A Smart Inspection and Control Method for the Operation and Maintenance of Photovoltaic Power Plants

By collecting data from photovoltaic equipment using drones, identifying and optimizing inspection paths, the problems of low operation and maintenance efficiency and insufficient environmental adaptability of traditional photovoltaic power plants have been solved, achieving high-precision and high-speed photovoltaic power plant detection and fault location.

CN123117499APending Publication Date: 2026-11-17CHINA ENERGY ENG GRP TIANJIN ELECTRIC POWER CONSTR CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611621083.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional photovoltaic power plant operation and maintenance relies on manual inspections, which is inefficient and makes it difficult to achieve high-frequency and precise inspections of large-area power plants. Furthermore, existing drone inspection technology has shortcomings in terms of environmental adaptability and fault location.

Method used

By collecting environmental data and output power of photovoltaic equipment using drones, target areas can be identified, the power generation attenuation process can be deduced, hot spot areas can be automatically identified, and inspection paths can be updated. Combined with infrared thermal imaging spectra and temperature gradient characteristics, the inspection paths can be optimized to improve detection accuracy and efficiency.

Benefits of technology

It achieves environmental adaptability and fault-oriented inspection of photovoltaic power plants, improves the targeting and execution efficiency of detection, ensures high detection rate and low missed detection rate, and realizes multi-UAV task load balancing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117499A_ABST
    Figure CN123117499A_ABST
Patent Text Reader

Abstract

This invention relates to the field of drone inspection technology, specifically an intelligent inspection and control method for photovoltaic power plant operation and maintenance. The method includes: identifying all photovoltaic devices within the inspection path's range based on the drone's previous inspection path; collecting environmental data and output power of the photovoltaic devices and determining target areas for each scenario; deducing the power generation attenuation process of the photovoltaic devices based on their output power within the target area and selecting candidate inspection points that characterize the risk of power generation attenuation; acquiring infrared thermal imaging maps of the areas corresponding to the candidate inspection points, automatically identifying and extracting hotspot areas from the maps; updating the risk information of the candidate inspection points based on the principal axis direction and temperature gradient of the hotspot areas; and iteratively setting the inspection path for the current round based on the risk information of the candidate inspection points. This improves the efficiency and priority of photovoltaic power plant inspections.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) inspection technology, specifically a method for intelligent inspection and control of photovoltaic power plant operation and maintenance. Background Technology

[0002] Traditional photovoltaic (PV) power plant operation and maintenance (O&M) mainly relies on manual on-site inspections, which suffers from drawbacks such as long inspection cycles and high labor costs, and makes it difficult to achieve high-frequency, high-precision inspections of large-area power plants. In recent years, drone inspection technology, equipped with infrared thermal imaging and visible light payloads, has gradually become widespread. It can quickly collect surface temperature data and image data of PV modules, significantly improving inspection efficiency and has become the mainstream technical means for PV power plant O&M.

[0003] For example, Chinese Patent Publication No. CN118012100A discloses a method and system for intelligent photovoltaic power station inspection using drones, which relates to the technical field of photovoltaic power stations. The method includes: obtaining an inspection route based on the distribution of several photovoltaic power stations; obtaining the inspection workload based on the coverage area of ​​each photovoltaic power station; matching a number of drones based on the inspection workload, inspection route, drone flight capability, and total inspection time; the total inspection time corresponding to the daily operating time of the photovoltaic power station; sending the inspection route and regional information of the photovoltaic power stations in the inspection route to the drone control system to trigger the drone to start flight; receiving images collected by the drone; and determining whether there are factors affecting the operation of the photovoltaic power station based on the images.

[0004] For example, Chinese Patent Publication No. CN120560305A discloses a method, system, medium, and program product for unmanned aerial vehicle (UAV) inspection of photovoltaic power stations, relating to the general field of control or regulation systems. The method includes: controlling a lightweight UAV to perform inspections according to a preset detection route, acquiring real-time inspection images and actual flight parameters; determining the real-time inspection progress and fault detection areas based on the real-time inspection images, and calculating the predicted inspection time; determining multiple target inspection areas including historical fault areas and fault detection areas, and constructing a real-time re-inspection route; calculating the equivalent flight parameters of a standard UAV based on the actual flight parameters of the lightweight UAV during inspection, and calculating the predicted re-inspection time for the standard UAV to complete its inspection on the real-time re-inspection route; when the predicted re-inspection time is greater than or equal to the predicted inspection time, controlling the standard UAV to perform inspections according to the real-time re-inspection route, obtaining inspection re-inspection images, and generating an inspection report. Existing technologies determine the set inspection route based on the spatial location of the photovoltaic power station; or detect faults in real-time through inspection images, and determine the UAV inspection report based on the inspection time and re-inspection time. In these processes, the inspection area targets lack an environmentally adaptive judgment process, and the identified relative anomalies do not distinguish between power generation attenuation anomalies and hot spot distribution anomalies, resulting in the detection results being greatly affected by external factors; it is difficult to locate the location of the anomaly and the inspection route after the anomaly occurs, leading to problems such as low analysis accuracy, poor environmental adaptability and insufficient fault direction in power plant scenarios. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a photovoltaic power plant operation and maintenance intelligent inspection and control method, including: S1, based on the inspection path of the UAV in the previous round, identifying all photovoltaic equipment within the range of the inspection path; collecting environmental data and output power of photovoltaic equipment, and determining the target area under each scenario; the environmental data includes the irradiance, ambient temperature and average temperature of the module backsheet in the current time period.

