optimization module is used to collect historical diagnostic cases, actual processing results, and subsequent power generation data recovery status as data references for iterative updates of the coverage impact model and fault knowledge base.

[0014] Furthermore, the power generation initial judgment module is interconnected with the image analysis module and the data acquisition module via a wireless network; the coverage analysis module is interconnected with the image analysis module, the data acquisition module, and the fault matching module via a wireless network; and the fault matching module is interconnected with the alarm triggering module via a wireless network.

[0015] (III) Beneficial Effects Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By integrating multi-source information such as meteorological data, real-time power generation data, images, and equipment attributes, and introducing a deep learning-based coverage impact model, the system can accurately correlate abnormal fluctuations in power generation data with shading by specific types and areas of coverings, and quantify the degree of impact. This improves the accuracy and operability of diagnosis, and provides a direct basis for subsequent precise maintenance.

[0016] 2. By automating data collection, analysis, matching, and alarm triggering, the system frees maintenance personnel from tedious data monitoring and preliminary judgment. For potential power generation losses caused by coverings, the system can identify and accurately locate them in the early stages, guiding the maintenance team to carry out targeted cleaning or treatment. This avoids the waste of manpower and resources caused by blind inspections or excessive cleaning, thereby effectively improving the overall availability and power generation revenue of the power station and reducing the total life cycle maintenance cost. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall framework of the present invention.

[0019] The labels in the diagram represent: 1. Initial power generation assessment module; 2. Image analysis module; 3. Data acquisition module; 4. Coverage analysis module; 5. Fault matching module; 6. Alarm triggering module; 7. Data interaction module; 8. Model optimization module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] The present invention will be further described below with reference to embodiments.

[0022] This embodiment presents an IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants, such as... Figure 1 As shown, it includes: The initial power generation assessment module 1 is used to collect meteorological data of the target power station's location and real-time power generation data of the target power station, and to retrieve historical power generation data. By comparing the fluctuation trends of real-time data and historical data, it makes a preliminary judgment on whether there are any anomalies in the current power generation status. The process of judging the power generation status is as follows: The collected real-time power generation sequence is filtered to eliminate instantaneous noise, and historical power generation data is classified according to the same meteorological condition type, which includes at least sunny, cloudy, rainy, and snowy weather. Based on the current real-time meteorological data, determine the meteorological condition type, and extract the power generation data of the same season and time period from the historical data of the corresponding category. Generate the power generation benchmark curve and its confidence interval under the condition through statistical analysis methods. The preprocessed real-time power generation sequence is synchronously compared with the reference curve to calculate the deviation of the real-time sequence from the reference curve. When the deviation continuously exceeds the upper or lower limit of the preset confidence interval and the duration reaches the preset time window threshold, it is determined that the current power generation status has an abnormal fluctuation trend. After determining that an anomaly exists, the spatial distribution characteristics of the deviation are obtained. If the anomaly only appears in a local combiner box or string circuit of the power station, it is initially located as a local fault. If the anomaly is widespread throughout the entire station, it is initially determined whether it is a global meteorological factor or a power station-level equipment failure by combining the current meteorological data. By introducing a dynamic baseline curve and confidence interval mechanism based on meteorological conditions classification, and combining it with the judgment of continuous deviation in the spatiotemporal dimension, a more reliable preliminary identification and location of abnormal fluctuations in photovoltaic power generation is achieved. This effectively reduces false alarms caused by normal changes in meteorological conditions or instantaneous interference, and improves the pertinence and diagnostic efficiency of anomaly detection.

[0023] Image analysis module 2 is used to activate a preset image acquisition device to acquire monitoring images of the target power plant's photovoltaic panel array when the power generation initial judgment module 1 determines that an anomaly exists. It then performs visual analysis on each group of photovoltaic panels in the image, extracting the surface covering characteristics. These characteristics include covering type, distribution density, and coverage area ratio. The extraction process is as follows: The acquired photovoltaic panel monitoring images are preprocessed, including image denoising, geometric correction, and illumination equalization. The pre-trained convolutional neural network model is used to perform semantic segmentation on the pre-processed image, classifying the image pixels into background, clean surface of photovoltaic panels, and different types of coverings, including at least dust, snow, bird droppings, and plant shadows. Based on the semantic segmentation results, for each group of photovoltaic panels, the proportion of the projected area of ​​various coverings in its effective light-receiving area is obtained as the coverage density feature. Extract the uniformity of the distribution of coverings on the surface of photovoltaic panels, and the spatial characteristics of their mixed distribution when multiple coverings are present; The characteristics of cover type, cover density, distribution uniformity, and mixed distribution space are fused and vectorized to form a comprehensive feature vector that characterizes the current state of the photovoltaic panel surface cover, which is used for subsequent cover impact analysis.

