Online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things

Abstract

The present application provides an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things, which relates to the field of photovoltaic power generation technology. It collects high-resolution remote sensing images and dynamically fuses the hidden state representations of each monitoring area at the feature domain level to establish a correlation model of key features and a significant information aggregation network to reconstruct the global operating state map of the photovoltaic power station. Finally, a lightweight classifier is combined to achieve cross-level enhanced expression of defect features, so as to capture the abnormal conduction relationship between components through correlation modeling in the feature space, thereby significantly improving the perception sensitivity of cross-regional defects such as microcracks and hidden hot spots. In this way, high-precision detection of abnormal operating status of distributed photovoltaic power stations is achieved, which helps to improve the defect perception capability of photovoltaic power stations and the intelligence level of distributed photovoltaic power station online monitoring systems.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic power generation technology, and more particularly, in an embodiment of the present application, relates to an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things. Background Art

[0002] With the rapid development of photovoltaic power generation technology, distributed photovoltaic power stations have become a key development direction in the new energy sector due to their advantages such as flexible deployment and localized consumption. However, because photovoltaic power stations are typically distributed over vast areas and their components are exposed to complex environments for long periods of time, defects such as panel surface contamination, hot spot effects, and line aging are common. Traditional manual inspection methods are inefficient and difficult to achieve in real-time monitoring and precise operation and maintenance.

[0003] In recent years, intelligent monitoring systems based on drones and the Internet of Things (IoT) have become a mainstream solution. Chinese patent CN116488323B proposes an IoT-based distributed photovoltaic power station monitoring system and method. This system uses drone swarms to perform regional inspections of the photovoltaic power station. The system divides the entire plant area into several sub-regions, and drone formations collect remote sensing images of each sub-region. These images are then stitched together to form a comprehensive monitoring image for overall analysis. However, this approach has significant drawbacks. First, analyzing each sub-region image individually lacks in-depth analysis of the global photovoltaic power station status and cross-region correlations, making it difficult to identify distributed anomalies caused by series-parallel module configurations or shadowing. Second, the overall analysis after simple image stitching is susceptible to interference from factors such as uneven lighting, artifacts, and differences in shooting angles. This leads to high feature distortion in high-dynamic lighting conditions or complex terrain. Furthermore, key defect signatures can be weakened or missed. In particular, the sensitivity to early-stage subtle defects (such as hidden cracks, microcracks, micro-arcing, and hot spot effects) is insufficient, seriously impacting the timeliness and accuracy of photovoltaic power station fault warnings.

[0004] Therefore, an optimized distributed photovoltaic power station online monitoring system is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things, which first collects high-resolution remote sensing images and dynamically fuses the hidden state representations of each monitoring area at the feature domain level, then establishes a correlation model of key features and a significant information aggregation network to reconstruct the global operating status map of the photovoltaic power station. Finally, a lightweight classifier is combined to achieve cross-level enhanced expression of defect features, so as to capture the abnormal conduction relationship between components through correlation modeling in the feature space, thereby significantly improving the perception sensitivity of cross-regional defects such as microcracks and hidden hot spots. In this way, high-precision detection of abnormal operating status of distributed photovoltaic power stations is achieved, which helps to improve the defect perception capability of photovoltaic power stations and the intelligence level of distributed photovoltaic power station online monitoring systems.

[0006] According to one aspect of the present application, there is provided an online intelligent monitoring system for a distributed photovoltaic power station based on the Internet of Things, which includes:

[0007] The UAV monitoring information generation module is used to obtain remote sensing images of the entire photovoltaic power station area and generate UAV monitoring paths and monitoring points;

[0008] A position deviation calculation module is used to control the UAV to fly along the monitoring path and determine the position deviation based on the current position of the UAV and the position of the monitoring point;

[0009] The photovoltaic power station status monitoring and analysis module is used to collect remote sensing images of the monitoring area through the camera on the drone if the position deviation is less than or equal to the predetermined threshold, generate photovoltaic power station monitoring information based on the remote sensing images of the monitoring area, and then transmit the photovoltaic power station monitoring information to the cloud server;

[0010] Among them, the photovoltaic power station status monitoring and analysis module includes: a power station monitoring area status feature extraction unit, which is used to capture the monitoring area status features of remote sensing images of multiple monitoring areas in the photovoltaic power station to obtain a set of photovoltaic power station monitoring area status features; a power station status monitoring unit, which is used to perform power station status monitoring based on regional status significant feature aggregation analysis on the set of photovoltaic power station monitoring area status features to obtain photovoltaic power station monitoring information.

[0011] Compared with the existing technology, the present application provides an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things. It first collects high-resolution remote sensing images and dynamically fuses the hidden state representations of each monitoring area at the feature domain level. It then establishes a correlation model of key features and a significant information aggregation network to reconstruct the global operating status map of the photovoltaic power station. Finally, a lightweight classifier is combined to achieve cross-level enhanced expression of defect features, so as to capture the abnormal conduction relationship between components through correlation modeling in the feature space, thereby significantly improving the perception sensitivity of cross-regional defects such as microcracks and hidden hot spots. In this way, high-precision detection of abnormal operating status of distributed photovoltaic power stations is achieved, which helps to improve the defect perception capability of photovoltaic power stations and the intelligence level of distributed photovoltaic power station online monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 This is a system block diagram of an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things according to an embodiment of the present application.

[0014] Figure 2 This is a block diagram of a photovoltaic power station status monitoring and analysis module in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application.

[0015] Figure 3 Schematic diagram of data flow of a photovoltaic power station status monitoring and analysis module in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application.

[0016] Figure 4 This is a block diagram of a power station monitoring area status feature extraction unit in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application.

[0017] Figure 5 This is a block diagram of a power station status monitoring unit in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application.

[0018] Figure 6 This is a block diagram of a feature aggregation subunit for regional status significant information in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application.

[0019] Figure 7This is a block diagram of a secondary subunit for linear grouping calibration of a photovoltaic power station monitoring area state in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0021] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0022] In addition, numerous specific details are provided in the following detailed description to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0024] Chinese patent CN116488323B proposes a distributed photovoltaic power station monitoring system and method based on the Internet of Things. However, this method has the following defects: First, in the local sub-region independent analysis mode, global operating status characteristics (such as the series-parallel topology of components) and cross-region correlation characteristics (such as abnormal conduction paths caused by shadow occlusion) are not effectively modeled, resulting in limited recognition of distributed anomalies (such as string mismatch and current backflow); second, the full-station image stitching and reconstruction method is susceptible to interference from dynamic lighting offset and multi-source artifacts (such as angle distortion from drone aerial photography and mirror reflection noise), resulting in distortion of the feature space in high-dynamic scenarios. This not only leads to distortion in the representation of key defects (such as misjudgment of hot spot area and deviation of the direction of hidden cracks), but also seriously weakens the detection sensitivity of early microscopic defects (<0.5mm level hidden cracks and micro-arcing discharge traces), resulting in an increase in the warning lag rate of high-risk faults such as hot spot effect and potential PID attenuation.

