Photovoltaic power station fault supervision system based on neural network model

Abstract

The invention relates to the technical field of photovoltaic system monitoring, in particular to a photovoltaic power station fault supervision system based on a neural network model, and the system comprises a data collection module which is used for collecting string current data and voltage data; the IV analysis module is used for obtaining a fault diagnosis result; the digital power station modeling module is used for identifying and numbering the coordinates of the photovoltaic string; an inspection route planning module; the unmanned aerial vehicle inspection module is used for performing inspection control on the unmanned aerial vehicle to obtain inspection data; an inspection data uploading module; the CV analysis module is used for performing defect identification on the inspection data to obtain a defect identification result; the fusion inspection analysis module is used for outputting a fusion diagnosis result and also used for selecting a final fusion diagnosis result; the visual presentation module is used for carrying out visual presentation; the report module is used for generating a report; and the navigation defect elimination module is used for carrying out navigation defect elimination. According to the invention, the automation level and inspection efficiency of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic system monitoring, and in particular to a photovoltaic power station fault monitoring system based on a neural network model. Background Art

[0002] With the completion of photovoltaic power stations, a large number of photovoltaic modules may encounter abnormal conditions such as dust, dirt, obstruction, bird droppings, fragmentation, diode damage, and string drop during operation, which will cause great power loss over time. At present, the operation and maintenance of photovoltaic power stations are mainly manpower-based, supplemented by monitoring systems. Although the main power generation parameters and meteorological data can be monitored in the background monitoring system, the photovoltaic power stations have many equipment and are widely distributed. The photovoltaic areas are at high altitudes and have strong radiation, which bring great difficulties to the operation and maintenance work. Therefore, it is urgent to introduce a fault diagnosis method that integrates IV scanning based on electrical quantities and CV inspection based on machine vision to make up for the shortcomings of conventional operation and maintenance methods and maximize the automation level of fault diagnosis and inspection efficiency.

[0003] Chinese patent publication number: CN116923156B discloses a cloud-based intelligent safety monitoring system for photovoltaic energy storage charging stations, which analyzes environmental information that affects charging equipment, realizes environmental risk control of charging equipment, and facilitates managers to respond to possible unexpected situations and risk probabilities in advance. Through the photovoltaic equipment monitoring unit, the comprehensive analysis of the operation time and power generation capacity of photovoltaic power generation equipment enables managers to timely understand the power generation efficiency of photovoltaic energy storage charging stations, carry out maintenance when abnormalities occur, reduce the dependence of photovoltaic energy storage power stations on power grid power supply, monitor the parameters of charging equipment during operation, avoid charging equipment running with faults, analyze the frequency of use of charging equipment, and generate reminders when the frequency of use of certain charging equipment is significantly lower than the normal frequency of use. However, this solution requires managers to be able to timely understand the power generation efficiency of photovoltaic energy storage charging stations and carry out maintenance when abnormalities occur. It still relies closely on manual intervention and cannot maximize the automation level of fault diagnosis and inspection efficiency. Summary of the invention

[0004] To this end, the present invention provides a photovoltaic power station fault monitoring system based on a neural network model to overcome the problem in the prior art that the operation and maintenance of photovoltaic power stations is mainly manpower-based, supplemented by a monitoring system, which leads to difficulties in operation and maintenance work.

[0005] To achieve the above object, the present invention provides a photovoltaic power station fault monitoring system based on a neural network model, comprising:

[0006] Data acquisition module, used to collect string current data and voltage data;

[0007] An IV analysis module, used to perform fault diagnosis according to the string current data and the voltage data to obtain a fault diagnosis result;

[0008] The digital power station modeling module is used to construct the basic model of the power station, identify and number the coordinates of the photovoltaic strings, adjust the coordinate numbers of the photovoltaic strings, and supplement the blank numbers;

[0009] Inspection route planning module, used to generate UAV flight strategies and plan UAV routes;

[0010] The UAV inspection module is used to control the UAV inspection and obtain the inspection data;

[0011] An inspection data uploading module, used to upload the inspection data to a CV analysis module;

[0012] A CV analysis module is used to obtain defect recognition results, to obtain defect location results, to automatically recognize defects, to optimize the defect recognition results according to the defect automatic recognition results, to automatically locate defects, to optimize the defect location results according to the defect automatic location results, to perform defect deduplication, and to adjust the optimization process of the defect location results according to defect deduplication;

[0013] The fusion inspection analysis module is used to output the fusion diagnosis results and select the final fusion diagnosis results;

[0014] A visualization module for visualization;

[0015] Report module, used to generate reports;

[0016] The navigation defect elimination module is used for navigation defect elimination.

[0017] Furthermore, the IV analysis module generates IV curve data according to the string current data and the voltage data, and sets the IV curve formula as n is the diode ideal factor, Rs is the series resistance of the photovoltaic string, Rsh is the parallel resistance of the photovoltaic string, q is the single electron charge, k is the Boltzmann constant, T is the thermodynamic temperature, and the IV analysis module inputs the IV curve data into the fault diagnosis model, and obtains the output result of the fault diagnosis model as the fault diagnosis result.

[0018] Furthermore, the digital power station modeling module performs aerial triangulation on the data collected by the drone aerial photographs and constructs a dense point cloud to generate orthophotos and digital surface models, and uses the orthophotos and digital surface models as the basic model of the power station.

