Photovoltaic power station power generation efficiency abnormal intelligent diagnosis system

Abstract

The application discloses a photovoltaic power station power generation efficiency anomaly intelligent diagnosis system, and relates to the technical field of photovoltaic power station anomaly diagnosis, solves the technical problems of single diagnosis trigger mechanism, lack of hierarchical processing capability and weak fine analysis capability of abnormal reasons, and the application divides the anomaly into three levels of mild, moderate and severe according to the threshold interval of power generation efficiency deviation, and starts first-level diagnosis and second-level diagnosis respectively, for slight efficiency loss, only first-level diagnosis is needed to complete the determination, without wasting unmanned aerial vehicle flight resources and complex computing power, for serious anomaly, high-precision multi-modal image diagnosis is automatically triggered, the hierarchical mechanism triggered on demand reduces the system operation cost and resource consumption compared with the fixed frequency inspection or single threshold trigger scheme in the prior art, while ensuring timely and accurate diagnosis of serious faults.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant anomaly diagnosis technology, specifically to an intelligent diagnostic system for abnormal power generation efficiency in photovoltaic power plants. Background Technology

[0002] As an important form of renewable energy generation, the power generation efficiency of photovoltaic (PV) power plants directly affects the economic benefits of the plant and the stable operation of the power grid. In actual operation, PV modules are often affected by various factors such as environmental factors, equipment aging, and sudden failures, leading to abnormal situations where power generation efficiency deviates from theoretical expectations. How to identify abnormal power generation efficiency and pinpoint the cause of the failure in a timely and accurate manner is a core issue in the intelligent operation and maintenance of PV power plants.

[0003] However, existing technologies often apply the same diagnostic process to all anomalies, indiscriminately using drone inspections or complex model analysis for minor efficiency losses and serious faults, resulting in a waste of computing resources and maintenance costs. Systems that rely solely on electrical parameter analysis are also unable to identify complex faults, leading to insufficient diagnostic accuracy. Furthermore, existing technologies terminate the diagnostic process after identifying anomaly clusters or regions, failing to further distinguish between independent faults and regional faults and to deeply analyze their root causes. This results in the inability to trace the source of the anomalies, making subsequent maintenance work inconvenient. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent diagnostic system for abnormal power generation efficiency in photovoltaic power plants, which solves the problems of a single diagnostic triggering mechanism, lack of hierarchical processing capabilities, and weak ability to conduct refined analysis of the causes of abnormalities.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants, comprising: The data acquisition and preprocessing module acquires real-time power generation data from the photovoltaic power station and performs preprocessing to obtain preprocessed real-time power generation data. The preprocessing includes missing value interpolation and outlier filtering correction. The real-time data analysis and processing module acquires pre-processed real-time power generation data, extracts the power generation characteristics of a single photovoltaic area, converts horizontal irradiance data into total irradiance on the tilted surface of the photovoltaic module, calculates the operating temperature of the photovoltaic module backsheet based on ambient temperature and wind speed data, calculates the theoretical output power and theoretical power generation efficiency based on the two, and calculates the real-time efficiency deviation. When the real-time efficiency deviation exceeds the preset threshold range, it marks the area's power generation efficiency as abnormal and generates an efficiency abnormality diagnostic signal. The intelligent diagnostic analysis module receives efficiency anomaly diagnostic signals, identifies the warning level based on real-time efficiency deviation, determines that a mild warning enters the first-level diagnostic analysis process, and a moderate and severe warning enters the second-level diagnostic process. For the first-level diagnostic analysis process, it calculates the current dispersion rate of each string in the power generation efficiency anomaly area, determines independent faults or regional faults based on the relationship between the dispersion rate and the dynamic threshold, and verifies the timing consistency of the determination results. For the secondary diagnostic analysis process, infrared and visible light images are collected by UAVs, and after preprocessing, they are input into a multimodal deep learning model to extract temperature distribution features and visible light spectral features respectively. The features of the two modes are then fused at the feature level to output the fault category prediction result.

