CN-121685543-B - Power equipment defect identification method and system based on machine vision

CN121685543BCN 121685543 BCN121685543 BCN 121685543BCN-121685543-B

Abstract

The application provides a machine vision-based power equipment defect identification method and system, which relate to the technical field of equipment defect identification and comprise the steps of acquiring and preprocessing standard electroluminescence images, standard infrared thermal images and standard visible light images of a target photovoltaic group string under different time windows through imaging equipment carried by an unmanned aerial vehicle, and recording environmental parameters acquired each time; the method comprises the steps of analyzing visual defect characteristics and attenuation gradient characteristics in a standard electroluminescent image, carrying out fusion verification by combining a standard infrared thermal image, a standard visible light image and electric performance data to obtain potential induced attenuation confirmation judgment, evaluating the evolution rate of potential induced attenuation obtained by using a time sequence analysis method based on environmental parameters of a photovoltaic module judged to have potential induced attenuation, mapping to generate potential induced attenuation risk levels, and generating a structured defect identification report. The accuracy of the defect identification of the power equipment is improved.

Inventors

Xu Doukui
LIANG BAOHONG
Xu Douwei
LONG HUA

Assignees

陕西丝路创城能源科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (8)

1. The power equipment defect identification method based on machine vision is characterized by comprising the following steps of: Acquiring electroluminescent images, infrared thermal images and visible light images of a target photovoltaic group string under different time windows through imaging equipment carried by an unmanned aerial vehicle, preprocessing the acquired images to obtain standard electroluminescent images, standard infrared thermal images and standard visible light images, and recording environmental parameters acquired each time; analyzing visual defect characteristics and attenuation gradient characteristics in the standard electroluminescent image, and combining the standard infrared thermal image, the standard visible light image and the electrical performance data for fusion verification to obtain potential induced attenuation diagnosis judgment; The confirmed diagnosis is judged to be provided with the potential induced attenuation photovoltaic module, a time sequence analysis method is used for evaluating and obtaining the evolution rate of the potential induced attenuation based on the environmental parameter, and the risk level of the potential induced attenuation is generated according to the evolution rate map; generating a structured defect identification report based on the definitive diagnosis of potential induced decay and the risk level of potential induced decay; The method comprises the steps of analyzing visual defect characteristics and attenuation gradient characteristics in the standard electroluminescent image, and combining the standard infrared thermal image, the standard visible light image and the electrical performance data for fusion verification to obtain potential induced attenuation diagnosis judgment, and comprises the following steps: respectively inputting standard electroluminescence images of each photovoltaic module in the target photovoltaic group string into a pre-trained edge blackening defect identification model, and calculating the area occupation ratio of an edge blackening defect area in the corresponding electroluminescence image to be used as a visual defect characteristic quantization value of the photovoltaic module; Arranging visual defect characteristic quantized values of all photovoltaic modules in the target photovoltaic group string according to an electrical connection sequence to form a visual defect characteristic quantized value sequence; Calculating a spearman grade correlation coefficient of the visual defect characteristic quantized value sequence as a gradient consistency quantized score; When the gradient consistency quantization score is lower than a preset negative threshold, determining that attenuation gradient characteristics exist, and generating a gradient verification effective signal; Setting a multisource consistency condition, verifying an effective signal and the multisource consistency condition based on the gradient, performing fusion decision, and outputting potential induced attenuation diagnosis determination when the decision passes; Setting a multi-source consistency condition, the multi-source consistency condition comprising: the standard infrared thermal image analysis result shows that at least one photovoltaic module generates abnormal heat; The second condition is that the analysis result of the standard visible light image shows that the number of the photovoltaic modules with appearance defect characteristics in the target photovoltaic group string is lower than a preset number threshold; the electrical performance data show that the insulation resistance value of the target photovoltaic group string is lower than a preset safety threshold value, or the output power attenuation rate of the photovoltaic modules in the target photovoltaic group string exceeds a preset normal range; When the gradient verification valid signal is valid and the number of the conditions which are met in the multi-source consistency conditions is not less than two, judging that multi-source evidence points to be consistent; and outputting a definite diagnosis judgment of the potential induced attenuation when judging that the multi-source evidence points consistently.