[0019] Furthermore, the digital power station modeling module identifies and numbers the photovoltaic string coordinates according to the intelligent numbering algorithm and the sub-array KML of the station to obtain the photovoltaic string coordinates and the photovoltaic string coordinate numbers.

[0020] Furthermore, the digital power station modeling module obtains the number of the photovoltaic string coordinates of the station construction CAD drawing, takes it as the preset photovoltaic string coordinate number, and compares the preset photovoltaic string coordinate number with the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number according to the comparison result, wherein:

[0021] When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is consistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is correct and does not adjust the photovoltaic string coordinate number;

[0022] When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is inconsistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is wrong, adjusts the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number of the same photovoltaic string coordinate to the preset photovoltaic string coordinate number.

[0023] Furthermore, the inspection route planning module inputs the power station basic model, the photovoltaic string coordinates and the photovoltaic string coordinate numbers into the management system to generate a UAV flight strategy, obtains the UAV flight strategy output by the management system, generates a refined inspection instruction according to the UAV flight strategy, and plans the UAV route according to the refined inspection instruction;

[0024] The UAV inspection module performs inspection control on the UAV according to the UAV route to obtain inspection data.

[0025] Furthermore, the CV analysis module selects YOLO-v5 as the basic algorithm model, trains the basic algorithm model to obtain a defect recognition algorithm, inputs the inspection data into the defect recognition algorithm, and obtains a defect recognition result output by the defect recognition algorithm corresponding to the inspection data.

[0026] Furthermore, when the CV analysis module trains the defect recognition algorithm, the xml file and the picture annotated by labelImg are used as defect recognition algorithm training data, and the defect recognition algorithm is trained according to a defect recognition training method, and the defect recognition training method includes:

[0027] Step S1, generating four files, train.txt, val.txt, test.txt and trainval.txt, according to the defect recognition algorithm training data, and naming them as training set, verification set and test set respectively;

[0028] Step S2, converting the defect recognition algorithm training data into yolo_txt format to obtain converted training data, that is, converting each xml file and picture in the defect recognition algorithm training data into txt format, each line of the txt file is information of a bbox target, and each line includes class, x_center, y_center, width, height;

[0029] Step S3, selecting the medium-sized architecture YOLO-v5m in YOLO-v5 as the basic algorithm model;

[0030] Step S4, training the basic algorithm model according to the training set in the converted training data to obtain a trained basic algorithm model, and verifying the trained basic algorithm model according to the verification set in the converted training data, outputting the trained basic algorithm model with an accuracy rate of 90% as a verified basic algorithm model, and retraining the trained basic algorithm model with an accuracy rate of less than 90%;

[0031] Step S5, testing the verified basic algorithm model according to the test set in the converted training data, outputting the verified basic algorithm model with an accuracy rate of 98% as the defect recognition algorithm, and retraining the verified basic algorithm model with an accuracy rate less than 90%.

[0032] Furthermore, when the CV analysis module locates defects in the inspection data, the accuracy of the GPS position is improved to the pixel level, and the elevation and position information of different pixel points in the image are calculated by SFM technology in combination with the flight altitude of the drone, the camera sensor size and the lens size, and the size of the global photo component is corrected to obtain the entire consistent ground resolution image, and the entire consistent ground resolution image is input into the defect location algorithm to solve the accurate position information of the faulty component;

[0033] The CV analysis module automatically identifies defects based on thermal imaging and machine vision to obtain automatic defect identification results;

[0034] The CV analysis module adds the automatic defect recognition result to the defect recognition result to optimize the defect recognition result;

[0035] The CV analysis module automatically locates the defect according to the photogrammetry principle and the aerial triangulation calculation method, and optimizes the defect location result according to the defect automatic location result;

[0036] The CV analysis module performs defect deduplication according to the thermal imaging fault recognition algorithm to obtain a defect deduplication result, and adjusts the optimization process of the defect location result according to the defect deduplication result.

[0037] Furthermore, when performing fusion diagnosis, the fusion inspection and analysis module generates a fusion analysis report according to the diagnosis mode, wherein:

[0038] When the diagnostic mode is the IV diagnostic trigger mode, the fusion inspection analysis module uses the fault diagnosis result obtained by the IV analysis module as the main analysis report, and the defect identification result output by the CV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report;

[0039] When the diagnosis mode is the CV analysis unit trigger mode, the fusion inspection analysis module uses the defect recognition result output by the CV analysis module as the main analysis report, and the fault diagnosis result obtained by the IV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report;

[0040] The fusion inspection and analysis module outputs each fusion analysis report as a fusion diagnosis result of each diagnosis mode;

[0041] The fusion inspection analysis module inputs the fusion diagnosis data into the diagnosis effect analysis model, obtains the effective value of the fusion diagnosis result of each diagnosis mode output by the diagnosis effect analysis model, and selects the fusion diagnosis result of the diagnosis mode corresponding to the maximum effective value as the final fusion diagnosis result.