[0006] S2, based on the output power of photovoltaic equipment in the target area, deduces the power generation degradation process of photovoltaic equipment and selects candidate inspection points that characterize the risk of power generation degradation.

[0007] S3: Obtain the infrared thermal imaging spectrum of the area corresponding to the candidate inspection point, automatically identify and extract the hot spot area in the spectrum; update the risk information of the candidate inspection point according to the main axis direction of the hot spot area and the temperature gradient.

[0008] S4, based on the risk information of candidate inspection points, iteratively sets the inspection path for the current round.

[0009] The beneficial effects of this invention are as follows: First, this invention defines the drone's path range based on the previous round of inspections, limiting the detection targets to photovoltaic equipment within the path's coverage area; it collects three types of data: irradiance, ambient temperature, and average backsheet temperature of the modules; it sets an environmental baseline based on irradiance and temperature, and calculates the corresponding theoretical backsheet temperature; it clusters photovoltaic equipment according to the environmental baseline intervals, grouping spatially adjacent equipment with the same operating conditions within the same interval into the same candidate target area; it compares the backsheet temperature residuals within the candidate target areas, and outputs the final target area when the residual exceeds the limit. This eliminates the interference of environmental factors on photovoltaic judgment in power plant scenarios, effectively narrowing the scope of subsequent detection and inspection, and improving the targeting and execution efficiency of inspections.

[0010] II. This invention constructs a power vector from the output power of a target area at multiple moments, and uses a clustering algorithm to obtain the cluster center of the area as a reference power. It then combines irradiance to screen effective sampling points, statistically analyzes the proportion of outliers among the effective sampling points, and quickly identifies short-term power anomalies to generate initial candidate inspection points. Next, a secondary verification is performed based on the power generation efficiency decay trend to distinguish between scenarios where power anomalies and decay trends are consistent. Further screening is done by combining the number of detections at the same location, outputting the final candidate inspection points. This two-stage processing ensures a high detection rate for short-term instantaneous faults, avoiding missed detections of instantaneous anomalies due to pure power generation efficiency, while also distinguishing between disturbances in power generation decay scenarios and equipment performance degradation, improving the accuracy and fault directionality of candidate inspection points.

[0011] Third, this invention uses infrared thermal imaging of the corresponding area of ​​candidate inspection points to binarize and extract high-temperature pixels to obtain hot spot regions. It then uses temperature-weighted centroid and weighted covariance matrix eigenvalue decomposition to fit the minimum bounding rectangle of the hot spot region, determining the main axis direction of the hot spot extension. By comparing the main axis with the photovoltaic module arrangement direction, the physical location and distribution pattern of the hot spot connectivity are determined. Subsequently, orthogonal dual temperature profiles are constructed along the hot spot main axis and perpendicular to the main axis, and the temperature gradient is calculated pixel-by-pixel. The maximum gradient, average gradient, gradient interval ratio, and time-series gradient change are extracted to form temperature gradient features. Combining the physical location of the hot spot connectivity and the temperature gradient features, the risk information of the candidate inspection points is updated. This quantifies the extension direction and spatial distribution of the hot spots, making the scene description of the hot spots and photovoltaic equipment more comprehensive, achieving refined description of hot spot identification and scene distribution recognition, and providing data basis for subsequent path planning.

[0012] IV. This invention expands the inspection set outward from candidate inspection points to include spatially adjacent photovoltaic devices, covering areas with potential fault spread. It then backtracks all candidate inspection points, adjusting risk weights based on power and temperature anomalies. An improved K-means spatial clustering method is employed: feature vectors are constructed using spatial coordinates and risk weights, and spatially dispersed high-risk points are selected as initial cluster centers to divide inspection sub-regions corresponding to the number of drones. Within each sub-region, ant colony optimization iterations are performed independently, integrating flight distance and point risk weights to generate inspection paths for each sub-region's drones. This avoids concentrating high-risk tasks, achieves balanced workload across multiple drones, and improves overall inspection efficiency and fault handling timeliness. Attached Figure Description

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] Figure 1 This is a flowchart illustrating an intelligent inspection and control method for the operation and maintenance of a photovoltaic power plant. Figure 2 This is a flowchart illustrating step S1 of an intelligent inspection and control method for the operation and maintenance of a photovoltaic power plant. Figure 3 This is a flowchart illustrating step S2 of an intelligent inspection and control method for the operation and maintenance of a photovoltaic power plant. Figure 4 This is a flowchart illustrating step S3 of a photovoltaic power plant operation and maintenance intelligent inspection and control method. Figure 5 This is a flowchart illustrating step S4 of a photovoltaic power plant operation and maintenance intelligent inspection and control method. Detailed Implementation

[0015] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0016] See Figure 1A method for intelligent inspection and control of photovoltaic power plant operation and maintenance includes: S1, based on the inspection path of the UAV in the previous round, identifying all photovoltaic equipment within the range of the inspection path; collecting environmental data and output power of the photovoltaic equipment, and determining the target area under each scenario; wherein, the environmental data includes the irradiance, ambient temperature and average temperature of the module backsheet at the current time, to interpret the environment in which the equipment operates; the target area is the area where the environment deviates from the overall baseline. Such areas often correspond to photovoltaic equipment clusters with severe shading, poor ventilation and heat dissipation, and high historical failure rates. Marking the area where such equipment is located as the target area in advance allows the subsequent UAV infrared inspection to directly focus on the points most likely to have hot spots and microcracks, reducing the missed detection rate in subsequent processing and avoiding misjudging the global environmental temperature rise as a local module failure.

