Operation and maintenance management methods for new energy power plants based on multi-source data

Abstract

This invention relates to a multi-source data-based operation and maintenance management method for new energy power plants, and to the field of operation and maintenance management technology. It includes acquiring electrical operating parameters of photovoltaic equipment. The invention constructs a two-dimensional contradiction deviation vector by comparing theoretical temperature inversion with infrared measured temperature and statistically analyzing the proportion of local hot spot areas. It then uses a false fault contradiction fingerprint database for matching. By matching the pre-constructed contradiction fingerprint database with a nearest neighbor classifier, four types of standard fingerprint databases are constructed to identify false faults. Each fingerprint type is jointly calibrated by a first contradiction deviation interval and a second contradiction deviation interval, improving the accuracy of fault identification. By tracking the temporal drift trajectory of the equipment on the contradiction fingerprint plane, the drift velocity and drift acceleration are calculated. The time required to reach the boundary of the target fingerprint region from the current state is predicted and calculated. Based on the false fault type and the critical evolution state, differentiated handling instructions are generated.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance management technology, and in particular to an operation and maintenance management method for new energy power plants with multi-source data. Background Technology

[0002] As a crucial component of the energy sector, the safe and stable operation of new energy power plants is of paramount importance to energy supply and grid security. With the rapid growth of installed new energy capacity, power plant scale is constantly expanding, and the number of devices is increasing dramatically. Traditional manual inspections and periodic maintenance are no longer sufficient to meet the operation and maintenance needs of large-scale new energy power plants. In recent years, with the rapid development of sensor technology, image acquisition technology, and communication technology, new energy power plants have widely deployed multi-source monitoring devices such as electrical quantity sensors, meteorological sensors, infrared thermal imaging equipment, and visible light cameras, accumulating massive amounts of operational data. Among existing new energy power plant operation and maintenance technologies, the mainstream fault diagnosis methods are mainly divided into two categories: one is the threshold judgment method based on electrical parameters, which triggers alarms by monitoring abnormal changes in electrical quantities such as voltage, current, and power; the other is the single-modal analysis method based on infrared thermal imaging or visible light images, which judges the equipment status by identifying abnormal temperatures or apparent defects on the equipment surface. However, the above-mentioned existing technologies still have the following significant shortcomings in practical applications:

[0003] In the existing technology, threshold judgment methods based on electrical parameters can only detect faults that have caused significant changes in electrical quantities. For early faults and progressive faults, the changes in electrical parameters are weak and easily masked by environmental fluctuations, leading to missed detections. Electrical parameter monitoring and image analysis are usually completed independently by different subsystems, and there is a lack of effective correlation and fusion between the two, resulting in reduced judgment efficiency.

[0004] Existing technologies treat all anomalies that deviate from the normal pattern as faults, lacking the ability to identify false faults such as sensor drift, dust accumulation and occlusion, shadow movement, and hot spot emergence. They also lack the ability to track and predict the evolution of false faults into real faults, and cannot provide early warnings before a fault occurs.

[0005] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0006] The purpose of this invention is to construct a two-dimensional contradiction deviation vector, compare theoretical temperature with infrared measured values, and compare electrical parameters with visible light appearance to form a false fault description. By using four types of standard fingerprint databases and a nearest neighbor classifier, false faults caused by dust accumulation can be identified in real time. Furthermore, by tracking the temporal drift trajectory on the contradiction fingerprint plane and calculating the drift velocity and acceleration, the invention aims to achieve critical early warning of the evolution of false faults into real faults and accurate allocation of operation and maintenance resources.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for operation and maintenance management of new energy power plants based on multi-source data, comprising the following steps:

[0008] Step 1: Obtain the electrical operating parameters of the photovoltaic equipment, collect infrared thermal images and visible light images at the same time and spatial location, and link the electrical operating parameters, infrared thermal images and visible light images with a unified timestamp as the association key to form a three-element database;

[0009] Step 2: Based on the three-element database, analyze the deviation between electrical measurement values ​​and infrared temperature distribution in spatial distribution to obtain the first degree of discrepancy between electrical operating parameters and infrared thermal imaging; analyze the deviation between electrical measurement values ​​and visible light apparent state in characteristic performance to obtain the second degree of discrepancy between electrical operating parameters and visible light image.

[0010] Step 3: Input the two-dimensional deviation vector formed by the first contradiction deviation degree and the second contradiction deviation degree into the pre-trained false fault classifier, and match it with the pre-constructed false fault contradiction fingerprint database. The classifier outputs the false fault type.