[0042] Compared with the prior art, the beneficial effects of the present invention are that the system accurately collects string current and voltage data through the data acquisition module to provide a reliable basis for fault diagnosis, the system uses the collected data to perform fault diagnosis and quickly identify abnormalities through the IV analysis module, the system builds a power plant model and accurately locates it through the digital power plant modeling module to support inspection route planning, the system generates a drone flight strategy through the inspection route planning module to improve inspection efficiency and coverage, the system performs automated inspection tasks through the drone inspection module to reduce manual operations and improve data collection efficiency, the system transmits inspection data in real time through the inspection data upload module to support real-time data processing, the system uses machine vision and thermal imaging technology through the CV analysis module to achieve high-precision automatic defect detection, the system combines multi-source data for comprehensive diagnosis through the fusion inspection analysis module to improve diagnostic accuracy, the system intuitively displays the analysis results through the visualization presentation module to facilitate decision-making and operation, the system generates a detailed analysis report through the report module to support the beneficial effects of operation and maintenance decisions, and the system guides accurate defect handling through the navigation and defect elimination module to improve maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the structure of the photovoltaic power station fault monitoring system based on the neural network model in this embodiment. DETAILED DESCRIPTION

[0044] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0046] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] See also Figure 1 As shown, it is a structural diagram of a photovoltaic power station fault monitoring system based on a neural network model in this embodiment, and the system includes:

[0048] Data acquisition module, used to collect string current data and voltage data;

[0049] An IV analysis module, used to perform fault diagnosis according to the string current data and the voltage data to obtain a fault diagnosis result;

[0050] The digital power station modeling module is used to construct the basic model of the power station based on the data collected by the drone aerial photos, identify and number the coordinates of the photovoltaic strings according to the intelligent numbering algorithm, adjust the coordinate numbers of the photovoltaic strings according to the station construction CAD drawings, and supplement the blank numbers according to the route planning software;

[0051] Inspection route planning module, used to generate UAV flight strategy according to the basic model of power station, PV string coordinates and PV string coordinate numbers, and plan the UAV route according to the UAV flight strategy;

[0052] The UAV inspection module is used to control the UAV inspection according to the UAV route and obtain the inspection data;

[0053] An inspection data uploading module, used to upload the inspection data to a CV analysis module;

[0054] A CV analysis module, for performing defect recognition on the inspection data according to a defect recognition algorithm to obtain a defect recognition result, for performing defect location on the inspection data according to a defect location algorithm to obtain a defect location result, for automatically recognizing defects according to thermal imaging and machine vision, for optimizing the defect recognition result according to the automatic defect recognition result, for automatically locating defects according to photogrammetry principles and aerial triangulation calculation methods, for optimizing the defect location result according to the automatic defect location result, for performing defect deduplication according to a thermal imaging fault recognition algorithm, and for adjusting the optimization process of the defect location result according to the defect deduplication;

[0055] The fusion inspection analysis module is used to perform fusion diagnosis and output fusion diagnosis results, and is also used to select the final fusion diagnosis results according to the diagnosis effect analysis model;

[0056] A visualization presentation module, used for visually presenting the fault diagnosis result, the defect identification result, the defect location result and the final fusion diagnosis result;

[0057] A reporting module, for generating a report according to the fault diagnosis result, the defect identification result, the defect location result and the fusion diagnosis result after visual presentation;

[0058] The navigation and defect elimination module is used to perform navigation and defect elimination according to the fault diagnosis result, the defect identification result, the defect location result and the fusion diagnosis result after the report is generated.

[0059] Specifically, the system is arranged in a photovoltaic power station fault monitoring terminal, and is applied to a photovoltaic power station fault monitoring architecture of the photovoltaic power station fault monitoring terminal to perform fault monitoring on the photovoltaic power station. A fault diagnosis method integrating IV scanning based on electrical quantities and CV inspection based on machine vision is introduced to make up for the deficiencies of conventional operation and maintenance means, thereby maximizing the automation level of fault diagnosis and inspection efficiency. The system accurately collects string current and voltage data through a data acquisition module to provide a reliable basis for fault diagnosis. The system uses the collected data through an IV analysis module to perform fault diagnosis and quickly identify anomalies. The system constructs a power station model and performs precise positioning through a digital power station modeling module to support inspection route planning. The system plans inspection routes through inspection routes. The module generates drone flight strategies to improve inspection efficiency and coverage. The system performs automated inspection tasks through the drone inspection module to reduce manual operations and improve data collection efficiency. The system transmits inspection data in real time through the inspection data upload module to support real-time data processing. The system uses machine vision and thermal imaging technology through the CV analysis module to achieve high-precision automatic defect detection. The system combines multi-source data for comprehensive diagnosis through the fusion inspection analysis module to improve diagnostic accuracy. The system intuitively displays the analysis results through the visualization presentation module to facilitate decision-making and operation. The system generates detailed analysis reports through the reporting module to support the beneficial effects of operation and maintenance decisions. The system guides the precise handling of defects through the navigation and defect elimination module to improve maintenance efficiency.

[0060] Specifically, the data acquisition module sends the IV scanning instruction to the inverter through the data acquisition device, and collects the string current data and voltage data through the inverter. The string current data includes the photovoltaic string operating current I, the photovoltaic string photocurrent IL and the diode reverse saturation current I0, and the voltage data includes the photovoltaic string operating voltage V.

[0061] Specifically, the photovoltaic string operating current I refers to the current output by the photovoltaic cell and photovoltaic module under actual working conditions, the photovoltaic string photocurrent IL refers to the current generated by the photovoltaic cell due to the photovoltaic effect after receiving light, the diode reverse saturation current I0 refers to the leakage current of the internal diode of the photovoltaic cell under reverse bias conditions, and the photovoltaic string operating voltage V refers to the output voltage of the photovoltaic module under actual working conditions.