[0024] Data acquisition module 3 is used to synchronously collect attribute data of photovoltaic panels. The attribute data includes photovoltaic panel model, installation tilt angle, azimuth angle, current theoretical peak power and regional environmental data. Regional environmental data is obtained through miniature sensors deployed on the back panel or bracket of the photovoltaic panel, including panel surface temperature, back panel irradiance and local wind speed.

[0025] Coverage analysis module 4 constructs a coverage impact model based on deep learning algorithms. The model takes the state characteristics of the cover and the attribute data of the photovoltaic panel as joint inputs. By analyzing the optical shading and thermal effects of the cover under specific attribute conditions, it outputs the impact coefficient of the current cover state on the power generation efficiency of the photovoltaic panel.

[0026] The fault matching module 5 receives the preliminary judgment of abnormal fluctuation trend characteristics and influence coefficients, and pre-stores a fault knowledge base containing various typical fault modes. Each fault mode is associated with a specific fluctuation trend characteristic range and influence coefficient threshold. The module combines the current abnormal fluctuation trend characteristics and influence coefficients with the patterns in the fault knowledge base to diagnose fault issues. The fault knowledge base performs pattern verification and correction based on statistical analysis of historical fault cases. It collects and labels actual fault events, corresponding power generation data fluctuation curves, on-site images, and final processing measures from historical operation and maintenance records. By extracting statistical patterns, it calibrates the basic fault modes generated by simulation and supplements composite faults and edge cases that occur in actual operation and maintenance. Through incremental learning or periodic batch retraining, it continuously updates the matching thresholds and feature weights of various fault modes in the fault knowledge base.

[0027] The alarm triggering module 6 is used to trigger alarm behaviors of corresponding levels based on the fault items matched by the fault matching module 5. During the operation phase, the alarm triggering module 6 classifies the alarm behaviors into different levels according to the magnitude of the impact coefficient of the fault item on the power generation efficiency, the risk of fault propagation, and the expected repair time. Different levels correspond to different notification channels, response time limits, and reporting processes. Alarm behaviors include generating alarm information on the monitoring interface, sending SMS or email notifications to operation and maintenance personnel, and generating work orders containing fault location and preliminary diagnosis reports.

[0028] The alarm triggering module 6 is connected to the data interaction module 7 via a wireless network. The data interaction module 7 is used to store all historical data, real-time data, model parameters and fault records required for system operation, and provides a data interface. It also synchronizes data and receives commands from external systems through a standardized API interface.

[0029] Coverage analysis module 4 and fault matching module 5 are interconnected via a wireless network to a model optimization module 8. The model optimization module 8 is used to collect historical diagnostic cases, actual processing results, and subsequent power generation data recovery status as data references for iterative updates of the coverage impact model and fault knowledge base, in order to improve the diagnostic accuracy and generalization ability of the model. Specifically, the internal parameters of the coverage impact model and the matching rule weights of the fault knowledge base are dynamically adjusted using diagnostic accuracy and fault recovery efficiency as reward signals.

[0030] The initial power generation judgment module 1 is interconnected with the image analysis module 2 and the data acquisition module 3 via a wireless network. The coverage analysis module 4 is interconnected with the image analysis module 2, the data acquisition module 3 and the fault matching module 5 via a wireless network. The fault matching module 5 is interconnected with the alarm triggering module 6 via a wireless network.

[0031] Compared with existing technologies, this method not only combines traditional power generation data anomaly monitoring with image recognition, but also introduces photovoltaic panel attribute data to construct a dynamic quantitative model of the impact of coverings. This enables more accurate matching of specific faults, improves the accuracy and automation level of fault diagnosis, and enhances the system's adaptability to different power plant environments and complex operating conditions through a continuous model self-optimization mechanism.