[0011] Step 4: For the two-dimensional deviation vector of the equipment at continuous time points, construct the temporal drift trajectory on the contradictory fingerprint plane, calculate the instantaneous drift velocity and drift acceleration at the current time point, generate the fault evolution critical state feature set, and judge the critical state evolution of false faults and real faults to output evolution warning signals and expected critical time windows.

[0012] Step 5: Generate graded handling instructions based on the identified false fault types and the determination results of the evolutionary critical state.

[0013] Furthermore, electrical operating parameters, infrared thermal images, and visible light images are linked by a unified timestamp to form a three-element database. The specific process is as follows:

[0014] The DC voltage, DC current, DC power, and string series resistance of photovoltaic modules are collected at a fixed sampling frequency. Each data point is appended with a millisecond-level timestamp and device identifier to form an electrical dataset.

[0015] The inspection and shooting were carried out by a drone equipped with an infrared camera, and the shooting timestamp, spatial coordinates and attitude angle were recorded. The image pixel value corresponds to the actual measured temperature.

[0016] A visible light camera, coaxially mounted with an infrared camera, is used to synchronously trigger acquisition, recording the same timestamp and spatial parameters. Pixel-level registration with the infrared image is achieved through affine transformation.

[0017] Using timestamps as the primary reference, the nearest neighbor matching method is employed to link electrical data and image data. Device ID, electrical parameters, infrared image path, visible light image path, and spatial coordinates are integrated into structured records and stored in a three-element database.

[0018] Furthermore, the spatial distribution deviation between electrical measurements and infrared temperature distribution is analyzed to obtain the first degree of discrepancy between electrical operating parameters and infrared thermal imaging. The specific process is as follows:

[0019] The electrical operating parameters of the equipment at the same point in time are extracted from the three-element database, and combined with the ambient temperature, the theoretical operating temperature of the equipment under the current working conditions is calculated.

[0020] Based on the infrared thermal image, the measured temperature of each pixel is extracted, and a measured temperature distribution matrix of the device surface is constructed. The pixel area corresponding to the photovoltaic module is located in the infrared image. The module outline is extracted by the image segmentation algorithm, the temperature value of all pixels in the module area is obtained, and the average temperature, maximum temperature and abnormal high temperature pixel set of the module area above the theoretical temperature are calculated.

[0021] Using the theoretical temperature as a baseline threshold, the proportion of abnormally high-temperature pixels is statistically analyzed in the infrared temperature distribution matrix, and the global deviation between the component's average temperature and the theoretical temperature is calculated.

[0022] By weighting and fusing the global temperature deviation with the proportion of local hot spot area, the consistency deviation between electrical measurements and infrared temperature distribution is analyzed to obtain the first degree of inconsistency.

[0023] Furthermore, the deviation between electrical measurements and visible light apparent states in terms of characteristic performance is analyzed to obtain the second discrepancy between electrical operating parameters and visible light images. The specific steps are as follows;

[0024] Visible light images at the same time are read from the three-element database. The images are classified pixel by pixel using a deep learning-based image segmentation network. The proportions of the area covered by dust and the area covered by shadow to the total area of ​​the component are calculated to obtain the appearance anomaly index.

[0025] Extract the measured electrical power, calculate the theoretical output power based on the ambient irradiance and component temperature, and obtain the relative deviation between the measured power and the theoretical power;

[0026] The second contradiction deviation degree is obtained by weighting and fusing the power deviation with the apparent anomaly index.

[0027] Furthermore, the two-dimensional deviation vector formed by the first and second contradiction deviations is input into a pre-trained pseudo-fault classifier and matched with a pre-built pseudo-fault contradiction fingerprint database. The classifier outputs the pseudo-fault type. The specific process is as follows:

[0028] Collect electrical-image dual-modal data samples of known fault types, calculate the first and second inconsistency deviations of each sample group, plot the sample distribution scatter plot of each type of fault on a two-dimensional plane, determine the standard fingerprint region for each type of false fault, and obtain the first and second inconsistency deviation intervals.

[0029] For the data to be identified, calculate its first contradiction deviation degree and second contradiction deviation degree to form a two-dimensional deviation vector, which is used as the input of the false fault classifier;

[0030] The nearest neighbor classifier is used for matching: the two-dimensional deviation vector is judged to be in the standard interval of the fingerprint type, and the false fault type corresponding to the fingerprint is directly output. If it is in multiple fingerprint intervals, the Euclidean distance from the two-dimensional deviation vector to the center vector of each overlapping fingerprint is calculated, and the fingerprint type with the smallest distance is selected as the output. If it does not fall into any fingerprint interval, it is marked as unrecognized.

[0031] The classifier outputs false fault types, only outputting type labels: dust accumulation occlusion, shadow movement, sensor drift, and hot spot budding.