[0025] In response to the above technical problems, this application proposes an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things. Figure 1FIG is a system block diagram of an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, an online intelligent monitoring system 100 for a distributed photovoltaic power station based on the Internet of Things includes: a drone monitoring information generation module 110, which is used to obtain remote sensing images of the entire area of ​​the photovoltaic power station and generate drone monitoring paths and monitoring points; a position deviation calculation module 120, which is used to control the drone to fly along the monitoring path and determine the position deviation based on the current position of the drone and the position of the monitoring point; a photovoltaic power station status monitoring and analysis module 130, which is used to collect remote sensing images of the monitoring area through a camera carried by the drone if the position deviation is less than or equal to a predetermined threshold, and generate photovoltaic power station monitoring information based on the remote sensing images of the monitoring area, and then transmit the photovoltaic power station monitoring information to a cloud server; a drone flight parameter adjustment and correction module 140, which is used to correct the drone position by adjusting the drone flight parameters if the position deviation is greater than a predetermined threshold.

[0026] In the aforementioned IoT-based distributed photovoltaic power station online intelligent monitoring system 100, the drone monitoring information generation module 110 is used to acquire full-area remote sensing images of the photovoltaic power station and generate drone monitoring routes and monitoring points. It should be understood that photovoltaic power stations typically cover vast areas (e.g., tens to hundreds of hectares) and have complex component layouts (e.g., series-parallel structures and undulating terrain). Acquiring full-area remote sensing images is fundamental to systematic monitoring, avoiding blind spots in manual inspections and ensuring that all photovoltaic modules are covered. Furthermore, anomalies in photovoltaic power stations (e.g., hot spots, hidden cracks, and shadows) often have cross-regional transmission characteristics. For example, current mismatch in a particular string group can affect the efficiency of modules in adjacent areas. Full-area remote sensing images can globally model the topological relationships of the modules. Combined with drone monitoring route planning, these images can capture distributed anomalies and their propagation paths, avoiding misjudgments caused by local analysis. Furthermore, generating pre-set routes and monitoring points based on remote sensing images can reduce ineffective drone flights (e.g., repeated scanning and missed areas), reducing energy consumption and time costs. At the same time, incorporating terrain data (such as slope and obstacle height) can mitigate flight risks (such as collisions and signal interference), improving mission safety. In particular, in one specific example of this application, a high-resolution infrared thermal imager and a visible light camera can be mounted on a drone to perform grid-based aerial photography of a photovoltaic power plant at a fixed altitude (e.g., 100 meters). Reference can also be made to step S1 of the IoT-based distributed photovoltaic power plant monitoring system and method proposed in existing patent CN116488323B.

[0027] In the aforementioned IoT-based distributed photovoltaic power station online intelligent monitoring system 100, the position deviation calculation module 120 is used to control the drone's flight path and determine the position deviation based on the drone's current position and the location of the monitoring points. It should be understood that in drone monitoring of photovoltaic power stations, precise flight path control and real-time position deviation correction are key components for ensuring monitoring quality and safety. Because photovoltaic power stations are often located in complex terrain (such as mountains, water, or building rooftops), drones must maintain stable flight along a pre-set path to avoid collisions with obstacles (such as support structures, trees, or high-voltage power lines) while ensuring comprehensive coverage of monitoring points. For example, a coastal tidal flat photovoltaic power station covers 80 hectares. The terrain is flat but winds are frequent (often reaching 8 m / s). If the drone relies solely on GPS navigation, it is susceptible to signal drift and wind interference, causing the actual flight path to deviate by more than 3 meters, potentially missing rust spots on component connections caused by tidal corrosion. To achieve this, the system integrates multi-source positioning data (such as centimeter-level RTK-GPS positioning and visual SLAM environmental feature matching) to track the drone's position in real time. It then spatially matches the coordinates of pre-set monitoring points (such as the longitude and latitude of each string center point) to calculate lateral and longitudinal deviations. If the deviation exceeds a threshold (e.g., a horizontal error > 0.5 meter or an altitude error > 1 meter), the flight control system initiates a dynamic adjustment strategy: It uses a PID controller to adjust motor thrust distribution, correcting yaw and pitch angles. It also dynamically adjusts flight altitude based on terrain elevation data (DEM) to avoid triggering emergency hovering due to unexpected obstacles (such as temporary construction equipment). For example, in strong winds, if the drone detects a 1.2-meter eastward deviation, the control system immediately increases the west rotor speed and reduces the east thrust, narrowing the deviation to within 0.3 meters within 2 seconds. This ensures accurate infrared image acquisition angles for the next monitoring point (e.g., the inverter cooling vent). Furthermore, in the event of a sudden failure (such as a sudden temperature rise in a component in a certain area), the system can temporarily insert monitoring points and replan the local path, using an algorithm to circumvent obstacles while simultaneously updating global path weights to reduce the total flight distance. This closed-loop control mechanism not only improves flight robustness in complex environments but also ensures consistent imaging of microscopic defects such as hot spots and hidden cracks through millimeter-level deviation correction. Reference can also be made to steps S2-S5 of the IoT-based distributed photovoltaic power station monitoring system and method proposed in the existing patent CN116488323B.

[0028] In the aforementioned IoT-based distributed photovoltaic power station online intelligent monitoring system 100, the photovoltaic power station status monitoring and analysis module 130 is configured to, if the position deviation is less than or equal to a predetermined threshold, use a drone-mounted camera to capture remote sensing images of the monitoring area, generate photovoltaic power station monitoring information based on the remote sensing images of the monitoring area, and then transmit the photovoltaic power station monitoring information to a cloud server. It should be understood that microscopic defects in photovoltaic modules (such as hidden cracks and hot spots) are highly sensitive to imaging angle and resolution. If the drone deviates from the preset monitoring point, it may cause image distortion or blurring of key features (such as hot spot boundary diffusion and hidden crack orientation shift), leading to misjudgment. For example, in a distributed photovoltaic power station, while flying on a calibrated path, the drone uses RTK-GPS and visual SLAM fusion positioning to calculate the deviation value in real time. When the horizontal error converges to within 0.3 meters (below the threshold of 0.5 meters), the system immediately activates the onboard multispectral camera (such as a visible light and infrared dual-mode sensor) to image the target area with a vertical pitch angle accuracy of ±5°, capturing the module surface temperature distribution (with a resolution of 0.1°C) and visible light texture details.

[0029] In specific implementation, using a mountain photovoltaic scenario as an example, raw images captured by a drone at a stable position are first subjected to edge enhancement and illumination compensation to eliminate artifacts caused by terrain shadows. Subsequently, a lightweight convolutional network running on an embedded processor extracts pixel clusters and linear features of hidden cracks in hotspot areas, generating structured monitoring information containing metadata such as geographic coordinates, temperature gradients, and defect types. This information is transmitted in real time to a cloud server via the multi-protocol network modules described in the patent (e.g., a hybrid ZigBee and NB-IoT network). The ZigBee module is responsible for local high-speed transmission (e.g., relay nodes within the power plant), while NB-IoT is used for wide-area, low-power backhaul. The cloud server constructs a digital twin model based on historical data and real-time information, dynamically updating the power plant health status map. For example, if a string temperature rises by 2°C due to dust accumulation, the system automatically marks it as requiring potential cleaning and pushes it to the operation and maintenance terminal.