[0062] Specifically, the IV analysis module generates IV curve data according to the string current data and the voltage data, and sets the IV curve formula as follows: n is the diode ideal factor, Rs is the series resistance of the photovoltaic string, Rsh is the parallel resistance of the photovoltaic string, q is the single electron charge, k is the Boltzmann constant, T is the thermodynamic temperature, and the IV analysis module inputs the IV curve data into the fault diagnosis model, and obtains the output result of the fault diagnosis model as the fault diagnosis result.

[0063] Specifically, the fault diagnosis result refers to the output result corresponding to the string current data and the voltage data obtained after the string current data and the voltage data are input into the fault diagnosis model, including current mismatch within the string, abnormal component current output, abnormal string voltage, too low string parallel resistance, string open circuit, too high string series resistance and low string short-circuit current. The fault diagnosis model refers to a neural network model that can analyze and output the fault diagnosis result corresponding to the IV curve data after inputting the IV curve data. The IV curve data refers to the curve graph data generated according to the string current data and the voltage data and the IV curve formula in which I is the vertical axis and V is the horizontal axis. In this embodiment, a convolutional neural network model is selected as the initial model of the fault diagnosis model, and a data group in the form of historical IV curve data-fault diagnosis result is used as the fault diagnosis model construction data, and 70% of the fault diagnosis model construction data is used as the fault diagnosis model. The fault diagnosis model training set is used as the fault diagnosis model test set. 30% of the fault diagnosis model construction data is used as the fault diagnosis model test set. The fault diagnosis model training set is input into the convolutional neural network model for model training to obtain the convolutional neural network model after fault diagnosis training. The fault diagnosis model test set is input into the convolutional neural network model after fault diagnosis training. The convolutional neural network model after fault diagnosis training, which has a fault diagnosis result consistent with that in the fault diagnosis model test set and accounts for 98%, is used as the fault diagnosis model for output. The diode ideal factor refers to a parameter related to the diode characteristics, which reflects the ideality of the diode. It is usually between 1 and 2. The closer the value is to 1, the closer the diode is to the ideal state. The photovoltaic string series resistance refers to the sum of the resistances inside the photovoltaic cell and at the connection point. The photovoltaic string parallel resistance refers to the leakage resistance inside the photovoltaic cell. The single electron charge refers to the basic charge of the electron, which is 1.602×10 -19 Coulomb, the Boltzmann constant is used to describe the behavior of particles in thermodynamics and is 1.381×10 -23 Joule / Kelvin, the thermodynamic temperature refers to the absolute temperature, expressed in Kelvin.

[0064] Specifically, the digital power station modeling module performs aerial triangulation on the data collected by the drone aerial photographs and constructs a dense point cloud, generates orthophotos and digital surface models, and uses the orthophotos and digital surface models as the basic model of the power station.

[0065] Specifically, the drone aerial photograph acquisition data refers to the image data taken by the camera equipment carried by the drone when it is flying in the air; the aerial triangulation encryption, that is, aerial triangulation encryption, refers to the use of the overlapping parts of multiple aerial photographs to perform geometric correction and position solution through mathematical methods to determine the shooting position and posture of each image, which is the basic step of three-dimensional modeling; the construction of dense point cloud refers to the process of extracting a large number of three-dimensional points from multi-perspective images to represent the terrain and structure; the orthophoto refers to the corrected aerial image, whose geometric distortion is eliminated, and the position of the objects is as if looking down vertically from above, which can be used for measurement and mapping and has a uniform scale; the digital surface model refers to a three-dimensional model that represents the surface including the height of all objects (such as buildings, trees, etc.); the power station basic model refers to the basic three-dimensional model of the power station area generated after processing the drone data.

[0066] Specifically, the digital power station modeling module identifies and numbers the photovoltaic string coordinates according to the intelligent numbering algorithm and the sub-array KML of the station to obtain the photovoltaic string coordinates and the photovoltaic string coordinate numbers.

[0067] Specifically, the intelligent numbering algorithm refers to an algorithm for automatically assigning unique identifiers to photovoltaic strings. This embodiment does not limit the setting method of the intelligent numbering algorithm. Those skilled in the art can freely set it according to actual conditions, as long as the need to automatically assign unique identifiers to photovoltaic strings is met. For example, numbering can be set based on geographic location, string arrangement mode or other related attributes; the subarray KML of the station refers to an XML format for representing geographic data. The subarray KML file of the station contains the geographic boundaries and location descriptions of different subarrays in the photovoltaic power station, which is used to display and manage various parts of the power station on a map; the photovoltaic string coordinates refer to the specific location coordinates of each photovoltaic string in geographic space. , usually expressed in the form of longitude and latitude, used to determine the physical location of each string; the photovoltaic string coordinate number refers to a unique number assigned to each photovoltaic string according to an intelligent numbering algorithm, which is used to manage and identify different strings; the process of identifying the photovoltaic string coordinates refers to the process of using image processing technology and spatial analysis algorithms, combined with the boundary information of the KML file, to accurately determine the location of each string; the process of numbering the photovoltaic string coordinates refers to assigning a unique number to each photovoltaic string coordinate through the intelligent numbering algorithm after identifying the photovoltaic string coordinates, and the unique number refers to a number that ensures the logic and uniqueness of the numbering after considering the arrangement and position of the strings and other factors that may affect the numbering system.

