Intelligent Analysis System and Method for Photovoltaic Power Plant Operation Data Based on Machine Vision

Abstract

This invention discloses a machine vision-based intelligent analysis system and method for photovoltaic power plant operation data, belonging to the field of big data analysis technology. The invention constructs a standardized dataset by collecting and preprocessing historical aging characteristics, historical operation data, and real-time operation data; extracting core features to establish a component-specific aging characteristic profile and a normal data fluctuation baseline; correcting the profile and baseline; generating a list of suspected fault data by comparing real-time operation data; distinguishing between normal aging fluctuations and actual faults through cross-validation with historical data from the same period of the component and reference data from similar components; and dynamically updating the profile and judgment threshold by collecting the verification results. The system includes modules for data acquisition and processing, feature profile construction, fault initial screening generation, cross-validation judgment, and profile threshold updating. This solution can reduce unnecessary downtime for maintenance, lower operation and maintenance costs, and provide technical support for the stable operation of photovoltaic power plants throughout their entire lifecycle.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a machine vision-based intelligent analysis system and method for photovoltaic power plant operation data. Background Technology

[0002] During long-term operation, photovoltaic (PV) power plants gradually undergo aging and degradation. This aging process causes core operating parameters such as module output power, operating current, and output voltage to exhibit continuous and degrading stable fluctuations. Current intelligent data analysis systems for PV power plants based on machine vision technology often fail to accurately distinguish between aging and degradation and fault damage. This can easily lead to confusion between normal performance degradation fluctuations and functional damage to modules, triggering unnecessary shutdowns or ignoring module damage, significantly increasing power plant operation and maintenance costs or preventing the PV power plant from operating at full power. Therefore, there is an urgent need for an intelligent data analysis solution for PV power plants that can adapt to the aging characteristics of modules and avoid misinterpreting normal data fluctuations. Summary of the Invention

[0003] The purpose of this invention is to provide a machine vision-based intelligent analysis system and method for photovoltaic power plant operation data to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent analysis method for photovoltaic power plant operation data based on machine vision, the intelligent analysis method comprising the following steps:

[0005] Collect historical aging characteristic data, historical operation data, and real-time operation data of photovoltaic power station components, preprocess the collected data, and construct a standardized dataset;

[0006] Based on the standardized dataset, extract the aging characteristic parameters of the components, construct the exclusive aging characteristic profile of each photovoltaic module, and draw the baseline of normal data fluctuation.

[0007] By combining real-time environmental data to correct the aging feature profile and the normal data fluctuation baseline, and by comparing real-time operation data with the corrected baseline, a list of suspected fault data is constructed and marked as a prediction result.

[0008] Cross-validate the list of suspected fault data to distinguish between normal aging fluctuations and actual faults, and output the validation results, which are labeled as secondary prediction results.

[0009] Collect the results of two predictions and manual inspections, and dynamically update the component aging feature profile and judgment threshold.

[0010] Furthermore, based on the standardized dataset, component aging feature parameters are extracted by classification. These aging feature parameters include appearance characterization and performance operation categories. The appearance characterization category is obtained through visual images. Redundant parameters are removed using feature filtering.

[0011] The feature filtering first extracts features according to appearance and performance, and then removes redundant parameters that are weakly correlated with aging and cannot distinguish different aging stages through logical filtering. It focuses on core features that are sensitive to aging changes and outputs appearance features and performance features.

[0012] The appearance characteristics are extracted through machine vision image analysis, including the percentage of panel scratch area, the percentage of pixels in yellowed areas, the degree of frame corrosion, the component tilt angle deviation, and the integrity of the junction box seal.

[0013] The performance operation features are extracted from the operating parameters, including output power attenuation rate, voltage stability, current fluctuation value, start-stop response time, and energy loss rate.

[0014] Furthermore, the process of constructing a unique aging feature profile for each photovoltaic module is as follows: using the unique identifier of the photovoltaic module as an index, the selected aging feature parameters are categorized according to the logic of "initial state - operating stage - aging trend" to establish a unique aging feature profile.

[0015] By combining the time dimension information in the historical aging feature data of components, the evolution pattern of each aging feature parameter with the runtime is marked, forming a unique aging feature profile that includes parameter baseline value, evolution rate, and stage threshold.