2. The machine vision-based power equipment defect identification method according to claim 1, wherein the steps of collecting electroluminescent images, infrared thermal images and visible light images of the target photovoltaic strings in different time windows by using an imaging device carried by an unmanned plane, preprocessing the collected images to obtain standard electroluminescent images, standard infrared thermal images and standard visible light images, and recording environmental parameters collected each time, comprise: Under the condition of all the night, driving the unmanned aerial vehicle to carry an electroluminescent imager to collect electroluminescent images of the target photovoltaic group string; Under the stable solar irradiation condition in sunny days, driving the unmanned aerial vehicle to carry an infrared thermal imager and a high-resolution visible light camera, and synchronously collecting an infrared thermal image and a visible light image of the target photovoltaic group string; Recording environmental parameters when the electroluminescent image, the infrared thermal image and the visible light image are acquired each time, wherein the environmental parameters at least comprise environmental temperature and environmental humidity; And respectively carrying out image registration processing, noise elimination processing and illumination normalization processing on the electroluminescence image, the infrared thermal image and the visible light image acquired in the past to obtain a standard electroluminescence image, a standard infrared thermal image and a standard visible light image.
3. The machine vision-based power equipment defect identification method according to claim 2, wherein performing image registration processing, noise elimination processing, and illumination normalization processing on the electroluminescence image, the infrared thermal image, and the visible light image acquired in the past, respectively, to obtain a standard electroluminescence image, a standard infrared thermal image, and a standard visible light image, comprises: extracting characteristic points in the same type of images acquired in different periods for the same photovoltaic module in the target photovoltaic group string, matching the characteristic points, and aligning the images acquired in different periods to the same coordinate system through space transformation to obtain an electroluminescence image after registration, an infrared thermal image after registration and a visible light image after registration; The registered electroluminescent image, the registered infrared thermal image and the registered visible light image are subjected to noise reduction processing by a spatial domain filtering algorithm respectively to obtain a denoised electroluminescent image, a denoised infrared thermal image and a denoised visible light image; The contrast and brightness of the denoised visible light image and the denoised infrared thermal image are respectively adjusted based on the global gray level distribution of the image, so that a normalized visible light image and a normalized infrared thermal image are obtained; Respectively defining the normalized visible light image, the normalized infrared thermal image and the denoised electroluminescent image as a standard electroluminescent image, a standard infrared thermal image and a standard visible light image, and carrying out association storage with a corresponding photovoltaic module identifier, a corresponding acquisition time stamp and a corresponding image type label; And traversing the target photovoltaic group string to obtain a standard electroluminescent image, a standard infrared thermal image and a standard visible light image of each photovoltaic module.
4. The machine vision-based power equipment defect identification method of claim 1, wherein a fusion decision is made based on the gradient verification valid signal and the multisource consistency condition, and a potential induced decay determination is output when the decision passes, comprising: Respectively inputting the standard infrared thermal images of each photovoltaic module in the target photovoltaic group string into a pre-trained infrared thermal image defect identification model, and outputting to obtain a two-classification judgment result of whether abnormal heating exists in each photovoltaic module, wherein the two-classification judgment result is used as a standard infrared thermal image analysis result; Respectively inputting standard visible light images of each photovoltaic module in a target photovoltaic group string into a pre-trained visible light image defect recognition model, and outputting to obtain a classification judgment result of whether each photovoltaic module has appearance defect characteristics or not, wherein the appearance defect characteristics at least comprise visible damage, stains and shielding, and the classification judgment result is used as a standard visible light image analysis result; acquiring electrical performance data of the target photovoltaic group string, wherein the electrical performance data at least comprises an insulation resistance value of the photovoltaic group string and an output power attenuation rate of a photovoltaic module; And judging whether the multi-source consistency condition is met or not based on the standard infrared thermal image analysis result, the standard visible light image analysis result and the electrical performance data.