[0030] It is worth mentioning that if the position deviation is greater than a predetermined threshold, the drone position is corrected by adjusting the drone flight parameters.

[0031] Figure 2 This is a block diagram of a photovoltaic power station status monitoring and analysis module in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application. Figure 3 Schematic diagram of data flow of photovoltaic power station status monitoring and analysis module in the distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to the embodiment of the present application. Figure 2 and Figure 3As shown, the photovoltaic power station status monitoring and analysis module 130 includes: a power station monitoring area status feature extraction unit 131, which is used to capture the monitoring area status features of remote sensing images of multiple monitoring areas in the photovoltaic power station to obtain a set of photovoltaic power station monitoring area status features; a power station status monitoring unit 132, which is used to perform power station status monitoring based on regional status significant feature aggregation analysis on the set of photovoltaic power station monitoring area status features to obtain photovoltaic power station monitoring information.

[0032] Accordingly, the technical concept of this application is to use drone formations to collect high-resolution remote sensing images in different regions according to intelligently planned paths. While maintaining the original image quality of each monitoring area, a multi-level feature extraction network is constructed to mine potential defect representations from the images of each monitoring area. Furthermore, a cross-regional saliency feature optimization integrated parsing network is used to capture the global state correlation of distributed photovoltaic power plants. Specifically, the latent state representations of each monitoring area are dynamically integrated at the feature domain level. By establishing a correlation model for key features such as hot spot diffusion paths and shadow migration trajectories between regions and a saliency information aggregation network, a global operating state map of the photovoltaic power plant is reconstructed. Finally, a lightweight classifier is combined to achieve cross-level enhanced representation of defect features. This allows the system to capture abnormal conduction relationships between components through feature space correlation modeling without the need for physical image stitching, significantly improving the sensitivity to cross-regional defects such as microcracks and hidden hot spots. This avoids the noise interference introduced by direct image stitching, while enabling three-dimensional identification of distributed defects through state correlation analysis in the feature domain. Ultimately, high-precision detection of distributed photovoltaic power plant operating status anomalies is achieved, which helps improve the defect perception capabilities of photovoltaic power plants and the intelligence level of distributed photovoltaic power plant online monitoring systems.

[0033] Figure 4 FIG is a block diagram of a power station monitoring area state feature extraction unit in a distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to an embodiment of the present application. Figure 4 As shown, in an embodiment of the present application, the power station monitoring area status feature extraction unit 131 includes: a multi-monitoring area remote sensing image acquisition subunit 1311, which is used to acquire remote sensing images of multiple monitoring areas of the photovoltaic power station through the camera to obtain a set of monitoring area remote sensing images; a monitoring area remote sensing image enhancement preprocessing subunit 1312, which is used to perform image enhancement preprocessing on each monitoring area remote sensing image in the set of monitoring area remote sensing images to obtain a set of enhanced monitoring area remote sensing images; a monitoring area status feature extraction subunit 1313, which is used to extract monitoring area status features from each enhanced monitoring area remote sensing image in the set of enhanced monitoring area remote sensing images to obtain a set of monitoring area status features of the photovoltaic power station.

[0034] Specifically, the multi-monitoring-area remote sensing image acquisition subunit 1311 is configured to use the camera to capture remote sensing images of multiple monitoring areas of a photovoltaic power station to obtain a collection of monitoring-area remote sensing images. It should be understood that the core reason for using cameras to capture a collection of remote sensing images of multiple monitoring areas in intelligent monitoring of photovoltaic power stations is that traditional single-image stitching methods suffer from the lack of global correlation features and sensitivity to dynamic interference. In existing technologies, simple stitching of full-station images is susceptible to interference from lighting offsets, shooting angle differences, and terrain artifacts (such as specular reflection noise during drone aerial photography), resulting in distortion of the spatial distribution characteristics of key defects (such as hidden cracks and hot spots). Especially in highly dynamic scenarios (such as rapidly moving clouds), localized temperature gradients or crack orientations may be misrepresented, significantly increasing the miss detection rate. For example, when stitching images from a single aerial shot at a mountain photovoltaic power station, the changing angle of afternoon sunlight caused overlapping shadows from components, leading to the misidentification of two normal connection points as hot spots, while actual micro-arcing areas (<2 mm in size) were not identified due to image blur. To this end, a regional acquisition strategy divides the power station into several monitoring sub-areas (such as by string or terrain units) and independently obtains remote sensing images of each area (such as visible light and infrared dual modalities). This can effectively avoid the distortion introduced by global stitching while preserving the integrity of local details.

[0035] Specifically, the monitoring area remote sensing image enhancement preprocessing subunit 1312 is configured to perform image enhancement preprocessing on each monitoring area remote sensing image in the set of monitoring area remote sensing images to obtain a set of enhanced monitoring area remote sensing images. It should be understood that in distributed photovoltaic power station monitoring scenarios, since photovoltaic power stations are often located in open outdoor environments, remote sensing images collected by drones are susceptible to interference such as sudden changes in light intensity (such as cloud cover and reflections), overlapping shadows on component surfaces (such as projections from brackets or adjacent components), and local contrast imbalance caused by dust adhesion. This directly leads to a decrease in the ability to distinguish defect features such as hot spot textures and microcrack orientations from background noise in the original image. Therefore, in order to address the problem of key feature loss caused by complex environmental interference, in the technical solution of the present application, image enhancement preprocessing is performed on each monitoring area remote sensing image in the set of monitoring area remote sensing images to obtain a set of enhanced monitoring area remote sensing images. By performing image enhancement preprocessing, specifically employing adaptive histogram equalization and a multi-scale Retinex fusion algorithm, we dynamically compensate for differences in illumination distribution across different monitoring areas, eliminating over- or under-exposure caused by sudden changes in ambient light. Furthermore, a shadow suppression model is used to separate component body shadows from abnormal shadows (such as hot spots), allowing the enhanced image to highlight subtle surface texture variations and defect edge features. The core purpose of this operation is to provide high-signal-to-noise ratio input data for the subsequent feature extraction network, preventing noise signals from being mistakenly identified as valid features due to interference from illumination artifacts during the subsequent feature extraction process for the PV power plant monitoring area. In particular, this ensures the integrity of the representation of gradient characteristics of microcracks and localized temperature rise areas caused by hot spot effects in the feature space.