[0017] In this step, based on the pre-planned drone inspection path, all photovoltaic devices within the coverage area of ​​the path are located. Different operating scenarios are divided according to the ambient temperature conditions, and target areas that need to be inspected in the power station are selected, providing spatial object input for subsequent temperature field analysis, power generation attenuation assessment, and dynamic path correction.

[0018] like Figure 2 As shown, one implementation of step S1 includes: S11 sets the drone's travel range based on the image range captured by the drone and the waypoint locations along the inspection path.

[0019] When setting the travel range, the drone is the center, the horizontal direction is the field of view of a single frame of the image, and the vertical direction covers the space from the bottom of the ground support to the drone's flight altitude, so that it forms the drone's travel range along the inspection path.

[0020] S12: For all photovoltaic devices within the route range, set an environmental baseline based on irradiance and ambient temperature, and solve for the theoretical backsheet temperature corresponding to the environmental baseline. The environmental baseline refers to the irradiance and ambient temperature at that location; for any photovoltaic device, the measured ambient temperature and irradiance at that location determine the backsheet temperature that should be achieved under unobstructed conditions.

[0021] Furthermore, historical normal operation data of this type of equipment in the power station is extracted, and the temperature rise coefficient corresponding to the irradiance is fitted according to the historical data. This temperature rise coefficient represents the temperature rise of the module relative to the ambient temperature under unit irradiance. The product of the temperature rise coefficient and the irradiance is used as the temperature rise of the photovoltaic equipment. This temperature is added to the ambient temperature to obtain the theoretical backsheet temperature.

[0022] S13, perform interval clustering of photovoltaic equipment according to environmental baseline to form candidate target areas.

[0023] When performing interval clustering of photovoltaic devices based on environmental baselines, the range of values ​​for irradiance and ambient temperature is divided into multiple intervals based on the quantiles of irradiance and ambient temperature in historical data. The interval in which each photovoltaic device falls is detected, and photovoltaic devices that are spatially adjacent and fall into the same interval are considered as a candidate target area.

[0024] When setting the irradiance range, the historical irradiance of the photovoltaic equipment is sorted from smallest to largest. Based on the quartiles, the 25th quartile, 50th quartile, and 75th quartile are selected, and the range is divided into four intervals according to the quartile values, forming four scenarios: weak light, low irradiance, medium irradiance, and high irradiance.

[0025] Similarly, for ambient temperature, the 50th and 75th percentiles are selected according to the temperature of the scene to divide it into three scenarios: low temperature, normal temperature, and high temperature. Multiple intervals are formed based on the combination of irradiance and ambient temperature.

[0026] S14. Within the same candidate target area, the average backsheet temperature of each photovoltaic device is compared with the theoretical backsheet temperature using residuals. If the residual exceeds the limit, the corresponding candidate target area is used as the output target area. Exceeding the residual limit indicates that the candidate target area where the device is located deviates from the environmental baseline and is therefore a key target for UAV inspection. If the backsheet temperature is lower than the theoretical backsheet temperature, it is generally due to sampling error and is not used as a selection criterion for the target area.

[0027] In one implementation scenario, based on normally operating samples within the same environmental baseline range, the average value ± 3 times the standard deviation is selected to set the range of normal residual values. When the temperature residual exceeds the upper limit of this range, it is considered that the residual exceeds the limit, and the current candidate target area and its related photovoltaic equipment are output.

[0028] S2, based on the output power of photovoltaic equipment within the target area, extrapolates the power generation degradation process of the photovoltaic equipment and selects candidate inspection points that characterize the risk of power generation degradation. In this step, the photovoltaic equipment is first compared longitudinally over time by the differences in output power at various times, and then the degradation trends of each device are compared to finally select candidate inspection points.

[0029] like Figure 3 As shown, one implementation of step S2 includes: S21. Calculate the power generation efficiency for each time period based on the output power at each moment. The power generation efficiency is defined by the PR value, which refers to the photovoltaic utilization rate. It is a ratio of the actual power generation capacity of the photovoltaic equipment to the theoretical maximum power generation capacity and is used to reflect the equipment's own losses.

[0030] The actual power generation capacity of photovoltaic (PV) equipment is determined by defining the cumulative output power per unit time. The theoretical maximum power generation capacity is determined by calculating the ratio of the current irradiance to the standard irradiance, multiplying this ratio by the cumulative rated power of the PV equipment per unit time, and obtaining the theoretical maximum power generation capacity. The standard irradiance is selected as 1 kW / m² as the reference irradiance intensity, and the power generation efficiency at each time period is quantified.

[0031] S22, based on the difference in output power at each moment, performs a preliminary inspection of the photovoltaic equipment in the time dimension and quickly generates candidate inspection points; the difference in output power represents the instantaneous power decay process, which is the basis for capturing instantaneous anomalies and generating candidate inspection points, and is used to adapt to scenarios of hot spot generation and short-term faults.

[0032] Preliminary time-based inspection of photovoltaic equipment includes: The output power of any target area at several times is obtained. For each photovoltaic device, the output power within the same time window is extracted and converted into a vector of corresponding values. The cluster center of the output power is obtained through a clustering algorithm, and the value corresponding to the cluster center is used as the reference power.

[0033] This clustering method divides the data into two categories: one representing normal power sample clusters and the other representing low-power abnormal clusters. The cluster center of each cluster is calculated, and the cluster center of the cluster with the largest sample size is selected. The power value of the center vector at each time moment is extracted as the reference power. The reference power is calculated from the feature mean of all samples in the cluster. It is the most representative feature point in the corresponding scenario and can accurately reflect the relative distribution of instantaneous output power at a specific moment. The photovoltaic equipment with short-term abnormality can be extracted through the reference power.