5. The machine vision-based power equipment defect identification method of claim 1, wherein evaluating the evolution rate of potential induced attenuation based on the environmental parameter using a time-series analysis method for the photovoltaic module determined to have potential induced attenuation, and generating a risk level of potential induced attenuation according to the evolution rate map comprises: aiming at the photovoltaic module with potential induced attenuation determined by the diagnosis, acquiring a visual defect characteristic quantification value in the conventional inspection, acquiring corresponding environment parameters, and correlating the two parameters according to time sequence to form a time sequence input sequence; calculating to obtain an environmental stress index based on environmental parameters recorded during the previous inspection; pre-constructing an evolution rate evaluation array comprising N evolution rate estimators; calculating and rounding the product of the environmental stress index and N to obtain an evaluation scale J, and randomly selecting J evolution rate estimators from the evolution rate estimation array, wherein J is more than or equal to 1 and less than or equal to N; inputting the time sequence input sequence into the J evolution rate estimators in parallel, outputting to obtain J preliminary evolution rate estimated values, and carrying out integrated calculation on the J preliminary evolution rate estimated values to obtain the evolution rate of potential induced attenuation; And comparing the evolution rate with a plurality of preset risk threshold intervals, and mapping to generate corresponding risk levels of potential induced attenuation.
6. The machine vision-based power equipment defect identification method of claim 5, wherein the calculating the environmental stress index based on the environmental parameters recorded during the past inspection comprises: obtaining an upper limit of a design environment parameter of the photovoltaic module, wherein the design environment parameter comprises an upper limit of a design temperature and an upper limit of a design humidity; based on the environmental parameters recorded during the previous inspection and the upper limit of the design environmental parameters, respectively calculating and obtaining a temperature deviation factor and a humidity deviation factor of the single inspection; acquiring a time interval between two adjacent patrol inspection processes, carrying out weighted summation on a temperature deviation factor and a humidity deviation factor of a single patrol inspection process based on the time interval to obtain an environmental stress amount in an adjacent patrol inspection period, and accumulating the environmental stress amount in each period to obtain an accumulated environmental stress amount; Dividing the accumulated environmental stress amount by the total time span value covered by the previous inspection to obtain an environmental stress index.
7. The machine vision-based power equipment defect identification method of claim 6, wherein constructing an evolution rate assessment array comprising N evolution rate assessors comprises: Acquiring complete visual defect characteristic quantized value time sequence data and corresponding environment parameter time sequence data of a plurality of photovoltaic modules from a historical operation and maintenance database, and carrying out one-to-one correspondence association according to module dimensions to form an original training sequence set; labeling a corresponding potential induced attenuation real evolution rate for each sample sequence in the original training sequence set to form a labeled training sequence set; Adopting a self-help sampling method, randomly extracting a plurality of sample sequences from the original training sequence set and the corresponding marked training sequence set in a put-back way, and generating N independent training subsets; N independent training subsets are used for training to obtain N evolution rate estimators respectively, and the N evolution rate estimators are integrated to form an evolution rate estimation array.
8. A machine vision-based power equipment defect identification system for performing the machine vision-based power equipment defect identification method of any of claims 1-7, comprising: The data acquisition module is used for acquiring electroluminescent images, infrared thermal images and visible light images of the target photovoltaic group strings under different time windows through imaging equipment carried by the unmanned aerial vehicle, preprocessing the acquired images to obtain standard electroluminescent images, standard infrared thermal images and standard visible light images, and recording environmental parameters acquired each time; the diagnosis confirming judging module is used for analyzing the visual defect characteristics and the attenuation gradient characteristics in the standard electroluminescent image, and combining the standard infrared thermal image, the standard visible light image and the electrical performance data for fusion verification to obtain diagnosis confirming judgment of potential induced attenuation; The risk level judging module is used for judging that the potential induced attenuation exists in the photovoltaic module through diagnosis, based on the environmental parameters, evaluating the evolution rate of the potential induced attenuation by using a time sequence analysis method, and mapping and generating a risk level of the potential induced attenuation according to the evolution rate; and the fusion output module is used for generating a structured defect identification report based on the diagnosis judgment of the potential induced attenuation and the risk level of the potential induced attenuation.