[0032] At other levels, this embodiment provides a process for constructing a coverage influence model, specifically as follows: Historical and real-time acquired photovoltaic panel monitoring images labeled with actual power generation efficiency loss rates and their corresponding photovoltaic panel attribute data are collected to form a training sample set; the power generation efficiency loss rate is calculated by comparing the power generation data of the same type of photovoltaic panel without cover during the same period with the actual power generation data. A deep neural network model is constructed, which includes a multi-layer convolutional feature extraction sub-network for processing monitoring images, a fully connected encoding sub-network for processing photovoltaic panel attribute data, and a feature fusion and regression output layer; the feature fusion layer concatenates or weights the high-dimensional features extracted by the two sub-networks through an attention mechanism. The deep neural network model is trained in a supervised manner using a training sample set. The predicted impact coefficient is calculated by forward propagation, and the loss function between the predicted impact coefficient and the labeled power generation efficiency loss rate is minimized by backpropagation and optimization algorithm until the model converges. The prediction accuracy and generalization ability of the trained model are evaluated using an independent validation dataset. Once the preset performance indicators are met, the model parameters are fixed and deployed to the coverage analysis module 4 for online real-time analysis.

[0033] By deeply integrating visual features with multi-dimensional attribute data of photovoltaic panels, the impact of coverings has been transformed from qualitative identification to quantitative analysis. This allows for the differentiation of covering types and, more importantly, the dynamic quantification of the actual impact coefficient of coverings on power generation efficiency by combining the physical state and microenvironment of specific photovoltaic panels. This provides a reliable data core for subsequent accurate fault diagnosis and performance evaluation, thereby improving the intelligence level and diagnostic accuracy of the monitoring system.

[0034] In summary, this invention achieves early identification and preliminary judgment of abnormal fluctuations through the linkage analysis of meteorological and power generation data. Combined with deep learning-based visual feature extraction and multi-dimensional attribute analysis, it can accurately quantify the specific impact of cover on power generation efficiency, upgrading traditional simple alarms to quantitative diagnosis with clear impact coefficients. By integrating image recognition, environmental perception, and device attribute data, and conducting comprehensive analysis through a coverage impact model, the built-in fault knowledge base and continuous optimization mechanism enable the system to have self-learning capabilities. The diagnostic logic evolves continuously with the accumulation of experience, resulting in a high degree of intelligence.

[0035] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote monitoring and fault diagnosis system for photovoltaic power plants based on the Internet of Things, characterized in that, include: The power generation preliminary judgment module is used to collect meteorological data of the area where the target power station is located, real-time power generation data of the target power station, and call historical power generation data. By comparing the fluctuation trend of real-time data and historical data, it makes a preliminary judgment on whether there is any abnormality in the current power generation status. The image analysis module is used to activate a preset image acquisition device to acquire monitoring images of the photovoltaic panel array of the target power station when the power generation preliminary judgment module determines that there is an anomaly, and to perform visual analysis on each group of photovoltaic panels in the image to extract the surface covering status features. A data acquisition module is used to synchronously acquire the attribute data of the photovoltaic panel; The coverage analysis module constructs a coverage impact model based on deep learning algorithms. The model takes the state characteristics of the cover and the attribute data of the photovoltaic panel as joint inputs and outputs the impact coefficient of the current state of the cover on the power generation efficiency of the photovoltaic panel. The fault matching module is used to receive the preliminary judgment of abnormal fluctuation trend characteristics and influence coefficients, pre-store fault knowledge base, combine the current abnormal fluctuation trend characteristics and influence coefficients, match them with the patterns in the fault knowledge base, and diagnose fault issues. The alarm triggering module is used to trigger alarm behaviors of the corresponding level based on the fault items matched by the fault matching module.

2. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The process for determining the power generation status in the initial power generation assessment module is as follows: The collected real-time power generation sequence is filtered to eliminate instantaneous noise, and historical power generation data are categorized according to the same meteorological conditions. Based on the current real-time meteorological data, determine the meteorological condition type, and extract the power generation data of the same season and time period from the historical data of the corresponding category. Generate the power generation benchmark curve and its confidence interval under the condition through statistical analysis methods. The preprocessed real-time power generation sequence is synchronously compared with the reference curve to calculate the deviation of the real-time sequence from the reference curve; when the deviation continuously exceeds the upper or lower limit of the preset confidence interval and the duration reaches the preset time window threshold, it is determined that the current power generation state has an abnormal fluctuation trend. After determining that an anomaly exists, the spatial distribution characteristics of the deviation are obtained. If the anomaly only appears in a local combiner box or string circuit of the power station, it is initially located as a local fault. If the anomaly is widespread throughout the entire station, it is initially determined whether it is a global meteorological factor or a power station-level equipment failure by combining the current meteorological data.

3. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The overlay status features in the image analysis module include overlay type, distribution density, and coverage area ratio, and their extraction process is as follows: Preprocess the acquired photovoltaic panel monitoring images; A pre-trained convolutional neural network model is used to perform semantic segmentation on the pre-processed image, classifying image pixels into background, clean photovoltaic panel surface, and different types of coverings; Based on the semantic segmentation results, for each group of photovoltaic panels, the proportion of the projected area of ​​various coverings in its effective light-receiving area is obtained as the coverage density feature. Extract the uniformity of the distribution of coverings on the surface of photovoltaic panels, and the spatial characteristics of their mixed distribution when multiple coverings are present; The covering type, covering density characteristics, distribution uniformity characteristics, and mixed distribution spatial characteristics are fused and vectorized to form a comprehensive feature vector characterizing the current state of the photovoltaic panel surface covering.

4. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The attribute data in the data acquisition module includes photovoltaic panel model, installation tilt angle, azimuth angle, current theoretical peak power, and regional environmental data. The regional environmental data is acquired by miniature sensors deployed on the back panel or support of the photovoltaic panel, including panel surface temperature, back panel irradiance, and local wind speed.

5. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The construction process of the cover impact model in the cover analysis module is as follows: Historical and real-time acquired photovoltaic panel monitoring images labeled with actual power generation efficiency loss rates and their corresponding photovoltaic panel attribute data are collected to form a training sample set; A deep neural network model is constructed, which includes a multi-layer convolutional feature extraction subnetwork for processing monitoring images, a fully connected encoding subnetwork for processing photovoltaic panel attribute data, and a feature fusion and regression output layer. The deep neural network model is trained in a supervised manner using a training sample set. The predicted impact coefficient is calculated by forward propagation, and the loss function between the predicted impact coefficient and the labeled power generation efficiency loss rate is minimized by backpropagation and optimization algorithm until the model converges. The prediction accuracy and generalization ability of the trained model are evaluated using an independent validation dataset. Once the preset performance metrics are met, the model parameters are fixed and deployed to the coverage analysis module.

6. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The fault knowledge base in the fault matching module performs pattern verification and correction based on statistical analysis of historical fault cases. It collects and labels actual fault events, corresponding power generation data fluctuation curves, on-site images and final processing measures in historical operation and maintenance records. By extracting statistical patterns, it calibrates the basic fault modes generated by simulation and supplements the composite faults and edge cases that occur in actual operation and maintenance.

7. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, During operation, the alarm triggering module classifies alarm behaviors into different levels based on the magnitude of the impact coefficient of the fault on power generation efficiency, the risk of fault propagation, and the expected repair time. Different levels correspond to different notification channels, response time limits, and reporting processes.

8. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The alarm triggering module is connected to a data interaction module via a wireless network. The data interaction module is used to store all historical data, real-time data, model parameters and fault records required for system operation, and provides a data interface. It also synchronizes data and receives instructions from external systems through a standardized API interface.

9. The IoT-based remote monitoring and fault diagnosis system for photovoltaic power plants according to claim 1, characterized in that, The coverage analysis module and the fault matching module are interconnected via a wireless network with a model optimization module. The model optimization module is used to collect historical diagnostic cases, actual processing results, and subsequent power generation data recovery status as data references for iterative updates of the coverage impact model and fault knowledge base.

10. A remote monitoring and fault diagnosis system for photovoltaic power plants based on the Internet of Things according to claim 1, characterized in that, The initial power generation judgment module is interconnected with the image analysis module and the data acquisition module via a wireless network. The coverage analysis module is interconnected with the image analysis module, the data acquisition module, and the fault matching module via a wireless network. The fault matching module is interconnected with the alarm triggering module via a wireless network.

Images