[0034] Based on the irradiance values ​​at each moment, the number of valid sampling points within the current time period is determined. Valid sampling points are the instantaneous power with irradiance exceeding a set threshold within the current time period. Conditions with weak power generation signals and unclear power characteristics under low irradiance conditions are excluded, and data from stable power generation scenarios are retained as input values ​​for subsequent deviation statistics and anomaly analysis. The threshold set here can be the 50th percentile of irradiance from the previous step, or the average irradiance corresponding to the reference power, to obtain instantaneous changes with clear power characteristics.

[0035] Based on the relative difference between the output power and the reference power at each moment, the total number of current equipment anomalies is counted; for the output power at each moment, an anomaly threshold for the relative difference is set according to the average value and standard deviation of the relative difference; when the relative difference is lower than the anomaly threshold, it means that the output power is much lower than the reference power, and there is a short-term power anomaly in photovoltaic power generation at the current moment.

[0036] In one implementation, the average of the relative differences minus twice the standard deviation is used as the set anomaly threshold. Data points smaller than the anomaly threshold are selected as anomalies for the current device.

[0037] Based on the proportion of abnormal points to the number of valid sampling points, it is determined whether the current device is abnormal, and the geographical coordinates of the abnormal devices are used to generate candidate inspection points.

[0038] When determining whether the current equipment is abnormal, if the proportion of abnormal points to the number of valid sampling points is greater than the proportion threshold, the photovoltaic equipment is determined to be abnormal in a short period of time; if it is less than the proportion, it means that the abnormality is a false warning caused by single-point noise or cloud disturbance in a short period of time. Only when the abnormal points continue to account for a certain proportion over time is the photovoltaic equipment's power generation process considered to be abnormal.

[0039] In one implementation scenario, the false alarm rate and false alarm rate of photovoltaic equipment failures are statistically analyzed in historical data. When the false alarm rate is less than or equal to 10% and the false alarm rate is less than or equal to 5%, the ratio of outliers to valid sampling points in the corresponding data is used as the proportion threshold to obtain the number of outliers in scenarios with low false alarm probability.

[0040] S23. For each candidate inspection point, a secondary inspection is performed according to the decay trend of the power generation efficiency of each device to determine the output candidate inspection point.

[0041] The decline in power generation efficiency represents a long-term screening of the equipment's performance. If only power generation efficiency is used to identify the decline, the instantaneous abnormal quantity will be diluted by time integration, causing the system to ignore the instantaneous abnormality corresponding to short-term faults and continuous losses.

[0042] A secondary inspection was conducted based on the decline trend of the power generation efficiency of each device, including: Based on the decline trend of power generation efficiency, determine whether the decline trend points to abnormal equipment performance. If so, output the corresponding candidate inspection point directly.

[0043] When determining the attenuation trend, the power generation efficiency for each time period is calculated based on the effective sampling points of the output power detection and the values ​​taken at multiple time periods.

[0044] Then, using the reference power, the power generation efficiency for the corresponding time period is calculated and used as the reference power generation efficiency. When the slope value of the fitted power generation efficiency for each time period is less than the slope value of the reference power generation efficiency, it is considered that there is a significant long-term decay and damage to the equipment itself. Drone inspections are then arranged for the corresponding candidate inspection points.

[0045] Otherwise, count the cumulative number of times the same location in the current target area is identified as a candidate inspection point, and output the candidate inspection points with a cumulative count greater than 1 directly.

[0046] If the power generation efficiency does not directly correspond to equipment performance abnormalities, it means that the slope value of the fitted power generation efficiency is greater than or equal to the slope value of the reference power generation efficiency, and there is no significant overall power generation efficiency degradation. The short-term anomalies tend to be in the form of intermittent faults, and the number of detections in the corresponding time period needs to be recorded. If the number is 1, it means that the corresponding location is an intermittent fault, and the number of detections in multiple time periods is very low. If the number is greater than 1, it means that the anomaly is detected in multiple time periods. Even if the photovoltaic equipment itself is in good working order, it is still necessary to check whether other anomalies have occurred in the equipment at that location to complete the inspection of the photovoltaic power station.

[0047] S3: Acquire infrared thermal imaging maps of the corresponding areas of the candidate inspection points, automatically identify and extract hot spot regions from the maps; update the risk information of the candidate inspection points based on the principal axis direction and temperature gradient of the hot spot regions; the infrared thermal imaging maps are based on the infrared thermal imaging maps collected in the previous round of UAV inspections and are used to plan the process of the next UAV inspection. The extracted hot spot regions are localized abnormal high-temperature areas on the equipment surface that are completely different from the normal uniform temperature field. They are localized high-temperature locations of photovoltaic equipment under normal operating conditions and can indicate abnormal operating conditions of the photovoltaic equipment.

[0048] like Figure 4 As shown, one implementation of step S3 includes: S31: The infrared thermal imaging spectrum is binarized, and pixels with temperatures higher than the regional baseline temperature are extracted and combined into hot spot regions. The regional baseline temperature refers to the minimum threshold for photovoltaic infrared thermal imaging detection. When the temperature at the corresponding location is lower than this value, the overall heat dissipation efficiency of the photovoltaic device is extremely high, the operating temperature rise is greatly suppressed, and the regional differences of hot spots are severely weakened, making it impossible to clearly identify early minor hot spots. Only when the temperature is higher than this value can the local differences of hot spots be clearly identified, ensuring the reliability of hot spot analysis.