[0032] Furthermore, for the two-dimensional deviation vector of the device at continuous time points, a temporal drift trajectory on the contradictory fingerprint plane is constructed, and the instantaneous drift velocity and drift acceleration at the current time point are calculated to generate a fault evolution critical state feature set. The specific steps are as follows:

[0033] Extract the two-dimensional deviation vector sequence of the same device at continuous time points from the three-element database, use the first contradiction deviation degree as the abscissa and the second contradiction deviation degree as the ordinate, draw scattered points at each time point on the two-dimensional plane, and connect adjacent scattered points in chronological order to form a time-series drift trajectory line.

[0034] Instantaneous drift velocity is calculated based on two adjacent time points on the time-series drift trajectory, and drift acceleration is calculated based on the drift velocity at two adjacent time points.

[0035] The distance from the two-dimensional deviation vector at the current time point to the boundary of each fingerprint region is taken as the minimum value of the fingerprint region distance feature;

[0036] From the time-series drift trajectory line, instantaneous drift velocity, and drift acceleration, we extract trajectory direction features, drift velocity features, drift acceleration, and fingerprint region distance features to form a fault evolution critical state feature set.

[0037] Furthermore, the critical state of the evolution from false faults to real faults is determined to output an evolution warning signal and the expected critical time window. The specific steps are as follows:

[0038] Obtain the current device's trajectory direction, instantaneous drift speed, drift acceleration, and distance to the fingerprint area boundary;

[0039] The evolutionary critical state is jointly determined by multiple conditions. When the drift trajectory points from the ash-covered area to the hot spot budding area, and the drift acceleration a>0, and the distance from the current point to the boundary of the hot spot area is less than a preset threshold, it is the pre-hot spot formation stage.

[0040] The current point is located in the hot spot budding area, the drift direction is pointing in the direction of increasing first contradiction deviation, the drift speed exceeds the threshold, and the infrared hot spot area increases three times in a row, which means that the power device is about to fail.

[0041] If the drift trajectory points from the shadow movement area to the hot spot budding area, and the acceleration changes from negative to positive, then it is the pre-stage of shadow-induced local overheating.

[0042] The current point is located within the dust accumulation and obscuring area, drifting towards the hot spot germination area, and its average velocity over past time points exceeds twice the historical average velocity. This indicates that it is in the pre-critical stage of accelerated dust accumulation.

[0043] Starting from the two-dimensional deviation vector of the current point, calculate the time to reach the boundary of the target fingerprint region, multiply it by the security factor, and output the estimated time window.

[0044] Furthermore, based on the identified false fault types and the determination results of the evolutionary critical state, a graded handling instruction is generated. The specific steps are as follows:

[0045] Based on the types of false failures and the critical state of evolution, a pre-set decision tree rule base is established, with planned handling corresponding to stable states and emergency handling corresponding to critical states.

[0046] Establish a priority order for handling hotspots: hotspot germination > sensor drift > shadow movement > dust accumulation and occlusion, for resource allocation decisions when multiple tasks are concurrent;

[0047] The command pushes the operation and maintenance scheduling system to create work orders. After execution, the system provides feedback on the results and updates the device status. The feedback data is also used for continuous optimization of the rule base.

[0048] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0049] This multi-source data-driven operation and maintenance management method for new energy power plants constructs a first discrepancy deviation degree by comparing theoretical temperature inversion with infrared measured temperature and statistically analyzing the proportion of local hot spot areas. Based on a second discrepancy deviation degree between electrical parameters and visible light appearance, a two-dimensional discrepancy deviation vector is constructed, forming a three-dimensional characterization of pseudo-faults. Using a pseudo-fault discrepancy fingerprint database for matching, it can immediately distinguish between pseudo-faults caused by dust accumulation, shadow movement, sensor drift, and hot spot budding, and real equipment faults after data acquisition. By matching the pre-constructed discrepancy fingerprint database with a nearest neighbor classifier, it can immediately output the pseudo-fault type after data acquisition, achieving accurate differentiation between pseudo-faults and real faults. Four types of standard fingerprint databases are constructed to identify false faults. Each fingerprint is jointly calibrated by the first and second contradiction deviation intervals, which improves the accuracy of fault identification. By tracking the temporal drift trajectory of the device on the contradiction fingerprint plane and calculating the drift velocity and drift acceleration, early warnings can be issued in advance at the critical state of the evolution from false faults to real faults. By predicting and calculating the time required to reach the boundary of the target fingerprint region from the current state, the estimated critical time window is output to achieve precise allocation of operation and maintenance resources. Differentiated handling instructions are generated according to the type of false fault and the evolution critical state. Attached Figure Description

[0050] Figure 1 A schematic diagram of the overall steps of the method of the present invention is shown;

[0051] Figure 2 This diagram illustrates the overall structure of the data transmission process of the present invention.