[0036] Specifically, the monitoring area state feature extraction subunit 1313 is configured to extract monitoring area state features from each enhanced monitoring area remote sensing image in the set of enhanced monitoring area remote sensing images to obtain a set of monitoring area state features for the photovoltaic power station. In an embodiment of the present application, the monitoring area state feature extraction subunit 1313 is configured to: pass each enhanced monitoring area remote sensing image in the set of enhanced monitoring area remote sensing images through a ConvNeXt-based photovoltaic power station monitoring area state feature extractor to obtain a set of photovoltaic power station monitoring area state feature vectors as the set of photovoltaic power station monitoring area state features. It should be understood that during the monitoring of distributed photovoltaic power stations, surface defects of photovoltaic modules (such as hidden cracks and micro-arcing) appear as micron-level texture changes or local grayscale anomalies in the enhanced monitoring area remote sensing images. However, traditional convolutional neural networks have limited ability to capture high-frequency detail features. In particular, in the context of metallized grid interference caused by light reflection and pseudo-texture caused by dust adhesion, normal module process textures are easily misidentified as defect features, making it difficult to effectively separate the coupled representations of photovoltaic panel texture features from environmental noise. Therefore, in the technical solution of this application, a ConvNeXt-based photovoltaic power station monitoring area state feature extractor is introduced. Through an improved deep separable convolution structure and channel attention mechanism, it can adaptively enhance the cross-scale characterization capability of component surface defects. For example, it uses a large kernel convolution layer to capture the continuous edge features of microcracks and enhances the temperature gradient difference between hot spot areas and normal components through a hierarchical feature pyramid. Specifically, a regional state feature space with strong discrimination is constructed, and the enhanced monitoring area remote sensing image data is mapped into a photovoltaic power station monitoring area state feature vector containing multi-dimensional state parameters such as component electrical performance degradation and physical damage degree. This provides high-dimensional semantic information for subsequent photovoltaic power station cross-region feature association modeling and full-region state representation.

[0037] Figure 5 FIG is a block diagram of a power station status monitoring unit in an Internet of Things-based distributed photovoltaic power station online intelligent monitoring system according to an embodiment of the present application. Figure 5 As shown, in an embodiment of the present application, the power station status monitoring unit 132 includes: a feature aggregation subunit 1321 of regional status significant information, which is used to integrate the set of the photovoltaic power station monitoring regional status feature vectors through a feature optimization integration analysis network of regional status significant information to obtain a photovoltaic power station full regional status representation vector; a photovoltaic power station monitoring information determination subunit 1322, which is used to perform photovoltaic power station status detection based on the photovoltaic power station full regional status representation vector to determine photovoltaic power station monitoring information, and the detection result is used to indicate whether there is any abnormality in the operating status of the distributed photovoltaic power station.

[0038] It should be understood that because photovoltaic power station components form an energy transmission network through series and parallel circuits, defects such as hot spot diffusion and hidden crack extension often exhibit cross-monitoring conduction characteristics (for example, a hot spot in one area triggers current imbalance in adjacent components). Traditional single-region feature analysis or physical image stitching solutions are unable to effectively capture such correlation characteristics. Therefore, to address the problem of the lack of feature correlation modeling for cross-region defect conduction patterns in existing technologies, the technical solution of this application further integrates the set of photovoltaic power station monitoring area state feature vectors through a feature optimization integration analysis network of regional state saliency information to obtain a photovoltaic power station full-region state representation vector. By introducing a feature optimization ensemble parsing network based on self-learning reinforcement learning to extract significant information about the state of feature regions, the paper first performs linear cluster parsing on the PV power plant monitoring region state feature vectors extracted by ConvNeXt. This constructs a linear cluster center encoding vector for the PV power plant monitoring region state as a global feature skeleton, revealing macroscopic correlation patterns in the states of each monitoring region (such as common trends in module degradation). The paper then uses a deep collaborative implicit encoding vector of the PV power plant monitoring region state to model the nonlinear interactions between regional features and the global linear structure, for example, capturing the potential transmission path between microcracks in one region and anomalies in adjacent regions. The paper also uses a feature cluster compensation incremental operator to dynamically correct the linear clustering results, eliminating feature offsets caused by environmental noise or local anomalies. This allows the reconstruction of a three-dimensional propagation map of cross-regional defects. By fusing the global framework provided by linear clustering with the compensation information enhanced by nonlinear energy relaxation coupling, the paper aggregates the discrete PV power plant monitoring region state feature vectors into a full-region PV power plant state representation vector that embodies implicit laws such as inter-module electrical correlation and thermodynamic conduction. This helps improve the detection reliability of complex defect patterns such as abnormal module string conduction and shadow migration cascading failures.

[0039] Specifically, the feature aggregation subunit 1321 of the regional state significant information is used to obtain a photovoltaic power station full-region state representation vector by subjecting the set of the photovoltaic power station monitoring area state feature vectors to a feature optimization integration parsing network of the regional state significant information. Figure 6 FIG is a block diagram of a feature aggregation subunit for regional status significant information in an online intelligent monitoring system for distributed photovoltaic power stations based on the Internet of Things according to an embodiment of the present application. Figure 6As shown, in an embodiment of the present application, the feature aggregation subunit 1321 of the regional state significant information includes: a photovoltaic power station monitoring area state initial linear clustering secondary subunit 1321-1, which is used to perform linear cluster analysis on the set of photovoltaic power station monitoring area state feature vectors to obtain the photovoltaic power station monitoring area state linear cluster center encoding vector; a photovoltaic power station monitoring area state linear grouping calibration secondary subunit 1321-2, which is used to perform grouping calibration on the set of photovoltaic power station monitoring area state feature vectors to obtain the photovoltaic power station monitoring area state linear cluster compensation component encoding vector; a photovoltaic power station full area state feature fusion secondary subunit 1321-3, which is used to fuse the photovoltaic power station monitoring area state linear cluster compensation component encoding vector and the photovoltaic power station monitoring area state linear cluster center encoding vector to obtain the photovoltaic power station full area state representation vector.

[0040] Specifically, the photovoltaic power station monitoring area state initial linear clustering secondary subunit 1321-1 is used to perform linear cluster analysis on the set of photovoltaic power station monitoring area state feature vectors to obtain the photovoltaic power station monitoring area state linear cluster center encoding vector, which is expressed as follows:

[0041]

[0042]

[0043] in, is the set of state feature vectors of the photovoltaic power station monitoring area, , , and They are the first, second, and third state feature vectors in the set of monitoring area state feature vectors of photovoltaic power station. and The state feature vector of the monitoring area of ​​a photovoltaic power station, yes The number of vectors in is the linear cluster center encoding vector of the monitoring area state of the photovoltaic power station. It should be understood that