[0049] In one implementation scenario, the regional baseline temperature is set to 5 degrees Celsius, and areas with temperatures higher than this are designated as hotspot areas at the locations of candidate inspection points.

[0050] S32, perform minimum bounding rectangle fitting on the hot spot region to determine the principal axis direction of the hot spot region extension.

[0051] Considering that the hot spot region is affected by the hot spot distribution and is prone to exhibiting an asymmetrical distribution pattern, it is necessary to lock the center of the hot spot region according to the temperature difference within the region, so that the minimum bounding rectangle corresponds to the center of the hot spot region, thereby improving the adaptability of the minimum bounding rectangle to the temperature trend within the hot spot.

[0052] Specifically, when fitting the minimum bounding rectangle to the hot spot region, the implementation methods include: For all pixels within the hotspot area, the difference between the temperature of each pixel and the lowest temperature in the hotspot area is used as the weight of that pixel, with pixels having a higher temperature having a higher weight.

[0053] The coordinates of all pixels are weighted and averaged to obtain the weighted average coordinates. These weighted average coordinates are then used as the weighted centroid of the hot spot region. This weighted centroid is automatically biased toward the hot spot region to locate it.

[0054] Calculate the weighted covariance matrix of all pixels relative to the weighted centroid; perform eigenvalue decomposition on the weighted covariance matrix to obtain two eigenvectors. The largest eigenvector is used as the principal axis direction for the hotspot region extension. This direction is dominated by pixels with high weights and is not affected by low-temperature edges, thus improving the accuracy of hotspot orientation recognition. Essentially, this process utilizes the temperature difference within the hotspot region as weights. These weighted values ​​form the covariance matrix, which is then decomposed into two eigenvectors using principal component decomposition. The largest eigenvector is used to select the primary extension direction, rotating the smallest bounding rectangle to the principal axis of temperature extension.

[0055] Using the weighted centroid of the hotspot region as the origin, the principal axis of the hotspot region as the horizontal axis, and the direction perpendicular to the principal axis as the vertical axis, the minimum bounding rectangle for fitting is determined. Using this weighted centroid as the origin, all pixels in the entire hotspot region are rotated so that the principal axis coincides with the horizontal axis of the coordinate system, and the direction perpendicular to the principal axis becomes the vertical axis. Within this coordinate system, all pixels in the hotspot region are traversed, and the maximum and minimum pixel coordinates in the horizontal and vertical axes are extracted respectively. These pixel coordinates are then used as the length and width sides of the rectangle, thus completing the fitting of the minimum bounding rectangle and determining the principal axis direction.

[0056] S33, compare the direction of the main shaft with the arrangement direction of the internal components of the photovoltaic equipment to determine the physical location of the hot spot region connection; The comparison of hotspot arrangement directions includes isolated points without a clear extension direction, directions along the edge of a single cell, directions longitudinally connected along the cell stringing direction, and directions extending across multiple cells or even across the entire module. These directions explain the distribution location of hotspots and are the shapes corresponding to the hotspot regions. The comparison process involves matching the hotspot's main axis with the module's cell arrangement direction to determine the physical location of the hotspot, thereby determining the range of influence of the hotspot and the physical location of that range.

[0057] S34, construct temperature profiles along the main axis direction and perpendicular to the main axis direction of the hot spot region respectively, and extract the temperature gradient features corresponding to each temperature profile; The constructed temperature profile involves using the previously fitted minimum bounding rectangle to lock the weighted centroid of the hotspot region. A first linear slice is made along the main axis of the hotspot's extension, extracting continuous temperature values ​​from all pixels along this tangent line to obtain the temperature profile along the main axis. A second linear slice is then made, starting from the hotspot center and perpendicular to the main axis, to extract the second temperature profile perpendicular to the main axis. Statistical analysis is performed on the temperature values ​​of the two profiles, including but not limited to features such as maximum gradient magnitude, average gradient magnitude, and gradient magnitude interval proportion. These features are summarized as the temperature gradient features of each profile, and then, based on the physical location of the hotspot region, they are summarized as risk information for candidate inspection points. During gradient calculation, a central difference method is used to calculate the temperature values ​​of the two profiles pixel-by-pixel, obtaining the gradient magnitude of different pixels. Then, the gradient magnitude interval proportion is determined according to the proportion of pixels occupied by different gradient magnitude intervals, thus completing the statistical analysis of the gradient features.

[0058] When extracting the temperature gradient features corresponding to each temperature profile, the implementation methods include: Based on the infrared thermal imaging spectrum over a continuous time period, the temperature gradient magnitude is calculated pixel by pixel at each temperature profile. The gradient magnitudes of all pixels on the profile are then arranged in order to obtain the temperature gradient sequence at the current moment.

[0059] The maximum gradient magnitude, average gradient magnitude, and gradient magnitude interval percentage corresponding to the temperature gradient sequence at the current moment are obtained to generate the gradient vector at each moment. Among them, the maximum gradient magnitude is the maximum value in the sequence, which can represent the intensity of the temperature jump of the hot spot at that moment and reflect the concentration of the heat source. The average gradient magnitude is the arithmetic mean of all values ​​in the sequence, which represents the overall heating intensity of the hot spot at that moment. The gradient magnitude interval percentage divides the gradient magnitude into three intervals: low gradient, medium gradient, and high gradient. The proportion of pixels in each interval to the total number of pixels in the profile is calculated to reflect the relative structure of the hot spot temperature distribution.