[0052] Figure 3 This diagram illustrates the structural steps of the contradiction deviation and fault classification method of the present invention.

[0053] Figure 4 A schematic diagram of the fault judgment and early warning output structure of the present invention is shown. Detailed Implementation

[0054] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0055] Example 1:

[0056] like Figure 1-4 As shown, the operation and maintenance management method for new energy power plants with multi-source data includes the following steps:

[0057] Step 1: Obtain the electrical operating parameters of the photovoltaic equipment, collect infrared thermal images and visible light images at the same time and spatial location, and link the electrical operating parameters, infrared thermal images and visible light images with a unified timestamp as the association key to form a three-element database;

[0058] Step 2: Based on the three-element database, analyze the deviation between electrical measurement values ​​and infrared temperature distribution in spatial distribution to obtain the first degree of discrepancy between electrical operating parameters and infrared thermal imaging; analyze the deviation between electrical measurement values ​​and visible light apparent state in characteristic performance to obtain the second degree of discrepancy between electrical operating parameters and visible light image.

[0059] Step 3: Input the two-dimensional deviation vector formed by the first contradiction deviation degree and the second contradiction deviation degree into the pre-trained false fault classifier, and match it with the pre-constructed false fault contradiction fingerprint database. The classifier outputs the false fault type.

[0060] Step 4: For the two-dimensional deviation vector of the equipment at continuous time points, construct the temporal drift trajectory on the contradictory fingerprint plane, calculate the instantaneous drift velocity and drift acceleration at the current time point, generate the fault evolution critical state feature set, and judge the critical state evolution of false faults and real faults to output evolution warning signals and expected critical time windows.

[0061] Step 5: Generate graded handling instructions based on the identified false fault types and the determination results of the evolutionary critical state.

[0062] In this scheme, when collecting electrical operating parameters of photovoltaic equipment, the DC voltage, DC current, and DC power of the photovoltaic modules are collected according to a fixed sampling frequency. Including series and group resistors, each data entry is appended with a millisecond-level timestamp and device identifier to form an electrical dataset;

[0063] The inspection and shooting were carried out by a drone equipped with an infrared camera, and the shooting timestamp, spatial coordinates and attitude angle were recorded. The image pixel value corresponds to the actual measured temperature.

[0064] A visible light camera, coaxially mounted with an infrared camera, is used to synchronously trigger acquisition, recording the same timestamp and spatial parameters. Pixel-level registration with the infrared image is achieved through affine transformation.

[0065] Using timestamps as the primary reference, the nearest neighbor matching method is employed to link electrical data and image data. Device ID, electrical parameters, infrared image path, visible light image path, and spatial coordinates are integrated into structured records and stored in a three-element database.

[0066] By analyzing the spatial distribution deviation between electrical measurements and infrared temperature distribution, the first discrepancy between electrical operating parameters and infrared thermal imaging is obtained. The specific process is as follows:

[0067] Extract electrical operating parameters of the equipment at the same point in time from the three-element database, and combine them with ambient temperature. The theoretical temperature of the calculation device under current operating conditions. ;

[0068] Theoretical temperature The calculation formula is:

[0069] in, U represents the theoretical photoelectric conversion efficiency, and U is the set basic comprehensive heat dissipation coefficient, which is obtained by performing multiple linear regression fitting on electrical power, ambient temperature and infrared measured temperature in historical normal operation data. A is the component area, and the theoretical temperature represents the expected temperature level of the equipment under normal heat dissipation conditions, serving as a benchmark for comparison with the infrared measured temperature.

[0070] Based on the infrared thermal image, the measured temperature of each pixel is extracted. A measured temperature distribution matrix of the device surface is constructed, the pixel area corresponding to the photovoltaic module is located in the infrared image, the module outline is extracted by the image segmentation algorithm, the temperature value of all pixels in the module area is obtained, and the average temperature and maximum temperature of the module area are calculated. And a collection of abnormally high-temperature pixels with temperatures exceeding theoretical values;

[0071] Using theoretical temperature as a baseline threshold, the proportion of pixels with abnormally high temperatures is statistically analyzed in the infrared temperature distribution matrix, and the markers in the matrix satisfy the following conditions: , With an allowable deviation of 1, count the number of these abnormally high-temperature pixels. Calculate the proportion of the total number of component pixels. Calculate the average temperature of the components Compared with theoretical temperature global deviation ;

[0072] Calculate the global deviation between the component's average temperature and the theoretical temperature:

[0073]

[0074] By weighting and fusing the global temperature deviation with the proportion of local hot spot area, the consistency deviation between electrical measurements and infrared temperature distribution is analyzed to obtain the first degree of discrepancy. :

[0075]

[0076] In the above formula, The set temperature instrument reference value, The larger the value, the greater the inconsistency between the electrical measurement value and the infrared temperature distribution.