[0016] Based on the specific aging characteristic profile, the normal fluctuation range of each aging characteristic parameter in different operating stages is statistically analyzed, and abnormal interference data is excluded; parameter fluctuation baselines are drawn according to appearance characterization and performance operation categories, and the normal data fluctuation baseline is formed by combining the change pattern of parameter fluctuation baselines in the time dimension.

[0017] The normal data fluctuation baseline is plotted by statistically analyzing the historical valid data of this feature within a certain operating phase, calculating the mean and standard deviation. The appearance characteristic baseline is presented as a "characteristic value - operating years" curve, and the performance operation baseline is presented as a "characteristic value - operating time" curve.

[0018] The construction of a dedicated aging feature profile uses the component's unique identifier as an index, categorizes core features along a timeline of "initial state - operational stage - aging trend," and marks the evolution logic of each feature over time. This evolution logic includes decay, stabilization, and abrupt change, forming a dedicated aging feature profile that includes baseline attributes, evolution logic, and stage boundaries. The generation of a normal fluctuation baseline is based on the performance of each feature in the dedicated aging feature profile at different operational stages. The natural fluctuation range after excluding abnormal interference is statistically analyzed, and baselines are drawn separately for appearance representation and performance operation, incorporating the changing patterns over time to ensure that the normal data fluctuation baseline is stage-adaptable.

[0019] Furthermore, real-time environmental data related to component operation is captured in real time by environmental sensors deployed in photovoltaic power plants, including light intensity, ambient temperature, air humidity, and wind speed. Based on the preset correlation logic between environmental data and component aging characteristic parameters, the collected real-time environmental data is substituted into the component's exclusive aging characteristic profile, the range of aging characteristic parameters at each stage in the profile is adjusted, and the baseline of normal data fluctuation is corrected according to the influence of environmental factors on the component's operating status.

[0020] Furthermore, according to the classification of appearance characteristics and performance operation, the real-time operation data is compared with the same type of parameters corresponding to the corrected normal data fluctuation baseline one by one. The values ​​of the real-time operation data in each parameter dimension are checked to see if they are within the normal fluctuation range of the corrected baseline. If a real-time operation data exceeds the range, the corresponding real-time operation data is marked as suspected fault data, and the specific parameters and deviation of the data exceeding the range are recorded to form a list of suspected fault data, which is marked as a prediction result.

[0021] The comparison process is as follows:

[0022] Step 1: Classify real-time operation data according to appearance and performance, ensuring consistency with the classification of the corrected normal data fluctuation baseline;

[0023] Step 2: For each aging characteristic parameter in each category, compare them one by one. For example, in the appearance characterization category, using the unique identifier of the photovoltaic module as an index, classify and collect the two prediction results to form a prediction result data set divided by module. Based on the collected prediction result data set and the manual inspection result "yellow area pixel ratio", using the unique identifier of the photovoltaic module as an index, classify and collect the two prediction results to form a prediction result data set divided by module. Based on the collected prediction result data set and the manual inspection result real-time value, compare it with the corrected baseline range of the parameter. In the performance operation category, using the unique identifier of the photovoltaic module as an index, classify and collect the two prediction results to form a prediction result data set divided by module. Based on the collected prediction result data set and the manual inspection result "output power", using the unique identifier of the photovoltaic module as an index, classify and collect the two prediction results to form a prediction result data set divided by module. Based on the collected prediction result data set and the manual inspection result real-time value, compare it with the corrected baseline range of the output power.

[0024] Step 3: Check whether the values ​​of real-time operation data in each parameter dimension are within the normal fluctuation range of the corrected baseline, and complete the comparison of all core characteristic parameters one by one.

[0025] If a real-time operation data point exceeds the normal fluctuation range of the corrected baseline, it is determined to be abnormal data, marked as suspected fault data, and the deviation range is recorded.

[0026] The list of suspected fault data is in tabular form and includes: component number, data collection timestamp, aging characteristic parameters, abnormal parameter name, real-time parameter value, corrected baseline range, and deviation magnitude.

[0027] Furthermore, the component-specific aging feature profile corresponding to the suspected fault data, the corrected normal data fluctuation baseline, and the historical appearance and performance data of the component under the same environmental conditions are retrieved. At the same time, the corresponding parameter data of the same type of component at the same aging stage and under the same environment are extracted. The suspected fault data list is compared with the two types of reference data in all dimensions, and the differences and overlaps with the two types of reference data are calculated. Combining the suspected fault data list with the evolution law of aging feature parameters at each stage in the component aging feature profile and the synergistic change relationship between parameters, the parameter changes of the suspected fault data are analyzed to see whether they conform to the natural evolution trend and parameter synergistic logic of the current aging stage of the component, and the analysis results list is output.