[0044] Because feature vectors collected by multi-source sensors (such as infrared thermal imagers and current and voltage sensors) typically exhibit high dimensionality (e.g., 512 dimensions) and complex coupling characteristics (e.g., strong correlations between temperature gradients and module aging characteristics), direct nonlinear modeling is susceptible to noise (e.g., temperature artifacts caused by transient cloud cover) and struggles to quickly identify common degradation patterns across regions (e.g., string-level current mismatch trends). For example, in monitoring data from a distributed power station, the Euclidean distance between a localized temperature rise (ΔT = 3°C) caused by dust accumulation and a true hot spot (ΔT > 8°C) in the original high-dimensional space is only 0.15, making it difficult to distinguish using traditional thresholding methods. Linear cluster parsing can capture the natural grouping patterns of data within the feature space, compress redundant dimensions into a low-rank subspace, and extract cluster center encoding vectors that represent the global linear structure, providing a benchmark anchor for subsequent nonlinear feature gain. Specifically, the linear cluster center encoding vectors for the monitored regional state of a photovoltaic power station are essentially linear principal component extractions of cross-modal interaction features (e.g., load time series fluctuations and battery state degradation trajectories). For example, the voltage droop patterns and diurnal load peak-valley cycle characteristics isolated during the feature dissociation phase were linearly clustered to form three typical centers: normal operation centers (low temperature variance, stable current), early decay centers (periodic voltage drift), and sudden abnormal centers (sudden temperature rise accompanied by current oscillation). These centers not only define the baseline categories of power plant health but also reveal fault transmission paths through their distribution in feature space (for example, a cluster center points to current back diffusion caused by string mismatch), providing a physically interpretable reference framework for dynamic threshold adjustment.

[0045] Specifically, the photovoltaic power station monitoring area state linear grouping calibration secondary subunit 1321 - 2 is used to perform grouping calibration on the set of photovoltaic power station monitoring area state feature vectors to obtain the photovoltaic power station monitoring area state linear cluster compensation component coding vector. Figure 7 FIG1 is a block diagram of a secondary subunit for linear grouping calibration of a photovoltaic power station monitoring area state in a distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to an embodiment of the present application. Figure 7As shown, in an embodiment of the present application, the photovoltaic power station monitoring area state linear cluster calibration secondary subunit 1321-2 includes: a photovoltaic power station monitoring area state deep collaborative implicit coding tertiary subunit 1321-21, which is used to construct a photovoltaic power station monitoring area state deep collaborative implicit coding vector between each photovoltaic power station monitoring area state feature vector in the set of the photovoltaic power station monitoring area state feature vector and the photovoltaic power station monitoring area state linear cluster center coding vector; a feature cluster compensation incremental operation operator calculation tertiary subunit 1321-22, which is used to calculate the photovoltaic power station monitoring area state deep collaborative implicit coding vector based on the photovoltaic power station monitoring area state linear cluster center coding vector. code vector, calculates the characteristic cluster compensation incremental operation operator of each photovoltaic power station monitoring area state feature vector in the set of the photovoltaic power station monitoring area state feature vector relative to the photovoltaic power station monitoring area state linear cluster center code vector to obtain multiple photovoltaic power station monitoring area state characteristic cluster compensation incremental operation operators; the photovoltaic power station monitoring area state linear grouping calibration component coding tertiary subunit 1321-23 is used to calculate the photovoltaic power station monitoring area state linear cluster compensation component code vector of the set of the photovoltaic power station monitoring area state feature vector based on the multiple photovoltaic power station monitoring area state characteristic cluster compensation incremental operation operators.

[0046] Specifically, the photovoltaic power station monitoring area state deep collaborative implicit coding three-level sub-unit 1321-21 is used to construct a photovoltaic power station monitoring area state deep collaborative implicit coding vector between each photovoltaic power station monitoring area state feature vector in the set of photovoltaic power station monitoring area state feature vectors and the photovoltaic power station monitoring area state linear cluster center coding vector, which is expressed as follows:

[0047]

[0048] in, It is the first in the set of state feature vectors of the photovoltaic power station monitoring area. The state feature vector of the monitoring area of ​​a photovoltaic power station, is the linear cluster center encoding vector of the photovoltaic power station monitoring area state, For the connection operation, is the first of multiple learnable collaborative weight matrices A learnable collaborative weight matrix, is the first of multiple collaborative bias vectors A collaborative bias vector, is the activation function, yes and Deep collaborative implicit encoding vector of regional status of photovoltaic power station monitoring.

[0049] It's understandable that the core rationale for constructing a deep collaborative implicit encoding vector for PV power plant monitoring area states in intelligent monitoring is to address the shortcomings of linear cluster analysis in modeling complex nonlinear relationships. Specifically, while linear cluster center encoding vectors for PV power plant monitoring area states can extract linear principal components of the global data distribution (such as the group string current mean and temperature baseline), they struggle to capture higher-order interaction patterns between PV power plant monitoring area state feature vectors and cluster centers (such as the dynamic correlation between temperature anomalies in a particular area and adjacent cluster centers). For example, in a distributed power plant, the feature vector of a monitoring area is initially classified as a "normal" center, but its implicit voltage droop trend is nonlinearly coupled with another decaying center (such as battery capacity fluctuations caused by diurnal temperature differences). Linear analysis alone cannot reveal these cross-cluster correlations. Through deep collaborative encoding, the model can exploit implicit contextual relationships between the original features and cluster centers (such as the path of a feature vector's deviation from the cluster center), thereby enhancing sensitivity to progressive degradation (such as microcrack propagation). In this way, a multi-scale feature fusion mechanism is constructed to organically combine the global priors of linear structures with the local dynamics of nonlinear details. For example, during the encoding process, the model performs attention calculations on the feature vector of a high-temperature region and the cluster center of the "hot spot" to capture the difference between the rate of change of the temperature gradient and the central mean. Simultaneously, it incorporates the decay trajectory of similar patterns in historical data (such as the nonlinear pattern of the hot spot area expansion over time) to generate an encoding vector that not only captures the current state but also implicitly reflects the evolutionary trend. This encoding not only preserves the local details of the original features (such as 0.5mm-level subtle crack textures) but also constrains noise interference through the global perspective of the cluster center (such as the temperature distribution statistics of the entire station). For example, it can distinguish temperature fluctuations caused by transient cloud shadows from persistent anomalies of true hot spots. In this way, the original PV power plant monitoring area state feature vector can be incorporated with contextual information relative to the global linear cluster structure, improving the richness and expressiveness of the feature representation, compensating for nonlinear data structures that may not be captured by linear cluster analysis, and providing more refined input for the subsequent feature compensation gain calculation.