[0006] As a further aspect of the present invention, the real-time power generation data includes the output power of the photovoltaic module, ambient temperature, solar irradiance, voltage and current at the grid connection point, and operating status data of each inverter. Preprocessing includes: interpolating and completing missing data; identifying and correcting outlier data that exceeds a reasonable range. Outlier data can be smoothed by using moving median filtering or removing outliers based on the 3σ criterion.

[0007] As a further aspect of the present invention, the method by which the real-time data analysis and processing module generates the efficiency anomaly diagnostic signal is as follows: According to the formula Tcell=Tamb+(GTI / GNOCT)×(T) NOCT -T amb，NOCT Calculate the operating temperature Tcell of the photovoltaic module backsheet, where GTI is the total irradiance on the photovoltaic tilt surface, GNOCT is the irradiance under NOCT test conditions, and T... NOCT T represents the standard operating temperature of the component under these conditions. amb，NOCT For the corresponding ambient temperature; Substituting the tilted surface irradiance GTI and the effective area A of the photovoltaic module into formula P in =GTI×A calculates the theoretical input power, and simultaneously uses the formula P theoretical =P max，stc ×(GTI / 1000)×η temp The theoretical output power P was calculated. theoretical , where P max，stc η is the standard power rating of the component at the time of manufacture. temp This is the temperature correction factor, and according to the formula η theoretical =P theoretical / P in The theoretical power generation efficiency η was calculated. theoretical Next, the actual power generation efficiency P is obtained. actual And according to the formula Δη=(P actual -Ptheoretical ) / P theoretical The real-time efficiency deviation Δη is calculated by multiplying by 100%. When the real-time efficiency deviation Δη exceeds the preset threshold range of normal fluctuation, the power generation efficiency of the region is marked as abnormal, and an efficiency abnormality diagnostic signal is generated.

[0008] As a further aspect of the present invention, the intelligent diagnostic analysis module identifies the warning level based on real-time efficiency deviation in the following way: When Δη∈[-A%,-B%), it indicates a mild warning and proceeds to the first-level diagnostic analysis process; when Δη∈[-C%,-A%), it indicates a moderate warning; and when Δη≤-C%), it indicates a severe warning and proceeds to the second-level diagnostic process.

[0009] As a further aspect of the present invention, the process for the primary diagnostic analysis workflow is as follows: Extract the real-time output current Ii of each string in the abnormal power generation efficiency area, and calculate the dispersion rate di of the current of each string. If the dispersion rate di is higher than the preset first threshold, it is determined that there is an independent fault in the string. If the dispersion rate di of all strings is lower than the first threshold, it is determined that there is a regional fault. Based on the identified anomalies, data from M consecutive diagnostic cycles are obtained. If at least K diagnostic cycles in the M consecutive diagnostic cycles are identified as having the same abnormal sequence or the same regional fault, then the abnormal event is determined to be valid. If the anomalies are inconsistent and gradually disappear, then the anomaly identification is stopped.

[0010] As a further aspect of the present invention, the process for the secondary diagnostic analysis workflow is as follows: The GPS coordinates of the power generation efficiency abnormal area are obtained, and an inspection command is issued to the UAV. The UAV collects infrared and visible light images of the power generation efficiency abnormal area according to the planned route, and performs preprocessing to obtain preprocessed area images. Image features corresponding to infrared and visible light images are extracted using a multimodal deep learning model. The acquired image features are then fused to obtain a fused feature vector, which is input into a classifier to output the fault category prediction result.

[0011] As a further aspect of the present invention, the multimodal deep learning model adopts a dual-branch encoder architecture: the first branch encoder receives infrared thermal imaging input based on an improved convolutional neural network and extracts temperature distribution features; the second branch encoder receives visible light image input based on an improved YOLO and extracts visible light spectral features.