[0028] The analysis of suspected fault data involves determining whether the parameter changes in the suspected data conform to the inherent laws of the current aging stage of the component, based on the natural evolution trend of each feature parameter in the component's unique aging feature profile and the synergistic change relationship between the parameters.

[0029] Furthermore, a three-dimensional judgment standard is established to qualitatively classify suspected fault data. At the same time, the difference parameters, matching degree indexes, and collaborative logic involved in the qualitative classification are recorded to form a verification result that includes the judgment basis, the correlation of key parameters, and the weight of the judgment dimension, which is then labeled as a secondary prediction result.

[0030] The three-dimensional judgment criteria are the results of evolution law verification and parameter coordination logic. The structured judgment criteria are established to classify suspected data into "normal fluctuations of aging" or "real faults" and record the judgment basis and key related information.

[0031] The verification results include scores for each dimension, key difference parameters, matching degree indicators, and collaborative logic analysis, forming a structured report to provide a basis for operation and maintenance decisions.

[0032] Furthermore, using the unique identifier of the photovoltaic module as an index, the results of the two predictions are classified and aggregated to form a prediction result data set divided by module. Based on the aggregated prediction result data set and manual inspection results, the aging characteristic parameters of the module in the current operating stage and the evolution relationship between parameters are supplemented. The parameter logic in the original dedicated aging characteristic profile that does not match the actual aging process is updated, and the newly formed parameter change patterns and relationships are incorporated into the profile framework. Combined with the updated dedicated aging characteristic profile of the module, the parameter boundary characteristics between normal aging fluctuations and real faults in the verification results are analyzed and verified, and the normal fluctuation judgment criteria of appearance characterization parameters and performance operation parameters at different stages are adjusted.

[0033] The parameter boundary feature extraction method is as follows: based on the verification results and manual inspection results within a preset period, the value range of each parameter is statistically analyzed according to normal aging fluctuations and real faults.

[0034] Based on the parameter boundary characteristics, the maximum value of the normal aging fluctuation is taken as the upper limit of the fluctuation range, and the minimum value is taken as the lower limit; if the fluctuation range of the original normal data fluctuation baseline exceeds the boundary, it is adjusted to be consistent with the boundary.

[0035] The verification results are collected using the unique identifier of the component as an index, and the judgment results of each component are classified and integrated. New aging characteristics and new relationships between parameters in the current running stage are extracted from them to provide a basis for profile updates.

[0036] The exclusive aging feature profile is used to correct the parameter evolution logic in the original profile that does not match the actual aging process. At the same time, newly discovered parameter change patterns and correlations are incorporated into the profile framework to ensure that the profile continuously matches the actual aging state of the component.

[0037] The threshold adjustment extracts the parameter boundary between normal aging fluctuations and actual faults from the verification results. Combined with the updated profile, it adjusts the judgment criteria for normal parameter fluctuations in different operating stages and different feature categories to improve the accuracy of subsequent analysis.

[0038] Furthermore, the standardized dataset includes historical aging feature data, historical operation data, and real-time operation data, which are integrated according to a unified data structure. The historical aging feature data includes appearance images and operation parameter records of components at different operating stages; the historical operation data includes the component's past running time, start-stop frequency, and load changes; and the real-time operation data includes visual images of the component's current operating status and synchronously generated operation parameter data.

[0039] The historical aging characteristic data is retrieved from the photovoltaic power plant operation and maintenance archives and equipment management system. It includes appearance images of each operating stage after the components are put into operation, inverter output power, open circuit voltage, short circuit current, fill factor and other operating parameter records. The appearance images are obtained through past manual inspections and historical archives of fixed cameras.

[0040] The historical operation data is extracted from the power plant's SCADA and operation and maintenance logs; the real-time operation data is obtained by collecting visual images of the current operating status of the components through high-definition machine vision cameras deployed in the photovoltaic array area, and real-time operating parameters are collected simultaneously through the components' built-in sensors and monitoring sensors deployed in the array area.