[0068] Specifically, the digital power station modeling module obtains the number of the photovoltaic string coordinates of the station construction CAD drawing, takes it as the preset photovoltaic string coordinate number, and compares the preset photovoltaic string coordinate number with the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number according to the comparison result, wherein:

[0069] When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is consistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is correct and does not adjust the photovoltaic string coordinate number;

[0070] When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is inconsistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is wrong, adjusts the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number of the same photovoltaic string coordinate to the preset photovoltaic string coordinate number.

[0071] Specifically, the station construction CAD drawings refer to computer-aided design drawings used in the construction process of photovoltaic power stations. The methods of obtaining the station construction CAD drawings include exporting from design software, downloading through the file transfer protocol (FTP) or obtaining from a project management platform; the photovoltaic string coordinate numbers of the station construction CAD drawings refer to the numbers assigned by the design team according to the layout logic or installation sequence, and are used to identify the layout and position of the strings in the drawings; the preset photovoltaic string coordinate numbers of the same photovoltaic string coordinates are consistent with the photovoltaic string coordinate numbers, which means that the preset numbers in the CAD drawings are consistent with the numbers identified in the digital modeling process, which indicates that no position misjudgment or recognition error occurs in the modeling process, the numbers are correct, and therefore no adjustment is required.

[0072] Specifically, after adjusting the photovoltaic string coordinate numbers, the digital power station modeling module inputs the photovoltaic string coordinates and the photovoltaic string coordinate numbers into the route planning software, and supplements the blank numbers through the route planning software.

[0073] Specifically, the route planning software refers to software used to supplement blank numbers, such as DroneDeploy, Pix4Dcapture, DJI Terra, UgCS, etc. When the blank numbers are supplemented by the route planning software, the input photovoltaic string coordinates and numbers are analyzed in the route planning software to identify blank areas. After identifying the blank areas, the route planning software automatically generates new numbers based on the surrounding numbered strings, and based on the spatial position, numbering sequence and preset logical rules.

[0074] Specifically, the inspection route planning module inputs the basic model of the power station, the photovoltaic string coordinates and the photovoltaic string coordinate numbers into the management system to generate a drone flight strategy, obtains the drone flight strategy output by the management system, generates refined inspection instructions based on the drone flight strategy, and plans the drone route based on the refined inspection instructions.

[0075] Specifically, the management system refers to a comprehensive software platform for integrating, processing and analyzing the basic model of the power station, the photovoltaic string coordinates and the photovoltaic string coordinate numbers. The management system is provided with a coverage optimization algorithm, and the UAV flight strategy is generated by the coverage optimization algorithm; the UAV flight strategy refers to a strategy for guiding the UAV on how to cover the entire power station for inspection, including path planning, time scheduling, resource allocation, etc.; the inspection route planning module inputs the UAV flight strategy into the expert refinement database, and obtains the stored UAV flight strategy consistent with the UAV flight strategy in the expert refinement database, and uses the inspection instructions corresponding to the stored UAV flight strategy as refined inspection instructions to generate refined inspection instructions. The refined inspection instructions refer to specific operation instructions that guide the UAV to perform specific actions during the inspection process, such as hovering and shooting at a certain point, changing the flight altitude, etc. According to the refined inspection instructions, the flight path of the UAV is designed using the route planning software.

[0076] Specifically, the UAV inspection module performs inspection control on the UAV according to the UAV route to obtain inspection data.

[0077] Specifically, the inspection data refers to the information collected by the drone during the inspection mission when the drone is inspected and controlled according to the drone route, including image data, sensor data, location information and equipment status data. The image data refers to photos and videos taken by the onboard high-definition camera or infrared camera. The sensor data refers to readings from drone sensors, such as temperature, humidity, air pressure, chemical composition, etc. The location information refers to information such as the drone's track, longitude and latitude, and flight altitude. The equipment status data refers to the status information of the drone itself, such as battery power, system health, flight time, etc.

[0078] Specifically, the CV analysis module selects YOLO-v5 as the basic algorithm model, trains the basic algorithm model to obtain a defect recognition algorithm, inputs the inspection data into the defect recognition algorithm, and obtains a defect recognition result output by the defect recognition algorithm corresponding to the inspection data.

[0079] Specifically, the YOLO-v5 refers to a deep learning algorithm for real-time target detection; the defect recognition result corresponding to the inspection data includes defect categories (such as cracks, dirt, occlusion, etc.), confidence scores, total number of defects and distribution of the number of defects in each category.

[0080] Specifically, when the CV analysis module trains the defect recognition algorithm, the xml file and the picture annotated by labelImg are used as defect recognition algorithm training data, and the defect recognition algorithm is trained according to the defect recognition training method, and the defect recognition training method includes:

[0081] Step S1, generating four files, train.txt, val.txt, test.txt and trainval.txt, according to the defect recognition algorithm training data, and naming them as training set, verification set and test set respectively;

[0082] Step S2, converting the defect recognition algorithm training data into yolo_txt format to obtain converted training data, that is, converting each xml file and picture in the defect recognition algorithm training data into txt format, each line of the txt file is information of a bbox target, and each line includes class, x_center, y_center, width, height;

[0083] Step S3, selecting the medium-sized architecture YOLO-v5m in YOLO-v5 as the basic algorithm model;

[0084] Step S4, training the basic algorithm model according to the training set in the converted training data to obtain a trained basic algorithm model, and verifying the trained basic algorithm model according to the verification set in the converted training data, outputting the trained basic algorithm model with an accuracy rate of 90% as a verified basic algorithm model, and retraining the trained basic algorithm model with an accuracy rate of less than 90%;

[0085] Step S5, testing the verified basic algorithm model according to the test set in the converted training data, outputting the verified basic algorithm model with an accuracy rate of 98% as the defect recognition algorithm, and retraining the verified basic algorithm model with an accuracy rate less than 90%.