[0012] As a further aspect of the present invention, it also includes a fault confirmation and analysis module. For the identified independent faults, it extracts the operating data of the abnormal sequence to construct a fault feature vector, matches it with a preset fault mode library, and outputs the fault cause and confidence level. For the identified regional faults, it constructs a photovoltaic power station knowledge graph and performs multi-level reasoning to output the specific cause of the regional fault.

[0013] As a further aspect of the present invention, the process for handling the identified independent faults is as follows: Construct a fault feature vector, which includes current dispersion rate, voltage deviation rate, power deviation rate, and short-circuit current, open-circuit voltage, maximum power point parameter, and fill factor extracted from the IV curve. Calculate the weighted Euclidean distance between the fault feature vector and the standard range of each fault type in the fault mode library, select the fault type with the smallest distance as the candidate fault cause, and determine the single or multiple candidate output based on the relationship between the minimum distance and the confidence threshold.

[0014] As a further aspect of the present invention, the process for handling the identified regional faults is as follows: Construct a photovoltaic power plant knowledge graph and extract the associated subgraphs corresponding to regional fault areas from the knowledge graph. Query whether the environmental parameters associated with all strings in the region are consistent within the fault time window. If the environmental parameters of all strings deviate from the normal range and the deviation direction is consistent, then start environmental association reasoning; if the environmental parameters are normal, then start common fault source reasoning. Environmental association reasoning means querying whether the readings of the irradiance sensors associated with the area are normal. If the sensor readings are normal but the actual irradiance conversion values ​​of each string in the area are low, it is inferred that there is a shared shading fault in the area; if the sensor readings are abnormal, it is inferred that there is a sensor fault. Common fault source reasoning involves querying the common device nodes associated with the region in the knowledge graph, and then checking the operating parameters of the common devices in turn to output the corresponding root cause.

[0015] This invention provides an intelligent diagnostic system for abnormal power generation efficiency in photovoltaic power plants. Compared with existing technologies, it has the following advantages: This invention categorizes anomalies into three levels—mild, moderate, and severe—based on the threshold range of the power generation efficiency deviation, and initiates primary and secondary diagnostics accordingly. For minor efficiency losses, a primary diagnostic test is sufficient for assessment, eliminating the need for drone flight resources and complex computing power. For severe anomalies, high-precision multimodal image diagnostics are automatically triggered. This on-demand triggering tiered mechanism, compared to existing fixed-frequency inspections or single-threshold triggering schemes, reduces system operating costs and resource consumption while ensuring timely and accurate diagnosis of severe faults.

[0016] Based on the primary diagnostic method of this invention, a fault feature vector matching method for independent faults and a knowledge graph multi-hop reasoning method for regional faults are designed. For independent faults, a multi-dimensional fault feature vector is constructed by extracting fault features and matching it with a preset fault pattern library, and a confidence score is given. For regional faults, a knowledge graph is constructed, and through multi-hop logic such as environmental homology verification, common fault source tracing, and cross-regional propagation reasoning, it can distinguish four types of regional root causes: dust-covered, shadow-occluded, attenuated, and systemic, and locate the specific source. Compared with the existing technology, which only outputs coarse-grained conclusions about regional faults, this invention improves the operability and guiding value of diagnosis. Attached Figure Description