[0050] Specifically, the feature cluster compensation incremental operation operator calculation three-level sub-unit 1321-22 is used to calculate the feature cluster compensation incremental operation operator of each photovoltaic power station monitoring area state feature vector in the set of the photovoltaic power station monitoring area state feature vector relative to the photovoltaic power station monitoring area state linear cluster center coding vector based on the photovoltaic power station monitoring area state deep collaborative implicit coding vector to obtain multiple photovoltaic power station monitoring area state feature cluster compensation incremental operation operators. In an embodiment of the present application, the feature cluster compensation incremental operation operator calculation three-level sub-unit 1321-22 is used to: activate the photovoltaic power station monitoring area state linear cluster center encoding vector to obtain the photovoltaic power station monitoring area state linear cluster center encoding vector; perform feature-level equalization correction on the photovoltaic power station monitoring area state linear cluster center encoding vector and the photovoltaic power station monitoring area state deep collaborative implicit encoding vector corresponding to the photovoltaic power station monitoring area state feature vector at a predetermined position to obtain the photovoltaic power station monitoring area state feature clustering intermediate compensation variable; perform complex oscillation interaction correction on the photovoltaic power station monitoring area state feature clustering intermediate compensation variable to obtain the photovoltaic power station monitoring area state feature clustering intermediate correction compensation variable; perform normalization processing on the photovoltaic power station monitoring area state feature clustering intermediate correction compensation variable to obtain the photovoltaic power station monitoring area state feature cluster compensation incremental operation operator corresponding to the photovoltaic power station monitoring area state feature vector at a predetermined position.

[0051] It's understandable that while the initial linear cluster centers (e.g., normal operating conditions, early degradation, and sudden failure)—the linear cluster center encoding vectors for the PV power plant's monitored area state—can capture coarse-grained patterns in global data distribution (e.g., mean temperature and voltage stability), they cannot effectively model dynamic nonlinear deviations between eigenvectors and cluster centers (e.g., the time-varying difference between the periodic diurnal temperature fluctuations in a region and the cluster center mean). For example, in a mountain power plant, two monitoring areas were both classified as "normal" centers in the initial linear clustering. However, one exhibited periodic temperature fluctuations due to shadowing, while the other exhibited voltage droop due to component aging. These nonlinear degradation paths were obscured in the original linear space. By calculating a compensation increment operator, the system quantifies the dynamic offset of each eigenvector from the cluster center (e.g., temperature fluctuation rate, voltage drift acceleration), thereby revealing underlying anomalous evolution trends. The core purpose of this step is to establish a nonlinear correction mechanism that deeply integrates the global benchmark provided by linear clustering with local dynamic details. In other words, the compensation operator is essentially a personalized adjustment strategy for the eigenvector state of each PV power plant's monitored area. For example, in areas where the temperature is slowly rising due to dust accumulation, the model uses a deep collaborative implicit encoding vector to learn the nonlinear relationship between temperature changes and the "normal" center, generating a compensation operator to amplify the deviation weight from the center, allowing the system to identify potential hot spot risks in advance. For temperature anomalies caused by transient cloud cover, the compensation operator suppresses the deviation contribution to avoid misjudgment. This "local nonlinear adjuster" mechanism not only retains the linear clustering's characterization of global stable patterns (such as the station-wide temperature distribution baseline), but also dynamically corrects the boundary between noise interference and real defects through data-driven adaptive compensation (such as self-learning reinforcement based on backpropagation).

[0052] Specifically, the photovoltaic power station monitoring area state linear cluster center code vector is activated to obtain the photovoltaic power station monitoring area state linear cluster center code vector, which is expressed as follows:

[0053]

[0054] in, is the activation function, is the linear cluster center encoding vector of the photovoltaic power station monitoring area state, is the linear cluster center encoding vector of the monitoring area status of the photovoltaic power station.

[0055] Specifically, the photovoltaic power station monitoring area state linear cluster center encoding vector and the photovoltaic power station monitoring area state feature vector at a predetermined position corresponding to the photovoltaic power station monitoring area state deep collaborative implicit encoding vector are subjected to feature-level equalization correction to obtain the photovoltaic power station monitoring area state feature clustering intermediate compensation variable, which is expressed as follows:

[0056]

[0057] in, yes Middle eigenvalues, is the logarithmic function value with base 2, yes Middle eigenvalues, yes and The length of the vector is and Same length, yes The corresponding intermediate compensation variables are clustered based on the state characteristics of the monitoring area of ​​the photovoltaic power station.

[0058] Specifically, the complex oscillation interaction correction is performed on the intermediate compensation variable of the photovoltaic power station monitoring area state feature cluster to obtain the intermediate correction compensation variable of the photovoltaic power station monitoring area state feature cluster, which is expressed as follows:

[0059] ;

[0060] ;

[0061]

[0062] in, represents the vector inner product, yes The corresponding photovoltaic power station monitoring area state feature clustering self-energy feature items, represents the variance of the vector, is a vector length, yes The corresponding photovoltaic power station monitoring area state characteristic self-energy scattering factor, yes The corresponding PV power station monitoring area state characteristic clustering dynamic balance factor, yes The corresponding PV power station monitoring area state feature clustering intermediate correction compensation variables.

[0063] It should be understood that when calculating the intermediate compensation variables of the clustering of the state characteristics of the photovoltaic power station monitoring area, due to the state characteristic vector of the photovoltaic power station monitoring area In addition to the representation of deep collaborative implicit encoding vector The linear clustering results of the incremental nonlinear information generate the addition of clustering elements within the clustering system, which leads to the non-equilibrium state of the cluster space distribution. In other words, the nonlinear compensation variables (such as temperature deviation weight and voltage drift correction coefficient) introduced in the deep collaborative coding process will cause dynamic non-equilibrium in the cluster space. For example, the intermediate compensation variables of a mountain power station show periodic oscillations (such as the temperature compensation value fluctuates in the range of ±3°C) due to the alternating effects of day and night temperature differences and shadows, resulting in a characteristic distribution tearing phenomenon in the molecular regions within the cluster space (such as the junction of the "normal" and "early attenuation" categories) - the monitoring data of the same group of strings are classified into different cluster centers at different time periods, which undermines the stability of the threshold recommendation. This oscillation originates from the imbalance of energy coupling between the nonlinear compensation increment and the linear clustering baseline, and needs to be corrected to eliminate the superposition effect of high-frequency noise interference and low-frequency drift. Therefore, the state characteristic vector of the photovoltaic power station monitoring area is first used and the linear cluster center encoding vector of the photovoltaic power station monitoring area state The inner product of The clustering self-energy characteristic term is constructed to calculate the dynamic balance factor of each fractal subdomain (such as the group string temperature gradient area and the inverter associated voltage area) in the clustering space to represent the scattering energy regulation effect of the fractal structure under the clustering space on the non-integer dimension of the spatial fractal. This can quantify its energy scattering intensity (for example, if the compensation variable in a certain area continuously deviates from the baseline due to salt spray corrosion, the dynamic balance factor is greater than 0.8). Then, considering the self-energy characteristic term Related cluster dynamic balance factors The energy regulation effect under the scattering relationship transfers the energy of high-frequency oscillation components (such as temperature compensation mutations caused by instantaneous cloud cover) to the low-frequency main mode (such as the voltage drop trend caused by battery aging) through a resonance coupling algorithm, achieving spectral reshaping of the compensation variable. This enhances the resonance coupling of the characteristic cluster compensation incremental operation operator under the fractal structure representation of the cluster space, thereby achieving a nonlinear incremental coupling cluster equilibrium state correction effect.