[0086] Specifically, the labelImg refers to an image annotation tool used to generate image annotation data, and the annotated xml file refers to a file generated by LabelImg in a format that follows Pascal The information of all annotated objects in the VOC standard image, such as category, bounding box coordinates, etc.; the annotated image refers to the original image file corresponding to the XML file used for model training and testing; the information of a bbox target refers to the rectangular box used to describe the target position in the image; the class refers to the category label of the target, usually expressed as an integer; the x_center refers to the x coordinate of the center of the bounding box, normalized relative to the image width; the y_center refers to the y coordinate of the center of the bounding box, normalized relative to the image height; the width refers to the width of the bounding box, normalized relative to the image width; the height refers to the height of the bounding box, normalized relative to the image height; the medium-sized architecture YOLO-v5m refers to a variant of the YOLO-v5 model, "m" stands for medium, it is one of the networks of different sizes in YOLO-v5, compared with YOLO-v5s (small) and YOLO-v5l (large), YOLO-v5m is suitable for application scenarios with moderate requirements for computing resources in terms of model complexity and inference speed.

[0087] Specifically, when locating defects in the inspection data, the CV analysis module improves the accuracy of the GPS position to the pixel level. At the same time, combined with the UAV flight altitude, camera sensor size and lens size, the SFM technology is used to calculate the elevation and position information of different pixel points in the image, and the global photo component size is corrected to obtain an entire consistent ground resolution image. The entire consistent ground resolution image is input into the defect location algorithm to solve the accurate location information of the faulty component.

[0088] Specifically, the defect location algorithm refers to an algorithm used to analyze images to identify and locate defects. For example, if the defect location algorithm is set to a convolutional neural network model algorithm, the method for improving the accuracy of the GPS position to the pixel level in this embodiment is to use a high-precision GPS device combined with image processing technology to accurately match the original GPS coordinates with the pixel coordinates of the image; the pixel level refers to the smallest unit of the image, that is, a single pixel; the UAV flight altitude refers to the vertical distance between the UAV and the ground when flying; the camera sensor size refers to the physical size of the camera sensor, usually in millimeters; the lens size refers to the focal length of the lens; the SFM technology refers to a method of reconstructing 3D structures from 2D images of multiple perspectives. Technology, which calculates the three-dimensional structure of the scene and the camera motion by identifying and matching feature points in the image; when this embodiment calculates the elevation and position information of different pixel points in the image through SFM technology, the three-dimensional coordinates and elevation information of each feature point are inferred by matching the feature points in the images taken from different perspectives. Based on known ground control points, this embodiment uses high-precision measuring equipment to adjust the geometric distortion in the image to ensure the accuracy of the size, and corrects the size of the global photo component. When solving the accurate location information of the faulty component, this embodiment inputs the corrected image into the defect location algorithm, identifies the defect location, and converts the defect location into the actual geographic location by combining the precise geographic coordinates obtained by GPS and SFM.

[0089] Specifically, the CV analysis module automatically identifies defects based on thermal imaging and machine vision to obtain automatic defect identification results.

[0090] Specifically, the thermal imaging refers to the temperature distribution map of the object surface captured by a thermal imaging camera, the machine vision refers to the visible light image provided by the machine vision camera, and the CV analysis module fuses thermal imaging and machine vision at the decision level, and uses a deep learning model to automatically identify defects in the thermal imaging and machine vision after the decision-level fusion.

[0091] Specifically, the CV analysis module adds the automatic defect recognition result to the defect recognition result to optimize the defect recognition result.

[0092] Specifically, the CV analysis module automatically locates the defect according to the photogrammetry principle and the aerial triangulation calculation method, and optimizes the defect location result according to the automatic defect location result.

[0093] Specifically, the photogrammetry principle refers to the principle of obtaining geometric information of an object by analyzing and measuring images, including stereoscopic vision, scale and image analysis; the aerial triangulation method refers to the method of calculating the three-dimensional coordinates of a ground target from multiple images, including:

[0094] Step S10, selecting a ground control point with known coordinates;

[0095] Step S20, matching feature points of the same target in multiple images and calculating their parallax;

[0096] Step S30, calculating the three-dimensional coordinates of the ground target by a geometric algebra method according to the ground control points with known coordinates and the parallax;

[0097] In this embodiment, when the automatic defect location result is inconsistent with the defect location result, the defect location result is replaced by the automatic defect location result.

[0098] Specifically, the CV analysis module performs defect deduplication according to the thermal imaging fault recognition algorithm to obtain a defect deduplication result, and adjusts the optimization process of the defect location result according to the defect deduplication result.

[0099] Specifically, the thermal imaging fault recognition algorithm refers to an algorithm that uses infrared imaging technology to analyze the thermal image of the target object to identify faults. For example, the thermal imaging fault recognition algorithm can be set as a convolutional neural network model algorithm. This embodiment does not limit the specific method of defect deduplication by the thermal imaging fault recognition algorithm. Technical personnel in this field can freely set it according to actual conditions, and only need to meet the need for removing duplicate defects. For example, a clustering algorithm can also be set to merge adjacent or similar defects into a single object. In this embodiment, the defect location result is replaced by the defect deduplication result to adjust the optimization process of the defect location result.