[0017] Figure 1 This is a system block diagram of the intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to the present invention; Figure 2 This is a flowchart illustrating the implementation of the intelligent diagnostic system for abnormal power generation efficiency in photovoltaic power plants according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] First Embodiment Please see Figure 1 and Figure 2 This application provides an intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants, including: The data acquisition and preprocessing module acquires real-time power generation data from the photovoltaic power station and performs preprocessing to obtain preprocessed real-time power generation data. The preprocessing includes missing value interpolation and outlier filtering correction. The real-time data analysis and processing module acquires pre-processed real-time power generation data, extracts the power generation characteristics of a single photovoltaic area, converts horizontal irradiance data into total irradiance on the tilted surface of the photovoltaic module, calculates the operating temperature of the photovoltaic module backsheet based on ambient temperature and wind speed data, calculates the theoretical output power and theoretical power generation efficiency based on the two, and calculates the real-time efficiency deviation. When the real-time efficiency deviation exceeds the preset threshold range, it marks the area's power generation efficiency as abnormal and generates an efficiency abnormality diagnostic signal. The intelligent diagnostic analysis module receives efficiency anomaly diagnostic signals, identifies the warning level based on real-time efficiency deviation, determines that a mild warning enters the first-level diagnostic analysis process, and a moderate and severe warning enters the second-level diagnostic process. For the first-level diagnostic analysis process, it calculates the current dispersion rate of each string in the power generation efficiency anomaly area, determines independent faults or regional faults based on the relationship between the dispersion rate and the dynamic threshold, and verifies the timing consistency of the determination results. For the secondary diagnostic analysis process, infrared and visible light images are collected by UAVs, and after preprocessing, they are input into a multimodal deep learning model to extract temperature distribution features and visible light spectral features respectively. The features of the two modes are then fused at the feature level to output the fault category prediction result.

[0020] Second Embodiment As a second embodiment of the present invention, the difference from the first embodiment is as follows: The photovoltaic data acquisition and processing module collects real-time power generation data from the photovoltaic power station. This real-time power generation data includes: the output power of the photovoltaic modules, ambient temperature, solar irradiance, voltage and current at the grid connection point, and operating status data of each inverter. The collected real-time power generation data is preprocessed, including: interpolating and completing missing data, specifically using linear interpolation or cubic spline interpolation methods based on the changing trend of data points before and after the missing data; identifying and correcting abnormal data that exceeds the reasonable range, where the reasonable range is pre-set based on the physical constraints of photovoltaic modules and inverters and environmental extreme values, and abnormal data can be smoothed by using sliding median filtering or removing outliers based on the 3σ criterion. The processed data is organized into a preset standard format and stored in the diagnostic database. The pre-processed real-time power generation data is then transmitted to the real-time data analysis and processing module.