[0064] Specifically, the photovoltaic power station monitoring area state feature cluster intermediate correction compensation variable is normalized to obtain the photovoltaic power station monitoring area state feature cluster compensation increment operation operator corresponding to the photovoltaic power station monitoring area state feature vector at the predetermined position, which is expressed as follows:

[0065] ;

[0066] in, is the normalization function, yes The corresponding photovoltaic power station monitoring area state feature cluster compensation incremental operation operator.

[0067] Specifically, the photovoltaic power station monitoring area state linear grouping calibration component encoding tertiary subunit 1321-23 is used to calculate the photovoltaic power station monitoring area state linear cluster compensation component encoding vector of the set of photovoltaic power station monitoring area state feature vectors based on the multiple photovoltaic power station monitoring area state feature cluster compensation increment operation operators, which is expressed as follows:

[0068] ;

[0069] in, yes The corresponding photovoltaic power station monitoring area state feature cluster compensation incremental operation operator, is the linear cluster compensation component encoding vector of the photovoltaic power station monitoring area state. It should be understood that

[0070] In intelligent monitoring of photovoltaic power plants, the core reason for calculating the linear cluster compensation component encoding vector for the state of a PV power plant's monitored area is to address the interference issues of local noise and redundant information in individual compensation increment operators. Specifically, since the compensation operators (such as temperature deviation weights and voltage drift correction coefficients) in each monitoring area may be affected by transient environmental interference (such as temperature artifacts caused by cloud cover or current fluctuations caused by bird occlusion), direct use in global analysis can easily lead to misjudgments. For example, in a mountain power plant, the compensation operator in one area may show significant temperature deviations, which are actually caused by transient shadows, while the compensation operator in another area exhibits persistent low-amplitude deviations due to component aging, yet the two are difficult to distinguish at the individual level. By aggregating a set of compensation operators, the system can filter out sporadic noise and extract common compensation patterns across regions (such as site-wide temperature gradient anomalies or synchronized string-level voltage decay), thereby enhancing the ability to capture true degradation trends. The goal is to construct a global nonlinear compensation representation that elevates the dispersed individual correction information into a holistic correction framework for the plant's health status. The encoding vector of the linear cluster compensation component for the state of a photovoltaic power plant's monitoring area is essentially a deep fusion of compensation operators across all monitoring areas. For example, through weighted summation using an attention mechanism or principal component analysis, the core compensation dimensions that influence the operation of the entire station (such as battery capacity decay rate and load peak-valley matching deviation) are extracted. In this process, the system automatically suppresses the contribution of local outliers (such as single-point temperature spikes) while amplifying cross-regional consistency signals (such as the synchronous voltage drop of multiple adjacent strings), forming a composite feature representation that retains the linear clustering baseline (such as the health status categories defined by the initial center) while incorporating nonlinear correction factors. For example, a coastal power station identified a rising trend in connector contact resistance caused by salt spray corrosion across the entire station by aggregating compensation components, while individual operators only reflected local point-like anomalies.

[0071] Specifically, the photovoltaic power station full-area state feature fusion secondary subunit 1321-3 is used to fuse the photovoltaic power station monitoring area state linear cluster compensation component encoding vector and the photovoltaic power station monitoring area state linear cluster center encoding vector to obtain the photovoltaic power station full-area state representation vector, which is expressed as follows:

[0072]

[0073] in, is the linear cluster compensation component encoding vector of the photovoltaic power station monitoring area state, and They are and The corresponding trainable weighted hyperparameters can be trained using the directional propagation of gradient descent. It is the state representation vector of the entire photovoltaic power station area.

[0074] In an embodiment of the present application, the photovoltaic power station monitoring information determination subunit 1322 is configured to pass the photovoltaic power station full-region state representation vector through a classifier-based photovoltaic power station state detector to obtain a detection result as the photovoltaic power station monitoring information. It should be understood that, because traditional rule-based or shallow model-based defect recognition methods have difficulty analyzing the high-dimensional nonlinear features output by the feature optimization integration parsing network for significant regional state information, i.e., the photovoltaic power station full-region state representation. In other words, because abnormal conditions in photovoltaic power stations often manifest as complex patterns of multi-source feature coupling (such as the nonlinear correlation between local current distortion caused by microcracks and temperature gradients in adjacent regions), the photovoltaic power station full-region state representation vector incorporates cross-domain correlation features such as inter-component thermodynamic conduction and electrical parameter drift through a feature gain aggregation network. The decision boundary between abnormal modes and normal states in its high-dimensional feature space exhibits strong nonlinear characteristics. Therefore, the photovoltaic power station full-region state representation vector is further passed through a classifier-based photovoltaic power station state detector to obtain a detection result as the photovoltaic power station monitoring information, which indicates whether the operating status of a distributed photovoltaic power station is abnormal.

[0075] In summary, the Internet of Things-based distributed photovoltaic power station online intelligent monitoring system 100 based on the embodiment of the present application is explained, which collects high-resolution remote sensing images and dynamically fuses the hidden state representations of each monitoring area at the feature domain level, establishes a correlation model of key features and a significant information aggregation network, and reconstructs the global operating state map of the photovoltaic power station. Finally, a lightweight classifier is combined to achieve cross-level enhanced expression of defect features, so as to capture the abnormal conduction relationship between components through correlation modeling in the feature space, thereby significantly improving the perception sensitivity of cross-regional defects such as microcracks and hidden hot spots. In this way, high-precision detection of abnormal operating status of distributed photovoltaic power stations is achieved, which helps to improve the defect perception capability of photovoltaic power stations and the intelligence level of distributed photovoltaic power station online monitoring systems.

[0076] As described above, the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things according to the embodiments of the present application can be implemented in various terminal devices. In one example, the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things can be integrated into the terminal device as a software module and / or hardware module. For example, the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things can also be one of the many hardware modules of the terminal device.

[0077] Alternatively, in another example, the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things and the terminal device may also be separate devices, and the distributed photovoltaic power station online intelligent monitoring system 100 based on the Internet of Things may be connected to the terminal device through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0078] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is merely a logical function division, and other division methods may be used in actual implementation.

[0079] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0080] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0081] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0082] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0083] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. Second-order terms are used to indicate names and do not imply any particular order.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit of the technical solutions of the present invention.