[0100] Specifically, when performing fusion diagnosis, the fusion inspection and analysis module generates a fusion analysis report according to the diagnosis mode, wherein:

[0101] When the diagnostic mode is the IV diagnostic trigger mode, the fusion inspection analysis module uses the fault diagnosis result obtained by the IV analysis module as the main analysis report, and the defect identification result output by the CV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report;

[0102] When the diagnosis mode is the CV analysis unit trigger mode, the fusion inspection analysis module uses the defect recognition result output by the CV analysis module as the main analysis report, and the fault diagnosis result obtained by the IV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report;

[0103] The fusion inspection and analysis module outputs each fusion analysis report as a fusion diagnosis result of each diagnosis mode.

[0104] Specifically, the diagnostic mode refers to different working modes that trigger or direct the system to perform analysis and generate reports, including IV diagnostic trigger mode and CV analysis unit trigger mode.

[0105] Specifically, the fusion inspection analysis module inputs the fusion diagnosis data into the diagnosis effect analysis model, obtains the effective value of the fusion diagnosis result of each diagnosis mode output by the diagnosis effect analysis model, and selects the fusion diagnosis result of the diagnosis mode corresponding to the maximum effective value as the final fusion diagnosis result.

[0106] Specifically, the diagnostic effect analysis model refers to a model used to evaluate the fused diagnostic results under different diagnostic modes. This embodiment does not limit the construction method of the diagnostic effect analysis model. Technical personnel in this field can freely set it according to actual conditions, and only need to meet the output requirements of the effective values ​​of the fused diagnostic results of each diagnostic mode. For example, the diagnostic effect analysis model can be obtained by training a deep learning model with historical data; the effective value of the fused diagnostic results of each diagnostic mode refers to the value used to measure the reliability or accuracy of the fused diagnostic results generated by each diagnostic mode, and is output by the diagnostic effect analysis model.

[0107] Specifically, the visualization module visualizes the fault diagnosis result, the defect identification result, the defect location result and the final fusion diagnosis result.

[0108] Specifically, visual presentation refers to the process of converting complex data and analysis results into an easy-to-understand visual format to help users understand and analyze information more intuitively, such as using charts, dashboards, heat maps, fault location maps, time series diagrams, 3D models, and interactive visualizations for visual presentation.

[0109] Specifically, the reporting module generates a report according to the fault diagnosis result, the defect identification result, the defect location result and the fusion diagnosis result after the visual presentation.

[0110] Specifically, generating a report refers to the process of systematically organizing the results of analysis and diagnosis into documents for easy storage, sharing and further analysis. The report includes a summary of the data, analysis results, conclusions and suggestions. In this implementation, the report is generated through a large language model.

[0111] Specifically, after generating the report, the navigation and defect elimination module performs navigation and defect elimination according to the fault diagnosis result, the defect identification result, the defect location result and the fusion diagnosis result.

[0112] Specifically, the navigation and defect elimination refers to the process of helping users or automatic systems to effectively troubleshoot and repair defects through guiding steps or strategies after faults and defects in the system or equipment are detected. In this embodiment, navigation and defect elimination are performed through automated tools and algorithms, such as convolutional neural network models.

[0113] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A photovoltaic power station fault monitoring system based on a neural network model, characterized in that: include: Data acquisition module, used to collect string current data and voltage data; An IV analysis module, used to perform fault diagnosis according to the string current data and the voltage data to obtain a fault diagnosis result; The digital power station modeling module is used to construct the basic model of the power station, identify and number the coordinates of the photovoltaic strings, adjust the coordinate numbers of the photovoltaic strings, and supplement the blank numbers; Inspection route planning module, used to generate UAV flight strategies and plan UAV routes; The UAV inspection module is used to control the UAV inspection and obtain the inspection data; An inspection data uploading module, used to upload the inspection data to a CV analysis module; A CV analysis module is used to obtain defect recognition results, to obtain defect location results, to automatically recognize defects, to optimize the defect recognition results according to the defect automatic recognition results, to automatically locate defects, to optimize the defect location results according to the defect automatic location results, to perform defect deduplication, and to adjust the optimization process of the defect location results according to defect deduplication; The fusion inspection analysis module is used to output the fusion diagnosis results and select the final fusion diagnosis results; A visualization module for visualization; Report module, used to generate reports; The navigation defect elimination module is used for navigation defect elimination.

2. The photovoltaic power station fault monitoring system based on the neural network model according to claim 1 is characterized in that: The IV analysis module generates IV curve data according to the string current data and the voltage data, and sets the IV curve formula as follows: n is the diode ideal factor, Rs is the series resistance of the photovoltaic string, Rsh is the parallel resistance of the photovoltaic string, q is the single electron charge, k is the Boltzmann constant, T is the thermodynamic temperature, and the IV analysis module inputs the IV curve data into the fault diagnosis model, and obtains the output result of the fault diagnosis model as the fault diagnosis result.

3. The photovoltaic power station fault monitoring system based on the neural network model according to claim 1 is characterized in that: The digital power station modeling module performs aerial triangulation on the data collected by the drone aerial photographs and constructs a dense point cloud to generate an orthophoto and a digital surface model, and uses the orthophoto and the digital surface model as the basic model of the power station.