[0060] In one implementation, the temperature gradient amplitude of the hot spot region sample is extracted, sorted in ascending order, and the 50th and 75th percentiles are selected to set three intervals: low gradient, medium gradient, and high gradient. The relative distribution of the temperature gradient of each pixel is recorded.

[0061] Guided by the gradient vectors at each time step, the difference between the gradient vector at the initial inspection time is calculated to obtain the gradient change at each time step. The trend of the gradient change is then used as the temperature gradient feature at the current time step. Here, the initial inspection time refers to the initial time point when the previous round of inspection began, and it represents the change pattern of a single hotspot region over a continuous period of time.

[0062] In this scenario, if the difference in gradient vectors shows a continuously increasing trend, it indicates that the temperature gradient of the hot spot is changing more and more, the heat source concentration is continuously increasing, and the hot spot is in a state of rapid deterioration. If it remains relatively stable, it indicates that the heating state of the equipment has not changed significantly, and there is only a problem of power generation attenuation. If the difference in gradient vectors continues to decrease, it indicates that the hot spot temperature is gradually dropping, and the abnormal heating is an instantaneous heating caused by shading and installation conditions, which is a temporary hot spot change.

[0063] Specifically, from the data samples of normal operation of photovoltaic equipment, the difference corresponding to the gradient vector at each time moment is extracted, and the average value and standard deviation of the difference are selected to set a confidence interval, which is the range of the average value ± 3 times the standard deviation. If the difference of the gradient vector at a certain time moment exceeds the upper limit of the interval and is increasing, its continuous increase is regarded as a continuously increasing trend. If it is always within the confidence interval, it is regarded as a relatively stable trend. If it gradually falls from the upper limit of the interval and continuously decreases into the confidence interval, it is regarded as a continuously decreasing trend. The trends of the gradient vector at each time moment are summarized into the gradient features at the corresponding time moment to explain the evolution process under a specific gradient value. Combined with the actual value and physical location of the temperature gradient, the risk update of each candidate inspection point is completed.

[0064] S35, based on the extracted temperature gradient features and the physical location of the hot spot regions, update the risk information of the candidate inspection points. After obtaining the temperature gradient features and their physical locations, process the hot spot regions of each candidate inspection point separately to generate risk information for different candidate inspection points.

[0065] It should be noted that the risk information not only includes the relative risk description of the temperature gradient change trend and the hot spot area orientation, but also covers the risk description of power generation attenuation in step S2. These descriptions will be used as the data content recorded at each candidate inspection point and synchronized to the path planning in step S4.

[0066] Furthermore, the physical location of the hot spot region indicates the working status of the photovoltaic equipment, and its temperature gradient further indicates the relative risk brought about by the temperature gradient; in this scenario, instead of regenerating candidate inspection points, we only rewrite the risk labels of the corresponding locations and record the relative risk of each location simultaneously.

[0067] S4, based on the risk information of candidate inspection points, iteratively sets the inspection path for the current round.

[0068] like Figure 5 As shown, one implementation of step S4 includes: S41, take the candidate inspection points as the base points and add them to the preset inspection path set of the current round; and traverse the spatial location of each candidate inspection point to identify the photovoltaic equipment that is spatially adjacent to the candidate inspection point.

[0069] In this step, based on the identified risk points, the adjacent photovoltaic equipment is automatically included in the inspection scope; these adjacent equipment are those that were not directly identified as having abnormal risks in the previous process, and are potential points of connection for anomalies.

[0070] S42, For photovoltaic equipment that is spatially adjacent to a candidate inspection point, if the equipment is not included in the preset inspection path set, then the equipment is expanded into a new candidate inspection point and added to the preset inspection path set.

[0071] The reason for selecting these devices for simultaneous inspection is that photovoltaic power plant faults have a natural spatial conductivity. For adjacent cells around hot spots, early latent heating may occur due to the spread of local shading or abnormal string current, even if it does not meet the threshold for hot spot identification. In addition, devices in the same string with abnormal power output will be affected by the faulty device, and will simultaneously experience problems such as reduced power generation performance or abnormal temperature gradients. It is necessary to check the adjacent photovoltaic devices together to improve the relative comprehensiveness of the current round of inspections.

[0072] It should be noted that the newly added candidate inspection points no longer recursively identify spatially adjacent devices to avoid unlimited expansion of the inspection range. These devices are only used as spatial expansion of the candidate inspection points.

[0073] S43, backtrack to all candidate inspection points in the preset inspection path set, and adjust the risk weight of all candidate inspection points according to the power anomaly and temperature anomaly marked by the risk information.

[0074] The main focus of the correction is the candidate inspection points introduced in step S41. These points have risk information labels and are the only processing targets for the current scenario correction.

[0075] During the correction, the risk information marked during temperature processing is first used as a reference. If the previous risk was marked as an abnormality in power, but the temperature is completely within the normal temperature range of the component, then the relevant risk is eliminated.

[0076] Secondly, the risk of hot spots is verified based on the risk information of power generation attenuation. If the power generation corresponding to the hot spot area does not show the performance degradation corresponding to the current hot spot change trend, the hot spot risk is determined to be a non-faulty pseudo hot spot, and the relevant risk is removed.

[0077] For the corrected candidate inspection points, the ratio of the change in the current time period's power generation efficiency fitting slope value relative to the reference power generation efficiency slope value, as described in step S2, will be used as the weight for performance degradation; the normalized value of the difference between the hot spot's highest temperature and the module's normal operating temperature will be used as the weight for temperature; finally, based on the hot spot's gradient vector being in a trend of continuous increase, relative stability, and continuous decrease, the normalized value of its gradient change will be used as the weight for trend evolution, and the sum of these three weights will be used as the risk weight for each candidate inspection point.