Claims

1. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things is characterized by: include: The UAV monitoring information generation module is used to obtain remote sensing images of the entire photovoltaic power station area and generate UAV monitoring paths and monitoring points; A position deviation calculation module is used to control the UAV to fly along the monitoring path and determine the position deviation based on the current position of the UAV and the position of the monitoring point; The photovoltaic power station status monitoring and analysis module is used to collect remote sensing images of the monitoring area through the camera on the drone if the position deviation is less than or equal to the predetermined threshold, generate photovoltaic power station monitoring information based on the remote sensing images of the monitoring area, and then transmit the photovoltaic power station monitoring information to the cloud server; The photovoltaic power station state monitoring and analysis module includes: a power station monitoring area state feature extraction unit, which is used to capture the monitoring area state features of remote sensing images of multiple monitoring areas in the photovoltaic power station to obtain a set of photovoltaic power station monitoring area state features; a power station state monitoring unit, which is used to perform power station state monitoring based on regional state significant feature aggregation analysis on the set of photovoltaic power station monitoring area state features to obtain photovoltaic power station monitoring information; The power station status monitoring unit includes a feature aggregation subunit for regional status significant information and a photovoltaic power station monitoring information determination subunit; The feature aggregation subunit of the regional state salient information includes: The secondary sub-unit of the initial linear clustering of the photovoltaic power station monitoring area state is used to perform linear cluster analysis on the set of the photovoltaic power station monitoring area state feature vectors to obtain the photovoltaic power station monitoring area state linear cluster center encoding vector, which is expressed as follows: in, is the set of state feature vectors of the photovoltaic power station monitoring area, , , and They are the first, second, and third state feature vectors in the set of monitoring area state feature vectors of photovoltaic power station. and The state feature vector of the monitoring area of ​​a photovoltaic power station, yes The number of vectors in is the linear cluster center encoding vector of the monitoring area state of the photovoltaic power station; A photovoltaic power station monitoring area state linear grouping calibration secondary subunit is used to perform grouping calibration on the set of photovoltaic power station monitoring area state feature vectors to obtain a photovoltaic power station monitoring area state linear cluster compensation component encoding vector; The photovoltaic power station full area state feature fusion secondary subunit is used to fuse the photovoltaic power station monitoring area state linear cluster compensation component encoding vector and the photovoltaic power station monitoring area state linear cluster center encoding vector to obtain the photovoltaic power station full area state representation vector.

2. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 1 is characterized in that: The power station monitoring area state feature extraction unit includes: a multi-monitoring area remote sensing image acquisition subunit, configured to acquire remote sensing images of multiple monitoring areas of the photovoltaic power station through the camera to obtain a set of remote sensing images of the monitoring areas; A monitoring area remote sensing image enhancement preprocessing subunit, configured to perform image enhancement preprocessing on each monitoring area remote sensing image in the set of monitoring area remote sensing images to obtain a set of enhanced monitoring area remote sensing images; The monitoring area state feature extraction subunit is used to extract monitoring area state features from each enhanced monitoring area remote sensing image in the set of enhanced monitoring area remote sensing images to obtain a set of monitoring area state features of the photovoltaic power station.

3. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 2 is characterized in that: The monitoring area state feature extraction subunit is used to: pass each enhanced monitoring area remote sensing image in the set of enhanced monitoring area remote sensing images through a photovoltaic power station monitoring area state feature extractor based on ConvNeXt to obtain a set of photovoltaic power station monitoring area state feature vectors as the set of photovoltaic power station monitoring area state features.

4. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 3 is characterized in that: The power station status monitoring unit includes: A feature aggregation subunit of regional state significant information is used to obtain a state representation vector of the entire photovoltaic power station region by subjecting the set of the photovoltaic power station monitoring region state feature vectors to a feature optimization integration parsing network of regional state significant information; The photovoltaic power station monitoring information determination subunit is used to perform photovoltaic power station status detection based on the photovoltaic power station full area status characterization vector to determine photovoltaic power station monitoring information, wherein the photovoltaic power station monitoring information is used to indicate whether there is any abnormality in the operating status of the distributed photovoltaic power station.

5. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 1 is characterized in that: The photovoltaic power station monitoring area state linear grouping calibration secondary subunit includes: The three-level sub-unit of deep collaborative implicit coding of the photovoltaic power station monitoring area state is used to construct a photovoltaic power station monitoring area state deep collaborative implicit coding vector between each photovoltaic power station monitoring area state feature vector in the set of photovoltaic power station monitoring area state feature vectors and the photovoltaic power station monitoring area state linear cluster center coding vector, which is expressed as follows: in, It is the first in the set of state feature vectors of the photovoltaic power station monitoring area. The state feature vector of the monitoring area of ​​a photovoltaic power station, is the linear cluster center encoding vector of the photovoltaic power station monitoring area state, For the connection operation, is the first of multiple learnable collaborative weight matrices A learnable collaborative weight matrix, is the first of multiple collaborative bias vectors A collaborative bias vector, is the activation function, yes and Deep collaborative implicit coding vector of the photovoltaic power station monitoring area status between them; a feature cluster compensation incremental operation operator calculation tertiary subunit, configured to calculate, based on the photovoltaic power station monitoring area state deep collaborative implicit coding vector, a feature cluster compensation incremental operation operator of each photovoltaic power station monitoring area state feature vector in the set of the photovoltaic power station monitoring area state feature vectors relative to the photovoltaic power station monitoring area state linear cluster center coding vector to obtain a plurality of photovoltaic power station monitoring area state feature cluster compensation incremental operation operators; The photovoltaic power station monitoring area state linear grouping calibration component encoding tertiary subunit is used to calculate the photovoltaic power station monitoring area state linear cluster compensation component encoding vector of the set of photovoltaic power station monitoring area state feature vectors based on the multiple photovoltaic power station monitoring area state feature cluster compensation increment operation operators.

6. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 5 is characterized in that: The feature cluster compensation incremental operation operator calculates the third-level sub-unit, which is used to: Activating the linear cluster center encoding vector of the photovoltaic power station monitoring area state to obtain a linear cluster center encoding vector of the photovoltaic power station monitoring area state; Performing feature-level equalization correction on the photovoltaic power station monitoring area state linear cluster center encoding vector and the photovoltaic power station monitoring area state deep collaborative implicit encoding vector corresponding to the photovoltaic power station monitoring area state feature vector at a predetermined position to obtain a photovoltaic power station monitoring area state feature clustering intermediate compensation variable; The photovoltaic power station monitoring area state feature cluster intermediate compensation variable is subjected to complex oscillation interaction correction to obtain the photovoltaic power station monitoring area state feature cluster intermediate correction compensation variable, which is expressed as: ； ； in, represents the vector inner product, yes The corresponding photovoltaic power station monitoring area state feature clustering self-energy feature items, represents the variance of the vector, is a vector length, yes The corresponding photovoltaic power station monitoring area state characteristic self-energy scattering factor, yes The corresponding PV power station monitoring area state characteristic clustering dynamic balance factor, yes The corresponding PV power station monitoring area state feature clustering intermediate correction compensation variables; Normalization processing is performed on the photovoltaic power station monitoring area state feature clustering intermediate correction compensation variable to obtain a photovoltaic power station monitoring area state feature cluster compensation increment operation operator corresponding to the photovoltaic power station monitoring area state feature vector at a predetermined position.

7. The distributed photovoltaic power station online intelligent monitoring system based on the Internet of Things according to claim 6 is characterized in that: The photovoltaic power station monitoring information determination subunit is configured to: pass the photovoltaic power station full-area state characterization vector through a photovoltaic power station state detector based on a classifier to obtain the photovoltaic power station monitoring information.

Images