[0021] The real-time data analysis and processing module is used to analyze and process pre-processed real-time power generation data, extract the power generation characteristics of a single photovoltaic area, specifically obtain a single set of real-time power generation data sequences, and use wavelet transform to perform multi-resolution decomposition of the power sequences. The wavelet transform selects the db4 wavelet basis function and sets the hierarchy to 4 levels, decomposing the original power sequence into low-frequency components and high-frequency components. The low-frequency components correspond to the basic power generation trend of the photovoltaic area affected by changes in solar intensity and temperature, while the high-frequency components correspond to local fluctuations, instantaneous anomalies, and potential fault characteristics. At the same time, the Pearson correlation coefficient between each high-frequency component and the solar intensity sequence is calculated, and components with correlation coefficients lower than the preset coefficient are filtered out. The specific value of the preset coefficient is set by the operator. In this embodiment, the value is 0.3, which is identified as noise components and removed. The remaining components are superimposed and reconstructed to obtain the noise-reduced power sequence. Next, based on the solar intensity meter and direct irradiance meter deployed at the power station site, the total horizontal irradiance, horizontal diffuse irradiance, and direct irradiance data of the photovoltaic power station site are collected in real time. At the same time, the installation tilt angle, azimuth angle, ground albedo, and solar position parameters of the photovoltaic modules are obtained, and the direct irradiance component, the diffuse irradiance component, and the ground reflected irradiance component of the tilted surface are calculated. The three components are summed to obtain the total irradiance of the tilted surface. The calculation of the direct irradiance component of the tilted surface is as follows: Based on the solar zenith angle θz and solar azimuth angle φs, as well as the tilt angle β and azimuth angle φp of the photovoltaic module, the solar incident angle θi is calculated. Specifically, cosθi = cosθz × cosβ + sinθz × sinβ × (φs - φp), according to formula G. b，t =G b，n The direct irradiation component G on the inclined surface is calculated by multiplying by max(0, cosθi). b，t G b，n Normal direct radiation intensity; The irradiance component scattered by the tilted surface is modeled using the Perez model: G d，t =G d，h [(1-F1)×[(1+cosβ) / 2]+F1×a / b+F2×sinβ], where G d，h Let F1 and F2 be the horizontal surface scattered irradiance, and let a = max(0, cosθi) and b = max(0.087, cosθz). The ground reflected irradiance component is based on the formula G. r，t =ρ×G h ×[(1-cosβ) / 2], where ρ is the ground albedo, and is generally taken as 0.2, G h This represents the total irradiance on the horizontal plane. Collect ambient temperature and wind speed data, and apply the formula Tcell=Tamb+(GTI / GNOCT)×(T) NOCT -T amb，NOCT Calculate the operating temperature Tcell of the photovoltaic module backsheet, where GTI is the total irradiance on the tilted surface of the photovoltaic module, and GNOCT is the irradiance under NOCT test conditions, typically taken as 800 W / m. 2 T NOCT T represents the standard operating temperature of the component under these conditions. amb，NOCT For the corresponding ambient temperature; Based on the obtained tilt surface irradiance GTI and the effective area A of the photovoltaic module, the theoretical input power P is calculated. in =GTI×A, and introduce a temperature correction factor η temp =1+α temp ×(T cell -25), where αtemp The power temperature coefficient is given, and the temperature correction factor is substituted into the formula P. theoretical =P max，stc ×(GTI / 1000)×η temp The theoretical output power P was calculated. theoretical , where P max，stc This is the standard power of the component at the time of manufacture, and according to the formula η theoretical =P theoretical / P in The theoretical power generation efficiency η was calculated. theoretical Next, the actual power generation efficiency P is obtained. actual And according to the formula Δη=(P actual -P theoretical ) / P theoretical The real-time efficiency deviation Δη is calculated by multiplying by 100%. The real-time efficiency deviation Δη is compared with a preset threshold. When the real-time efficiency deviation Δη exceeds the preset threshold range of normal fluctuation, the power generation efficiency of the area is marked as abnormal. The specific value of the preset threshold is set by the operator, and an efficiency abnormality diagnosis signal is generated.

[0022] The intelligent diagnostic analysis module receives and analyzes efficiency anomaly diagnostic signals. Based on the obtained real-time efficiency deviation Δη, it performs early warning level identification. When Δη∈[-A%,-B%), it indicates a mild early warning and enters the first-level diagnostic analysis process. When Δη∈[-C%,-A%), it indicates a moderate early warning. When Δη≤-C%, it indicates a severe early warning and enters the second-level diagnostic process. The specific values ​​of A, B, and C are set by the operator, and B is greater than 0 and less than A, and A is less than C. For the first-level diagnostic analysis process, the real-time output current Ii of each string in the abnormal power generation efficiency area is extracted, and the dispersion rate di of the current of each string is calculated as follows: di = |Ii - median(I)| / median(I) × 100%, where median(I) is the median of the output current of all strings in the area. If the dispersion rate di is higher than the preset first threshold, it is determined that there is an independent fault in the string. If the dispersion rate di of all strings is lower than the first threshold, it is determined that there is a regional fault. The first threshold is obtained by acquiring the historical data of the current dispersion rate of each string under normal operating conditions in the abnormal efficiency area for the past 30 days, calculating its mean and standard deviation, and setting a dynamic threshold d. th =μ d +2.5σ d ; Based on the identified anomalies, data from M consecutive diagnostic cycles are obtained. If at least K diagnostic cycles in the M consecutive diagnostic cycles are determined to be the same abnormal sequence or the same regional fault, it indicates that the abnormal event is established and the diagnosis is marked with a high confidence level. If they are inconsistent but it has been confirmed that the average value of the current diagnostic cycle index deviates significantly from the normal range, the confidence level is reduced to low. If they are inconsistent and the anomalies gradually disappear, it indicates that the previous anomalies may be a timely phenomenon caused by environmental changes, and the anomaly identification is stopped. For the secondary diagnostic process, the GPS coordinates of the power generation efficiency abnormal area are obtained, and the inspection command is issued to the UAV. The UAV collects images of the power generation efficiency abnormal area according to the planned route. The collected images include infrared images and visible light images. At the same time, the collected images are preprocessed to obtain preprocessed area images. The preprocessing operations include image denoising, image enhancement, size standardization and coordinate registration. The preprocessed region image is input into a pre-trained multimodal deep learning model, which employs a dual-branch encoder architecture: the first branch encoder, based on an improved convolutional neural network, receives the infrared thermal image input and extracts temperature distribution features to identify temperature-related faults such as hot spots, solder detachment, and junction box overheating; the second branch encoder, based on an improved YOLO, receives the visible light image input and extracts visible light spectral features to identify visually visible faults such as dust accumulation, bird droppings, physical cracks, surface contamination, and weed obstruction, extracting corresponding image features and fusing the acquired image features to generate the infrared feature map F. IR and visible light feature map F RGB After undergoing one-dimensional global average pooling, global feature vectors v for the two modes are obtained. IR and v RGB The attention weights α and β for each of the two modalities are calculated using a two-layer fully connected network (FC-ReLU-FC) and a softmax function, where α + β = 1. The final fused feature vector v fusion =α×v IR +β×v RGB The fused feature vector v fusion The input is fed into a subsequent classifier, which outputs the fault category prediction result.