[0078] For candidate inspection points that have been removed from the list of risks, after the risk weight calculation is completed, the calculated risk weight is reduced proportionally to half of the original calculated value, thereby lowering the relative priority of the inspection at that location.

[0079] S44 performs spatial clustering based on the spatial location and risk weight of candidate inspection points, divides multiple inspection sub-regions, and iteratively generates the inspection path for each inspection sub-region as the output inspection path.

[0080] When generating inspection paths, K-means clustering is first performed based on the spatial location and risk weight of each point. This creates a cluster feature vector composed of the spatial coordinates of each point and its risk weight. Points that are spatially adjacent and have similar risk levels are automatically divided into the same inspection sub-region, with the number of sub-regions corresponding one-to-one with the number of drones. In this scenario, the drones will identify high-risk inspection sub-regions and set inspection paths according to the distribution of these sub-regions.

[0081] At the same time, from all points in the power plant, priority is given to selecting several points with the highest risk weight as initial cluster centers. The number of points selected is the same as the number of drones, ensuring that each drone corresponds to at least one high-risk candidate inspection point, while avoiding all high-risk points from clustering within the operating range of one drone, which could lead to load imbalance.

[0082] Subsequently, ant colony optimization iterations are performed independently in each sub-region to generate inspection paths corresponding to each drone, ultimately generating multiple parallel inspection paths without spatial overlap or flight conflicts.

[0083] Specifically, when setting up the ant colony algorithm, the spatial distance and risk weight between points are first normalized, and then the transition probability of selecting the next point from the current point is calculated. The calculation formula is as follows: ; in, This represents the transition probability of selecting the next candidate inspection point j when the inspected sub-region k is at candidate inspection point i: This represents the pheromone concentration from candidate inspection point i to candidate inspection point j. The pheromone concentration is affected by the average risk weight of all points and the average shortest spatial distance between the points. Here, the product of the number of candidate inspection points in the current sub-region and the average risk weight is used as the numerator, and the average shortest spatial distance is used as the denominator. The ratio of the numerator to the denominator is used as the initial pheromone concentration of the current scene, and it is continuously adjusted according to each iteration. A factor representing the importance of pheromones, used to control the weight of pheromones in path selection; The heuristic function value represents the heuristic value from candidate inspection point i to candidate inspection point j. This value is defined by the ratio of the risk weight of candidate inspection point j to the spatial distance between the two points. It is the prior expected return for each point movement. The closer the distance and the higher the risk of the point, the larger the heuristic value. This represents the importance factor of the heuristic function, used to control the weight of the heuristic function in path selection; s represents the set of unvisited candidate inspection points in the inspection sub-region k, where s is the index of the set; and and represent the pheromone concentration and heuristic function value from candidate inspection point i to unvisited candidate inspection point s, respectively. In this formula, the transition probability is calculated only if candidate inspection point j has not been visited; otherwise, the corresponding point is not selected.

[0084] The calculated probability serves as the basis for single-step decision-making during path optimization. The next point is randomly selected based on this probability value. After traversing all candidate inspection points and forming a complete access sequence, the total cost of the entire path is calculated.

[0085] The cost is set based on the difference between the sum of the paths to each point in the access sequence and the sum of the risk weights; alternatively, an adjustment coefficient can be set on the sum of the risk weights to balance the relative filtering of spatial distance and risk weights.

[0086] Next, evaporation is performed on the paths between all points to reduce the pheromone concentration of the old paths and prevent the algorithm from locking into local optima. This means proportionally reducing the pheromone concentration of all paths. Then, all candidate inspection points are traversed again, and the ratio of a fixed constant to the total path cost is used as the pheromone increment between every two candidate inspection points, thus completing the pheromone adjustment. The total path cost is iterated multiple times. When the total cost no longer changes after several consecutive iterations, or when a preset number of iterations is reached, the sorting order of the candidate inspection points is considered the output inspection path, thus implementing the path setting for the ant colony algorithm. The preset number of iterations can be set to 200 to prevent excessive computation in local areas, which could reduce the system's computing power.

[0087] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. A method for intelligent inspection and control of photovoltaic power plant operation and maintenance, characterized in that, include: S1, based on the inspection path of the drone in the previous round, identifies all photovoltaic devices within the range of the inspection path; Collect environmental data and output power of photovoltaic equipment, and determine the target area in each scenario; the environmental data includes the current irradiance, ambient temperature and average temperature of the module backsheet; S2, based on the output power of photovoltaic equipment in the target area, extrapolate the power generation degradation process of photovoltaic equipment and select candidate inspection points that characterize the risk of power generation degradation; S3, acquire the infrared thermal imaging spectrum of the area corresponding to the candidate inspection point, and automatically identify and extract the hot spot area in the spectrum; Based on the main axis direction of the hot spot region and the temperature gradient, update the risk information of candidate inspection points; S4, based on the risk information of candidate inspection points, iteratively sets the inspection path for the current round.

2. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 1, characterized in that, Step S1, which determines the implementation method of the target region in each scenario, includes: S11, based on the image range captured by the drone and the waypoints along the inspection path, set the drone's travel range; S12, for all photovoltaic devices within the route range, set an environmental baseline according to irradiance and ambient temperature, and solve for the theoretical backsheet temperature corresponding to the environmental baseline; S13, perform interval clustering of photovoltaic equipment according to environmental baseline to form candidate target areas; S14. Within the same candidate target area, the average backsheet temperature of each photovoltaic device is compared with the theoretical backsheet temperature using residuals. When the residual exceeds the limit, the corresponding candidate target area is taken as the output target area.

3. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 2, characterized in that, When performing interval clustering of photovoltaic equipment based on environmental baselines, the implementation methods include: Based on the quantiles of irradiance and ambient temperature in historical data, the range of values ​​for both is divided into multiple intervals. Each photovoltaic device is detected to fall into an interval, and photovoltaic devices that are spatially adjacent and fall into the same interval are considered as candidate target areas.

4. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 1, characterized in that, The implementation methods for candidate inspection points in step S2 include: S21, Calculate the power generation efficiency for each time period based on the output power at each moment; S22, based on the differences in output power at different times, performs a preliminary inspection of the photovoltaic equipment in the time dimension and quickly generates candidate inspection points; S23. For each candidate inspection point, a secondary inspection is performed according to the decay trend of the power generation efficiency of each device to determine the output candidate inspection point.

5. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 4, characterized in that, Preliminary time-based inspection of photovoltaic equipment includes: The output power of any target region at several times is obtained, and the output power is converted into a vector of corresponding values. The cluster centers of the output power are obtained through a clustering algorithm, and the values ​​corresponding to the cluster centers are used as reference power. Based on the irradiance values ​​at each time point, determine the number of valid sampling points within the current time period; Based on the relative difference between the output power and the reference power at each moment, the total number of current equipment anomalies is counted. Based on the proportion of abnormal points to the number of valid sampling points, it is determined whether the current device is abnormal, and the geographical coordinates of the abnormal devices are used to generate candidate inspection points.

6. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 4, characterized in that, A secondary inspection was conducted based on the decline trend of the power generation efficiency of each device, including: Based on the decline trend of power generation efficiency, determine whether the decline trend points to abnormal equipment performance. If it does, output the corresponding candidate inspection point directly. Otherwise, count the cumulative number of times the same location in the current target area is identified as a candidate inspection point, and output the candidate inspection points with a cumulative count greater than 1 directly.

7. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 1, characterized in that, The methods for updating the risk information of candidate inspection points in step S3 include: S31, perform binarization processing on the infrared thermal imaging spectrum, extract pixels with temperatures higher than the base temperature of the region, and combine them into hot spot regions; S32, perform minimum bounding rectangle fitting on the hot spot region to determine the principal axis direction of the hot spot region's extension; S33, compare the direction of the main shaft with the arrangement direction of the internal components of the photovoltaic equipment to determine the physical location of the hot spot region connection; S34, construct temperature profiles along the main axis direction and perpendicular to the main axis direction of the hot spot region respectively, and extract the temperature gradient features corresponding to each temperature profile; S35 updates the risk information of candidate inspection points based on the extracted temperature gradient features and the physical location of the hot spot regions.

8. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 7, characterized in that, When fitting the minimum bounding rectangle to the hot spot region, the following methods can be used: For all pixels within the hot spot region, the difference between the temperature of each pixel and the lowest temperature in the hot spot region is used as the weight of that pixel. The coordinates of all pixels are weighted and averaged to obtain the weighted average coordinates. These weighted average coordinates are then used as the weighted centroid of the hot spot region. Calculate the weighted covariance matrix of all pixels relative to the weighted centroid; perform eigenvalue decomposition on the weighted covariance matrix, and use the largest eigenvector obtained from the decomposition as the principal axis direction for the extension of the hot spot region; Using the weighted centroid of the hot spot region as the origin, the principal axis of the hot spot region as the horizontal axis, and the direction perpendicular to the principal axis as the vertical axis, the minimum bounding rectangle for fitting is determined.

9. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 7, characterized in that, When extracting the temperature gradient features corresponding to each temperature profile, the implementation methods include: Based on the infrared thermal imaging spectrum over a continuous time period, the temperature gradient magnitude is calculated pixel by pixel at each temperature profile. The gradient magnitudes of all pixels on the profile are arranged in order to obtain the temperature gradient sequence at the current moment. Obtain the maximum gradient magnitude, average gradient magnitude, and gradient magnitude interval percentage corresponding to the temperature gradient sequence at the current moment, and generate the gradient vector at each moment. Using the gradient vector at each moment as a guide, the difference between the gradient vector at the initial moment of the inspection is calculated to obtain the gradient change at each moment, and the trend of the gradient change is used as the temperature gradient feature at the current moment.

10. The intelligent inspection and control method for operation and maintenance of a photovoltaic power station according to claim 1, characterized in that, The implementation methods of the inspection path in step S4 include: S41, take the candidate inspection points as the basic points and add them to the preset inspection path set of the current round; and traverse the spatial position of each candidate inspection point to identify the photovoltaic equipment that is spatially adjacent to the candidate inspection point. S42, For photovoltaic equipment that is spatially adjacent to a candidate inspection point, if the equipment is not included in the preset inspection path set, then the equipment is expanded into a new candidate inspection point and added to the preset inspection path set; S43, backtrack all candidate inspection points in the preset inspection path set, and adjust the risk weight of all candidate inspection points according to the power anomaly and temperature anomaly marked by the risk information. S44 performs spatial clustering based on the spatial location and risk weight of candidate inspection points, divides multiple inspection sub-regions, and iteratively generates the inspection path for each inspection sub-region as the output inspection path.

Citation Information

Patent Citations

  • Unmanned aerial vehicle intelligent photovoltaic power station inspection method and system

    CN118012100A

  • Unmanned aerial vehicle inspection method and system for photovoltaic station, medium and program product

    CN120560305A