[0041] All data underwent data cleaning and format standardization to ensure clarity. The data was then categorized and aggregated using the unique identifier of the photovoltaic module as an index, forming a dataset of prediction results by module. Based on this aggregated dataset and the results of manual inspections, deviations from the mean were eliminated. The results are categorized and aggregated to form a prediction result data set divided by component. Based on the aggregated prediction result data set and the data beyond the standard deviation of the manual inspection results, for the missing operating parameters of a certain component at a certain time period, the two prediction results are categorized and aggregated with reference to the unique identifier of the photovoltaic component as the index, forming a prediction result data set divided by component. Based on the aggregated prediction result data set and the parameter change trend of the manual inspection results at similar time periods, combined with the average parameter values ​​of the same type of component under the same environmental conditions, linear interpolation is used to complete the data.

[0042] The unified format includes converting visual images into a standardized pixel format, uniformly retaining operating parameters indexed by the unique identifier of the photovoltaic module, classifying and aggregating the two prediction results to form a prediction result data set divided by module, and using the unique identifier of the photovoltaic module as an index to classify and aggregate the two prediction results to form a prediction result data set divided by module, based on the aggregated prediction result data set and the manual inspection result 2, with decimal places, and the timestamp uniformly using the unique identifier of the photovoltaic module as an index to classify and aggregate the two prediction results to form a prediction result data set divided by module, based on the aggregated prediction result data set and the manual inspection result UTC, with the unique identifier of the photovoltaic module as an index to classify and aggregate the two prediction results to form a prediction result data set divided by module, based on the aggregated prediction result data set and the manual inspection result UTC, with the unique identifier of the photovoltaic module as an index to classify and aggregate the two prediction results to form a prediction result data set divided by module, based on the aggregated prediction result data set and the manual inspection result format.

[0043] Furthermore, the intelligent analysis system includes a data acquisition and processing module, a feature profile construction module, a fault initial screening generation module, a cross-validation judgment module, and a profile threshold update module;

[0044] The data acquisition and processing module is used to collect historical aging characteristic data, historical operation data and real-time operation data of photovoltaic power station components, and after completing the missing data items, integrate them according to a unified structure to construct a standardized dataset.

[0045] The feature profile construction module is used to extract component aging strongly correlated feature parameters from the standardized dataset, and after filtering out redundant parameters, construct an exclusive aging feature profile with the component's unique identifier as the index, and draw a normal data fluctuation baseline according to the category.

[0046] The fault screening generation module is used to capture real-time environmental data through environmental sensors, correct the aging characteristic profile and fluctuation baseline, compare real-time operation data with the corrected baseline, mark abnormal data and record relevant information to generate a list of suspected fault data.

[0047] The cross-validation judgment module is used to retrieve multi-dimensional reference data and compare it with suspected fault data, analyze the matching degree between parameter changes and component aging trends, qualitatively classify data types based on three-dimensional judgment criteria, and output structured verification results.

[0048] The profile threshold update module is used to collect verification results with the component's unique identifier, supplement and improve the component's aging feature parameters and related logic, update the exclusive aging feature profile, and adjust the normal fluctuation judgment threshold of parameters at different stages.

[0049] The output of the data acquisition and processing module is connected to the input of the feature profile construction module; the output of the feature profile construction module is connected to the input of the fault initial screening generation module; the output of the fault initial screening generation module is connected to the input of the cross-validation judgment module; and the output of the cross-validation judgment module is connected to the input of the profile threshold update module.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. This invention effectively avoids misjudging normal aging decay fluctuations as faults by constructing a unique aging feature profile and dynamically correcting the normal data fluctuation baseline, thereby reducing unnecessary downtime for maintenance, lowering operation and maintenance costs, and increasing the trust of operation and maintenance personnel in the system's early warning.

[0052] 2. This invention integrates historical data from the same period of the component, reference data of the same type of component, and parameter coordination logic through multi-dimensional data cross-validation and three-dimensional judgment criteria. It distinguishes between normal aging fluctuations and real faults, and outputs verification results with judgment criteria to provide guidance for operation and maintenance and ensure the stable operation of photovoltaic power plants.

[0053] 3. This invention continuously optimizes the judgment threshold by dynamically supplementing the component aging characteristic parameters and evolution correlation, so that the analysis system always fits the actual aging state of the components, improves the accuracy of fault identification, and provides reliable technical support for the entire life cycle of photovoltaic power plants. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the intelligent analysis method for photovoltaic power plant operation data based on machine vision according to the present invention. Detailed Implementation

[0055] 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.