[0023] Third Embodiment As a third embodiment of the present invention, it is implemented based on the first embodiment, and the difference from the first embodiment is that it also includes a fault confirmation and analysis module. This module is used to perform specific fault cause analysis on the independent faults or regional faults identified in the first-level diagnostic process. The specific identification and analysis of independent faults in the string is as follows: The system acquires marked abnormal strings and extracts corresponding continuous operating data, along with corresponding fault features. These features include current discrete characteristics, voltage offset rate, power deviation rate, and short-circuit current, open-circuit voltage, maximum power point parameters, and fill factor extracted from the IV curve. A fault feature vector is then constructed based on these extracted features and matched against a pre-established fault mode library. For example, if the fault is caused by partial occlusion, the corresponding features include significantly low short-circuit current (Isc), basically normal open-circuit voltage (Voc), and a step-like distortion in the IV curve. If the fault is caused by surface dirt / dust coverage, the corresponding features include significantly decreased short-circuit current (Isc), basically normal open-circuit voltage (Voc), and an IV curve that is intact but proportionally narrowed. The specific matching method is as follows: Calculate the weighted Euclidean distance between the fault feature vector and the standard range of each fault type, and select the fault type with the smallest distance as the candidate fault cause of the anomaly string. At the same time, compare the smallest distance with the confidence threshold. When the smallest distance is less than the first confidence threshold, directly output the fault type and its corresponding confidence. When the smallest distance is between the first confidence threshold and the second confidence threshold, output multiple candidate fault causes and sort them in descending order of confidence. The specific identification and analysis of regional faults is as follows: First, a photovoltaic power station knowledge graph is constructed using the Neo4j graph database. The knowledge graph uses electrical connection relationships, spatial proximity relationships, environmental sharing relationships, and historical fault association relationships as relation edges. Nodes include at least equipment entity nodes, environmental condition nodes, and fault event nodes. The initial weights of the relation edges are calculated based on the equipment's rated parameters and topological distance. For regions marked as having regional faults, a related subgraph is extracted from the knowledge graph. This subgraph contains all string nodes within the region, as well as associated meteorological nodes, shadow source nodes, and historical fault nodes. Regional fault analysis is then performed based on this related subgraph, and the specific analysis process is as follows: The query checks whether the environmental parameters associated with all strings within the query area are consistent within the fault time window. These environmental parameters include irradiance and temperature values. If the environmental parameters of all strings deviate from the normal range and the direction of deviation is consistent (e.g., all are too low or all are too high), then environmental association reasoning is initiated. If the environmental parameters are normal, then the query switches to common fault source reasoning. For environmental management reasoning, check whether the readings of the irradiance sensors associated with the area are normal. If the sensor readings are normal but the actual irradiance conversion values ​​of each string in the area are low, it is inferred that the area is a shared shading fault. If the sensor readings are abnormal, it is inferred that the sensor is faulty. For common fault source reasoning, if the environmental parameters are normal but the power deviation rate of all strings in the area has decreased significantly and the decrease is similar, query the common device nodes associated with this area in the knowledge graph and check the operating parameters of the common devices in turn: if the output voltage of the combiner box has an abnormal voltage drop relative to the sum of the input voltages of each string, output combiner box connection fault root cause; if the inverter conversion efficiency decreases significantly and exceeds the normal fluctuation range during this period, output inverter efficiency abnormal root cause.