4. The photovoltaic power station fault monitoring system based on the neural network model according to claim 3 is characterized in that: The digital power station modeling module identifies and numbers the photovoltaic string coordinates according to the intelligent numbering algorithm and the sub-array KML of the station to obtain the photovoltaic string coordinates and the photovoltaic string coordinate numbers.

5. The photovoltaic power station fault monitoring system based on the neural network model according to claim 4 is characterized in that: The digital power station modeling module obtains the number of the photovoltaic string coordinates of the station building CAD drawing, takes it as the preset photovoltaic string coordinate number, and compares the preset photovoltaic string coordinate number with the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number according to the comparison result, wherein: When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is consistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is correct and does not adjust the photovoltaic string coordinate number; When the preset photovoltaic string coordinate number of the same photovoltaic string coordinate is inconsistent with the photovoltaic string coordinate number, the digital power station modeling module determines that the number is wrong, adjusts the photovoltaic string coordinate number, and adjusts the photovoltaic string coordinate number of the same photovoltaic string coordinate to the preset photovoltaic string coordinate number.

6. The photovoltaic power station fault monitoring system based on the neural network model according to claim 1 is characterized in that: The inspection route planning module inputs the power station basic model, the photovoltaic string coordinates and the photovoltaic string coordinate numbers into the management system to generate a UAV flight strategy, obtains the UAV flight strategy output by the management system, generates a refined inspection instruction according to the UAV flight strategy, and plans the UAV route according to the refined inspection instruction; The UAV inspection module performs inspection control on the UAV according to the UAV route to obtain inspection data.

7. The photovoltaic power station fault monitoring system based on the neural network model according to claim 1 is characterized in that: The CV analysis module selects YOLO-v5 as the basic algorithm model, trains the basic algorithm model to obtain a defect recognition algorithm, inputs the inspection data into the defect recognition algorithm, and obtains a defect recognition result output by the defect recognition algorithm that corresponds to the inspection data.

8. The photovoltaic power station fault monitoring system based on the neural network model according to claim 7 is characterized in that: When training the defect recognition algorithm, the CV analysis module uses the XML file and the picture annotated by label Img as the defect recognition algorithm training data, and trains the defect recognition algorithm according to the defect recognition training method, which includes: Step S1, generating four files, train.txt, val.txt, test.txt and trainval.txt, according to the defect recognition algorithm training data, and naming them as training set, verification set and test set respectively; Step S2, converting the defect recognition algorithm training data into yolo_txt format to obtain converted training data, that is, converting each xml file and picture in the defect recognition algorithm training data into txt format, each line of the txt file is information of a bbox target, and each line includes class, x_center, y_center, width, height; Step S3, selecting the medium-sized architecture YOLO-v5m in YOLO-v5 as the basic algorithm model; Step S4, training the basic algorithm model according to the training set in the converted training data to obtain a trained basic algorithm model, and verifying the trained basic algorithm model according to the verification set in the converted training data, outputting the trained basic algorithm model with an accuracy rate of 90% as a verified basic algorithm model, and retraining the trained basic algorithm model with an accuracy rate of less than 90%; Step S5, testing the verified basic algorithm model according to the test set in the converted training data, outputting the verified basic algorithm model with an accuracy rate of 98% as the defect recognition algorithm, and retraining the verified basic algorithm model with an accuracy rate less than 90%.

9. The photovoltaic power station fault monitoring system based on the neural network model according to claim 8 is characterized in that: When the CV analysis module locates defects in the inspection data, the accuracy of the GPS position is improved to the pixel level. At the same time, the elevation and position information of different pixel points in the image are calculated through the SFM technology in combination with the flight altitude of the drone, the camera sensor size and the lens size, and the size of the global photo component is corrected to obtain the entire consistent ground resolution image. The entire consistent ground resolution image is input into the defect location algorithm to solve the accurate position information of the faulty component; The CV analysis module automatically identifies defects based on thermal imaging and machine vision to obtain automatic defect identification results; The CV analysis module adds the automatic defect recognition result to the defect recognition result to optimize the defect recognition result; The CV analysis module automatically locates the defect according to the photogrammetry principle and the aerial triangulation calculation method, and optimizes the defect location result according to the defect automatic location result; The CV analysis module performs defect deduplication according to the thermal imaging fault recognition algorithm to obtain a defect deduplication result, and adjusts the optimization process of the defect location result according to the defect deduplication result.

10. The photovoltaic power station fault monitoring system based on the neural network model according to claim 9, characterized in that: When performing fusion diagnosis, the fusion inspection and analysis module generates a fusion analysis report according to the diagnosis mode, wherein: When the diagnostic mode is the IV diagnostic trigger mode, the fusion inspection analysis module uses the fault diagnosis result obtained by the IV analysis module as the main analysis report, and the defect identification result output by the CV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report; When the diagnosis mode is the CV analysis unit trigger mode, the fusion inspection analysis module uses the defect recognition result output by the CV analysis module as the main analysis report, and the fault diagnosis result obtained by the IV analysis module as the secondary analysis report, and combines the main analysis report and the secondary analysis report to generate a fusion analysis report; The fusion inspection and analysis module outputs each fusion analysis report as a fusion diagnosis result of each diagnosis mode; The fusion inspection analysis module inputs the fusion diagnosis data into the diagnosis effect analysis model, obtains the effective value of the fusion diagnosis result of each diagnosis mode output by the diagnosis effect analysis model, and selects the fusion diagnosis result of the diagnosis mode corresponding to the maximum effective value as the final fusion diagnosis result.

Images