[0077] The deviation between electrical measurements and visible light appearance is analyzed to obtain the second discrepancy between electrical operating parameters and visible light images. The specific steps are as follows:

[0078] Visible light images from the same time point were retrieved from a three-element database. A deep learning-based image segmentation network was used to classify the images pixel by pixel and to calculate the area of ​​dust accumulation. and shadow occlusion area Total area of ​​components The proportion is used to obtain the apparent abnormality index. ;

[0079]

[0080] Extract the measured electrical power, calculate the theoretical output power based on the ambient irradiance and component temperature, and obtain the measured power and theoretical power. The relative deviation is weighted and fused with the power deviation and the apparent anomaly index to obtain the second inconsistency deviation degree. The calculation process is as follows:

[0081]

[0082] Calculate the measured power Relative deviation from theoretical power :

[0083]

[0084]

[0085] In the above formula, the irradiance G is obtained from the weather station. .

[0086] The two-dimensional deviation vector formed by the first and second contradiction deviations is input into a pre-trained pseudo-fault classifier and matched with a pre-built pseudo-fault contradiction fingerprint database. The classifier outputs the pseudo-fault type. The specific process is as follows:

[0087] Collect electrical-image dual-modal data samples of known fault types, calculate the first and second inconsistency deviations of each sample group, plot the sample distribution scatter plot of each type of fault on a two-dimensional plane, determine the standard fingerprint region for each type of false fault, and obtain the first and second inconsistency deviation intervals.

[0088] Collect confirmed false fault samples from historical operation and maintenance data, with at least 500 samples per category. The fingerprinting rules are as follows:

[0089] Dust accumulation and obstruction: Comparative data from before and after cleaning. scope , scope 1.0);

[0090] Shadow movement: caused by object occlusion. scope , scope ,and and The time series exhibits periodic fluctuations;

[0091] Sensor drift: Case study of replacing a faulty sensor scope , scope ;

[0092] Early hot spot development: Case studies of early hot spot detection using infrared detection. scope , scope ;

[0093] For the current data to be identified, calculate the first contradiction deviation degree and the second contradiction deviation degree to form a two-dimensional deviation vector, which is used as the input of the false fault classifier;

[0094] The nearest neighbor classifier is used for matching: the two-dimensional deviation vector is judged to be in the standard interval of the fingerprint type, and the false fault type corresponding to the fingerprint is directly output. If it is in multiple fingerprint intervals, the Euclidean distance from the two-dimensional deviation vector to the center vector of each overlapping fingerprint is calculated, and the fingerprint type with the smallest distance is selected as the output. If it does not fall into any fingerprint interval, it is marked as unrecognized.

[0095] The classifier outputs false fault types, only outputting type labels: dust accumulation occlusion, shadow movement, sensor drift, hot spot budding;

[0096] The boundaries of each fingerprint class are determined by adding or subtracting twice the standard deviation from the sample mean, and are stored in the fingerprint database table.

[0097] For the two-dimensional deviation vector of the device at continuous time points, construct the temporal drift trajectory on the contradictory fingerprint plane, calculate the instantaneous drift velocity and drift acceleration at the current time point, and generate the fault evolution critical state feature set. The specific steps are as follows:

[0098] Will and Combined into a two-dimensional deviation vector Extract the two-dimensional deviation vector sequence of the same device at consecutive time points from the three-element database. i = {1, 2, ..., N}, with the first contradiction deviation as the x-axis and the second contradiction deviation as the y-axis, scatter points at each time point are plotted on a two-dimensional plane, and adjacent scatter points are connected in chronological order to form a time-series drift trajectory line;

[0099] Instantaneous drift velocity is calculated based on two adjacent time points on the time-series drift trajectory, and drift acceleration is calculated based on the drift velocity at two adjacent time points.

[0100] The distance from the two-dimensional deviation vector at the current time point to the boundary of each fingerprint region is taken as the minimum value of the fingerprint region distance feature;

[0101] For two adjacent points on the trajectory and Calculate the Euclidean distance between two points as a feature of the region distance. : ;

[0102] Time interval : Then the drift velocity at the i-th time point : ;

[0103] Calculate the drift acceleration from the drift velocity at two adjacent time points. : ;

[0104] From the time-series drift trajectory line, instantaneous drift velocity, and drift acceleration, we extract trajectory direction features, drift velocity features, drift acceleration, and fingerprint region distance features to form a fault evolution critical state feature set;

[0105] Trajectory direction feature: The angle between the current trajectory direction and the boundary of each fingerprint region, used to determine whether the drift is pointing to an adjacent fingerprint region;

[0106] Drift speed characteristics: current drift speed, and average speed at past time points;

[0107] Drift acceleration characteristics: current drift acceleration, and zero-crossing detection when acceleration changes from negative to positive or from positive to negative;

[0108] Fingerprint region distance feature: the shortest distance from the current time point to the boundary of each fingerprint region.