[0024] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0025] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A photovoltaic power plant power generation efficiency anomaly intelligent diagnostic system, characterized in that, include: The data acquisition and preprocessing module acquires real-time power generation data from the photovoltaic power station and performs preprocessing to obtain preprocessed real-time power generation data. The preprocessing includes missing value interpolation and outlier filtering correction. The real-time data analysis and processing module acquires pre-processed real-time power generation data, extracts the power generation characteristics of a single photovoltaic area, converts horizontal irradiance data into total irradiance on the tilted surface of the photovoltaic module, calculates the operating temperature of the photovoltaic module backsheet based on ambient temperature and wind speed data, calculates the theoretical output power and theoretical power generation efficiency based on the two, and calculates the real-time efficiency deviation. When the real-time efficiency deviation exceeds the preset threshold range, it marks the area's power generation efficiency as abnormal and generates an efficiency abnormality diagnostic signal. The intelligent diagnostic analysis module receives efficiency anomaly diagnostic signals, identifies the warning level based on real-time efficiency deviation, determines that a mild warning enters the first-level diagnostic analysis process, and a moderate and severe warning enters the second-level diagnostic process. For the first-level diagnostic analysis process, it calculates the current dispersion rate of each string in the power generation efficiency anomaly area, determines independent faults or regional faults based on the relationship between the dispersion rate and the dynamic threshold, and verifies the timing consistency of the determination results. For the secondary diagnostic analysis process, infrared and visible light images are collected by UAVs, and after preprocessing, they are input into a multimodal deep learning model to extract temperature distribution features and visible light spectral features respectively. The features of the two modes are then fused at the feature level to output the fault category prediction result.

2. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, The real-time power generation data includes the output power of photovoltaic modules, ambient temperature, solar irradiance, voltage and current at the grid connection point, and operating status data of each inverter. Preprocessing includes: interpolating and completing missing data; identifying and correcting outlier data that exceeds a reasonable range. Outlier data can be smoothed by using moving median filtering or removing outliers based on the 3σ criterion.

3. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, The real-time data analysis and processing module generates efficiency anomaly diagnostic signals in the following way: The backsheet operating temperature Tceii of a photovoltaic module is calculated according to the formula Tceii = Tamb + (GTI / GNOCT) x (T NOCT - T amb，NOCT ), where GTI is the total irradiance on the photovoltaic tilted surface, GNOCT is the irradiance under NOCT test conditions, T NOCT is the standard operating temperature of the module under these conditions, and Tamb is the corresponding ambient temperature. amb，NOCT ​ Substituting the tilted surface irradiance GTI and the effective area A of the photovoltaic module into formula P in =GTI×A calculates the theoretical input power, and simultaneously uses the formula P theoretical =P max，stc ×(GTI / 1000)×η temp The theoretical output power P was calculated. theoretical , where P max，stc η is the standard power rating of the component at the time of manufacture. temp This is the temperature correction factor, and according to the formula η theoretical =P theoretical / P in The theoretical power generation efficiency η was calculated. theoretical Next, the actual power generation efficiency P is obtained. actual And according to the formula Δη=(P actual -P theoretical ) / P theoretical The real-time efficiency deviation Δη is calculated by multiplying by 100%. When the real-time efficiency deviation Δη exceeds the preset threshold range of normal fluctuation, the power generation efficiency of the region is marked as abnormal, and an efficiency abnormality diagnostic signal is generated.

4. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, The intelligent diagnostic analysis module identifies the warning level based on real-time efficiency deviation in the following way: When Δη∈[-A%,-B%), it indicates a mild warning and enters the first-level diagnostic analysis process. When Δη∈[-C%,-A%), it indicates a moderate warning. When Δη≤-C%, it indicates a severe warning and enters the second-level diagnostic process, and B is greater than 0 and less than A, and A is less than C.

5. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, The process for the primary diagnostic analysis workflow is as follows: Extract the real-time output current Ii of each string in the abnormal power generation efficiency area, and calculate the dispersion rate di of the current of each string. If the dispersion rate di is higher than the preset first threshold, it is determined that there is an independent fault in the string. If the dispersion rate di of all strings is lower than the first threshold, it is determined that there is a regional fault. Based on the identified anomalies, data from M consecutive diagnostic cycles are obtained. If at least K diagnostic cycles in the M consecutive diagnostic cycles are determined to be the same abnormal sequence or the same regional fault, then the abnormal event is determined to be valid. If the inconsistency gradually disappears, then the anomaly identification process should be stopped.

6. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, The process for the secondary diagnostic analysis workflow is as follows: The GPS coordinates of the power generation efficiency abnormal area are obtained, and an inspection command is issued to the UAV. The UAV collects infrared and visible light images of the power generation efficiency abnormal area according to the planned route, and performs preprocessing to obtain preprocessed area images. Image features corresponding to infrared and visible light images are extracted using a multimodal deep learning model. The acquired image features are then fused to obtain a fused feature vector, which is input into a classifier to output the fault category prediction result.

7. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 6, characterized in that, The multimodal deep learning model adopts a dual-branch encoder architecture: the first branch encoder receives infrared thermal imaging input based on an improved convolutional neural network and extracts temperature distribution features; The second branch encoder receives visible light image input based on an improved YOLO and extracts visible light spectral features.

8. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 1, characterized in that, It also includes a fault confirmation and analysis module. For the identified independent faults, it extracts the operating data of the abnormal series to construct a fault feature vector, matches it with the preset fault mode library, and outputs the fault cause and confidence level. For the identified regional faults, it constructs a photovoltaic power station knowledge graph and performs multi-level reasoning to output the specific cause of the regional fault.

9. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 8, characterized in that, The process for handling identified independent faults is as follows: Construct a fault feature vector, which includes current dispersion rate, voltage deviation rate, power deviation rate, and short-circuit current, open-circuit voltage, maximum power point parameter, and fill factor extracted from the IV curve. Calculate the weighted Euclidean distance between the fault feature vector and the standard range of each fault type in the fault mode library, select the fault type with the smallest distance as the candidate fault cause, and determine the single or multiple candidate output based on the relationship between the minimum distance and the confidence threshold.

10. The intelligent diagnostic system for abnormal power generation efficiency of photovoltaic power plants according to claim 8, characterized in that, The process for handling the identified regional faults is as follows: Construct a photovoltaic power plant knowledge graph and extract the associated subgraphs corresponding to regional fault areas from the knowledge graph. Query whether the environmental parameters associated with all strings in the region are consistent within the fault time window. If the environmental parameters of all strings deviate from the normal range and the deviation direction is consistent, then start environmental association reasoning; if the environmental parameters are normal, then start common fault source reasoning. Environmental association reasoning means querying whether the readings of the irradiance sensors associated with the area are normal. If the sensor readings are normal but the actual irradiance conversion values ​​of each string in the area are low, it is inferred that there is a shared shading fault in the area. If the sensor reading is abnormal, it is inferred that the sensor is faulty; Common fault source reasoning involves querying the common device nodes associated with the region in the knowledge graph, and then checking the operating parameters of the common devices in turn to output the corresponding root cause.

Images