Description

Power equipment defect identification method and system based on machine vision Technical Field The invention relates to the technical field of equipment defect identification, in particular to a machine vision-based power equipment defect identification method and system. Background In the running process of a large photovoltaic system, hundreds of photovoltaic modules are connected in series to generate direct-current voltage up to 1500V, the high voltage can drive metal ions in the packaging material of the photovoltaic modules to directionally migrate, the migrated metal ions gather on the surface of a battery piece and then damage the passivation layer of the battery piece, and finally irreversible attenuation of the power generation performance of the photovoltaic modules is caused, namely potential induced attenuation. However, the traditional potential induced attenuation identification mainly relies on manual inspection and off-line detection, and has the problems of low detection efficiency, strong subjectivity, high risk of misjudgment due to missed judgment, difficulty in adapting to the normalized operation and maintenance requirements of large-scale electric power facilities and the like. Disclosure of Invention Aiming at the technical problem of insufficient accuracy in identifying potential induced attenuation defects in the prior art, the invention provides a machine vision-based power equipment defect identification method and system. The technical scheme includes that according to the first aspect, the invention provides a power equipment defect identification method based on machine vision, which comprises the steps of collecting electroluminescent images, infrared thermal images and visible light images of a target photovoltaic group string under different time windows through imaging equipment carried by an unmanned aerial vehicle, preprocessing the collected images to obtain standard electroluminescent images, standard infrared thermal images and standard visible light images, and recording environmental parameters collected each time. And analyzing visual defect characteristics and attenuation gradient characteristics in the standard electroluminescent image, and combining the standard infrared thermal image, the standard visible light image and the electrical performance data for fusion verification to obtain potential induced attenuation diagnosis judgment. And evaluating the evolution rate of the potential induced attenuation by using a time sequence analysis method based on the environment parameter and generating a risk grade of the potential induced attenuation according to the evolution rate map. Based on the determined diagnosis of potential induced decay, the risk level of potential induced decay, a structured defect identification report is generated. The invention provides a machine vision-based power equipment defect identification system, which comprises a data acquisition module, a machine vision-based power equipment defect identification module and a machine vision-based power equipment defect identification module, wherein the data acquisition module is used for acquiring electroluminescent images, infrared thermal images and visible light images of a target photovoltaic group string under different time windows through imaging equipment carried by an unmanned aerial vehicle, preprocessing the acquired images to obtain standard electroluminescent images, standard infrared thermal images and standard visible light images, and recording environmental parameters acquired each time. And the diagnosis confirming and judging module is used for analyzing the visual defect characteristics and the attenuation gradient characteristics in the standard electroluminescent image, and combining the standard infrared thermal image, the standard visible light image and the electrical performance data for fusion verification to obtain the diagnosis confirming and judging of the potential induced attenuation. And the risk level judging module is used for evaluating the evolution rate of the potential induced attenuation based on the environment parameters and the photovoltaic module which is judged to have the potential induced attenuation by the definite diagnosis, and generating the risk level of the potential induced attenuation according to the evolution rate map. And the fusion output module is used for generating a structured defect identification report based on the diagnosis judgment of the potential induced attenuation and the risk level of the potential induced attenuation. Compared with the prior art, the method has the advantages that the unmanned aerial vehicle-mounted imaging equipment collects the electroluminescent image, the infrared thermal image and the visible light image of the photovoltaic string under a plurality of time windows and completes standardized preprocessing, and the synchronously recorded environmental parameters and electrical performance data are co