[0109] The critical state of the evolution from false faults to real faults is determined, and an evolution warning signal and the expected critical time window are output. The specific steps are as follows:

[0110] Obtain the current device's trajectory direction, instantaneous drift speed, drift acceleration, and distance to the fingerprint area boundary;

[0111] The evolutionary critical state is jointly determined by multiple conditions. When the drift trajectory points from the ash-covered area to the hot spot budding area, and the drift acceleration a>0, and the distance from the current point to the boundary of the hot spot area is less than a preset threshold, it is the pre-hot spot formation stage.

[0112] The current point is located in the hot spot budding area, the drift direction is pointing in the direction of increasing first contradiction deviation, the drift speed exceeds the threshold, and the infrared hot spot area increases three times in a row, which means that the power device is about to fail.

[0113] If the drift trajectory points from the shadow movement area to the hot spot budding area, and the acceleration changes from negative to positive, then it is the pre-stage of shadow-induced local overheating.

[0114] If the current point is located within the dust accumulation and obscuring area, and the drift direction is pointing towards the hot spot germination area, and the average speed of past time points is more than twice the historical average speed, then it is in the critical stage before the accelerated accumulation of dust. The average speed of past time points is selected from 5 time points, and the number of time points can be set automatically.

[0115] Starting from the two-dimensional deviation vector of the current point, calculate the time to reach the boundary of the target fingerprint region, multiply it by the security factor, and output the estimated time window.

[0116] For the critical state, calculate the estimated time to reach the boundary of the target fingerprint region. :

[0117]

[0118] in, Let be the Euclidean distance from the current point to the boundary of the target fingerprint region, and take a security factor κ = 0.8. Current instantaneous velocity, For the current instantaneous acceleration, Given a time step, output the expected time window. :

[0119]

[0120] like If the value is less than 0.005, the output cannot be estimated. It is recommended to strengthen monitoring and encapsulate the judgment result into a structured early warning information, including equipment identification, current false fault type, expected critical time window, and recommended handling actions, including emergency cleaning, hot spot confirmation inspection, and continuous monitoring. The early warning information should be pushed to the operation and maintenance scheduling system to trigger the corresponding level of handling instructions.

[0121] Based on the identified false fault types and the determination results of the evolutionary critical state, a graded handling instruction is generated. The specific steps are as follows:

[0122] Based on the types of false failures and the critical state of evolution, a pre-set decision tree rule base is established, with planned handling corresponding to stable states and emergency handling corresponding to critical states.

[0123] Establish a priority order for handling hotspots: hotspot germination > sensor drift > shadow movement > dust accumulation and occlusion, for resource allocation decisions when multiple tasks are concurrent;

[0124] The command push operation and maintenance scheduling system creates work orders, and after execution, it feeds back the results to update the device status. At the same time, the feedback data is used for continuous optimization of the rule base.

[0125] In this scheme, the DC voltage, current, power, and series resistance of the photovoltaic strings are continuously collected. Simultaneously, a drone conducts weekly scheduled inspections, acquiring infrared thermal images and visible light images. For each set of correlated data, the first discrepancy deviation is calculated. Second contradiction deviation degree By comparing the theoretical temperature derived from electrical parameters with the measured infrared temperature, the degree of inconsistency between the electrical measurement values ​​and the infrared temperature distribution in spatial distribution is quantified. By comparing measured power with theoretical power and combining this with the proportion of dust or shadow area identified in visible light images, the degree of inconsistency between electrical measurements and visible light appearance is quantified.

[0126] Two-dimensional deviation vector Input a false fault classifier and match it against a pre-built fingerprint database containing four standard fingerprint categories: dust occlusion, shadow movement, sensor drift, and hot spot budding. Each category is represented by... and The interval joint calibration allows the classifier to directly output the false fault type based on the fingerprint interval it falls into, without going through the two-stage process of first judging the authenticity and then tracing the source.

[0127] For a two-dimensional deviation vector at consecutive time points, the system constructs a temporal drift trajectory on the contradictory fingerprint plane, calculates the instantaneous drift velocity and acceleration, and extracts features such as trajectory direction and distance to the fingerprint region boundary. When the drift trajectory migrates from one type of fingerprint region to another, the drift velocity exceeds the threshold, or the drift acceleration changes from negative to positive, the system determines that the device is in a critical state of evolution from a false fault to a real fault, and uses linear extrapolation to estimate the expected critical time window.

[0128] Finally, the system generates tiered handling instructions based on the type of pseudo-fault and the evolving critical state: planned maintenance instructions are generated in a stable state, emergency handling instructions are generated in a critical state, multiple tasks are sorted by priority when they occur concurrently, and tasks in the same geographical area are automatically merged into batch handling instructions, thereby realizing intelligent scheduling of operation and maintenance resources.

[0129] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0130] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0131] In the embodiments provided in this application, it should be understood that the disclosed methods can be implemented in other ways; for example, the method embodiments described above are merely illustrative. For example, the division of modules is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection between the shown or discussed components may be indirect coupling or communication connection through some interfaces, devices or modules, and may be electrical, mechanical or other forms.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for operation and maintenance management of new energy power plants based on multi-source data, characterized in that, Includes the following steps: Step 1: Obtain the electrical operating parameters of the photovoltaic equipment, collect infrared thermal images and visible light images at the same time and spatial location, and link the electrical operating parameters, infrared thermal images and visible light images with a unified timestamp as the association key to form a three-element database; Step 2: Based on the three-element database, analyze the deviation between electrical measurement values ​​and infrared temperature distribution in spatial distribution to obtain the first degree of discrepancy between electrical operating parameters and infrared thermal imaging; analyze the deviation between electrical measurement values ​​and visible light apparent state in characteristic performance to obtain the second degree of discrepancy between electrical operating parameters and visible light image. Step 3: Input the two-dimensional deviation vector formed by the first contradiction deviation degree and the second contradiction deviation degree into the pre-trained false fault classifier, and match it with the pre-constructed false fault contradiction fingerprint database. The classifier outputs the false fault type. Step 4: For the two-dimensional deviation vector of the equipment at continuous time points, construct the temporal drift trajectory on the contradictory fingerprint plane, calculate the instantaneous drift velocity and drift acceleration at the current time point, generate the fault evolution critical state feature set, and judge the critical state evolution of false faults and real faults to output evolution warning signals and expected critical time windows. Step 5: Generate graded handling instructions based on the identified false fault types and the determination results of the evolutionary critical state.

2. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, Electrical operating parameters, infrared thermal images, and visible light images are linked together using a unified timestamp as the association key to form a three-element database. The specific process is as follows: The DC voltage, DC current, DC power, and string series resistance of photovoltaic modules are collected at a fixed sampling frequency. Each data point is appended with a millisecond-level timestamp and device identifier to form an electrical dataset. The inspection and shooting were carried out by a drone equipped with an infrared camera, and the shooting timestamp, spatial coordinates and attitude angle were recorded. The image pixel value corresponds to the actual measured temperature. A visible light camera, coaxially mounted with an infrared camera, is used to synchronously trigger acquisition, recording the same timestamp and spatial parameters. Pixel-level registration with the infrared image is achieved through affine transformation. Using timestamps as the primary reference, the nearest neighbor matching method is employed to link electrical data and image data. Device ID, electrical parameters, infrared image path, visible light image path, and spatial coordinates are integrated into structured records and stored in a three-element database.

3. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, By analyzing the spatial distribution deviation between electrical measurements and infrared temperature distribution, the first discrepancy between electrical operating parameters and infrared thermal imaging is obtained. The specific process is as follows: The electrical operating parameters of the equipment at the same point in time are extracted from the three-element database, and combined with the ambient temperature, the theoretical operating temperature of the equipment under the current working conditions is calculated. Based on the infrared thermal image, the measured temperature of each pixel is extracted, and a measured temperature distribution matrix of the device surface is constructed. The pixel area corresponding to the photovoltaic module is located in the infrared image. The module outline is extracted by the image segmentation algorithm, the temperature value of all pixels in the module area is obtained, and the average temperature, maximum temperature and abnormal high temperature pixel set of the module area above the theoretical temperature are calculated. Using the theoretical temperature as a baseline threshold, the proportion of abnormally high-temperature pixels is statistically analyzed in the infrared temperature distribution matrix, and the global deviation between the component's average temperature and the theoretical temperature is calculated. By weighting and fusing the global temperature deviation with the proportion of local hot spot area, the consistency deviation between electrical measurements and infrared temperature distribution is analyzed to obtain the first degree of inconsistency.

4. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, The deviation between electrical measurements and visible light appearance is analyzed to obtain the second discrepancy between electrical operating parameters and visible light images. The specific steps are as follows: Visible light images at the same time are read from the three-element database. The images are classified pixel by pixel using a deep learning-based image segmentation network. The proportions of the area covered by dust and the area covered by shadow to the total area of ​​the component are calculated to obtain the appearance anomaly index. Extract the measured electrical power, calculate the theoretical output power based on the ambient irradiance and component temperature, and obtain the relative deviation between the measured power and the theoretical power; The second contradiction deviation degree is obtained by weighting and fusing the power deviation with the apparent anomaly index.

5. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, The two-dimensional deviation vector formed by the first and second contradiction deviations is input into a pre-trained pseudo-fault classifier and matched with a pre-built pseudo-fault contradiction fingerprint database. The classifier outputs the pseudo-fault type. The specific process is as follows: Collect electrical-image dual-modal data samples of known fault types, calculate the first and second inconsistency deviations of each sample group, plot the sample distribution scatter plot of each type of fault on a two-dimensional plane, determine the standard fingerprint region for each type of false fault, and obtain the first and second inconsistency deviation intervals. For the data to be identified, calculate its first contradiction deviation degree and second contradiction deviation degree to form a two-dimensional deviation vector, which is used as the input of the false fault classifier; The nearest neighbor classifier is used for matching: the two-dimensional deviation vector is judged to be in the standard interval of the fingerprint type, and the false fault type corresponding to the fingerprint is directly output. If it is in multiple fingerprint intervals, the Euclidean distance from the two-dimensional deviation vector to the center vector of each overlapping fingerprint is calculated, and the fingerprint type with the smallest distance is selected as the output. If it does not fall into any fingerprint interval, it is marked as unrecognized. The classifier outputs false fault types, only outputting type labels: dust accumulation occlusion, shadow movement, sensor drift, and hot spot budding.

6. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, For the two-dimensional deviation vector of the device at continuous time points, construct the temporal drift trajectory on the contradictory fingerprint plane, calculate the instantaneous drift velocity and drift acceleration at the current time point, and generate the fault evolution critical state feature set. The specific steps are as follows: Extract the two-dimensional deviation vector sequence of the same device at continuous time points from the three-element database, use the first contradiction deviation degree as the abscissa and the second contradiction deviation degree as the ordinate, draw scattered points at each time point on the two-dimensional plane, and connect adjacent scattered points in chronological order to form a time-series drift trajectory line. Instantaneous drift velocity is calculated based on two adjacent time points on the time-series drift trajectory, and drift acceleration is calculated based on the drift velocity at two adjacent time points. The distance from the two-dimensional deviation vector at the current time point to the boundary of each fingerprint region is taken as the minimum value of the fingerprint region distance feature; From the time-series drift trajectory line, instantaneous drift velocity, and drift acceleration, we extract trajectory direction features, drift velocity features, drift acceleration, and fingerprint region distance features to form a fault evolution critical state feature set.

7. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, The critical state of the evolution from false faults to real faults is determined, and an evolution warning signal and the expected critical time window are output. The specific steps are as follows: Obtain the current device's trajectory direction, instantaneous drift speed, drift acceleration, and distance to the fingerprint area boundary; The evolutionary critical state is jointly determined by multiple conditions. When the drift trajectory points from the ash-covered area to the hot spot budding area, and the drift acceleration a>0, and the distance from the current point to the boundary of the hot spot area is less than a preset threshold, it is the pre-hot spot formation stage. The current point is located in the hot spot budding area, the drift direction is pointing in the direction of increasing first contradiction deviation, the drift speed exceeds the threshold, and the infrared hot spot area increases three times in a row, which means that the power device is about to fail. If the drift trajectory points from the shadow movement area to the hot spot budding area, and the acceleration changes from negative to positive, then it is the pre-stage of shadow-induced local overheating. The current point is located within the dust accumulation and obscuring area, drifting towards the hot spot germination area, and its average velocity over past time points exceeds twice the historical average velocity. This indicates that it is in the pre-critical stage of accelerated dust accumulation. Starting from the two-dimensional deviation vector of the current point, calculate the time to reach the boundary of the target fingerprint region, multiply it by the security factor, and output the estimated time window.

8. The operation and maintenance management method for new energy power plants based on multi-source data according to claim 1, characterized in that, Based on the identified false fault types and the determination results of the evolutionary critical state, a graded handling instruction is generated. The specific steps are as follows: Based on the types of false failures and the critical state of evolution, a pre-set decision tree rule base is established, with planned handling corresponding to stable states and emergency handling corresponding to critical states. Establish a priority order for handling hotspots: hotspot germination > sensor drift > shadow movement > dust accumulation and occlusion, for resource allocation decisions when multiple tasks are concurrent; The command pushes the operation and maintenance scheduling system to create work orders. After execution, the system provides feedback on the results and updates the device status. The feedback data is also used for continuous optimization of the rule base.

Images