[0056] Example 1: As Figure 1 As shown, this invention provides a technical solution: an intelligent analysis method for photovoltaic power plant operation data based on machine vision. The intelligent analysis method includes the following steps:

[0057] Collect historical aging characteristic data, historical operation data, and real-time operation data of photovoltaic power station components, preprocess the collected data, and construct a standardized dataset;

[0058] The standardized dataset includes historical aging feature data, historical operation data, and real-time operation data, which are integrated according to a unified data structure. The historical aging feature data includes appearance images and operation parameter records of components at different operating stages. The historical operation data includes the component's past running time, start-stop frequency, and load changes. The real-time operation data includes visual images of the component's current operating status and synchronously generated operation parameter data.

[0059] For example:

[0060] Historical aging characteristics data: retrieved from the photovoltaic power plant operation and maintenance archives, including quarterly appearance inspection photos of the modules after commissioning, inverter output power, open circuit voltage, short circuit current and other operating parameter records, covering all operating stages of the entire life cycle of the modules.

[0061] Historical operating data: extracted from the power plant's SCADA system, including cumulative component runtime, daily start-up and shutdown frequency, and load change curves.

[0062] Real-time operation data: Visual images of the current operating status of the components are collected by machine vision cameras deployed on the photovoltaic array, and real-time operating parameters are collected simultaneously by sensors;

[0063] Missing value completion: For missing output power data of a component at a certain time period, the missing data is linearly completed by referring to the power change pattern of the same component at similar time periods in the previous 3 days and combining the average power value of the same type of component under the same environment.

[0064] Data standardization and integration: All data are integrated according to a unified structure, visual images are converted into a standardized pixel format, and running parameters are uniformly retained to two decimal places to form a standardized dataset.

[0065] Based on the standardized dataset, extract the aging characteristic parameters of the components, construct the exclusive aging characteristic profile of each photovoltaic module, and draw the baseline of normal data fluctuation.

[0066] Based on a standardized dataset, aging feature parameters of components are extracted by classification. These aging feature parameters include appearance characterization and performance operation categories. The appearance characterization category is obtained through visual images. Redundant parameters are removed using feature filtering.

[0067] The process of constructing a unique aging feature profile for each photovoltaic module is as follows: using the unique identifier of the photovoltaic module as an index, the selected aging feature parameters are categorized according to the logic of "initial state - operating stage - aging trend" to establish a unique aging feature profile; combining the time dimension information in the historical aging feature data of the module, the evolution law of each aging feature parameter with the operating time is marked to form a unique aging feature profile containing parameter baseline value, evolution rate, and stage threshold; based on the unique aging feature profile, the normal fluctuation range of each aging feature parameter in different operating stages is statistically analyzed to exclude abnormal interference data; parameter fluctuation baselines are drawn according to appearance characterization category and performance operation category, and the normal data fluctuation baseline is formed by combining the change law of parameter fluctuation baseline in the time dimension.

[0068] For example:

[0069] Classification and extraction: Appearance characteristics include the percentage of panel scratch area, the percentage of pixels in yellowed areas, the degree of frame corrosion, and the deviation of component tilt angle; performance characteristics include the output power attenuation rate, open circuit voltage stability, short circuit current fluctuation value, and start-stop response time.

[0070] Feature selection: A combination of "Pearson correlation coefficient + logical elimination" is used for selection. First, the Pearson correlation coefficient between each feature and the aging process of the component is calculated, and candidate features with an absolute correlation coefficient value ≥ 0.7 are selected; then, redundant parameters that cannot distinguish between different aging stages are eliminated, and finally the core features are retained.

[0071] Index creation: Using the component's unique code as the index, core features are categorized according to the logic of "initial state - running stage - aging trend".

[0072] Evolutionary pattern annotation: Combining time dimension information, annotate the evolutionary logic and quantitative indicators of each feature.

[0073] Fluctuation range statistics: Based on the exclusive profile, the 3σ principle is used to exclude abnormal interference data caused by extreme weather, and the natural fluctuation range of each core feature at different stages is statistically analyzed.

[0074] Baseline plotting: Baselines are plotted separately for appearance characteristics and performance operation, incorporating the changing patterns over time. Appearance-related baselines are presented as a curve representing the percentage of yellowed area pixels versus the number of years of operation, while performance-related baselines are presented as a curve representing the output power versus the operating time.

[0075] By combining real-time environmental data to correct the aging feature profile and the normal data fluctuation baseline, and by comparing real-time operation data with the corrected baseline, a list of suspected fault data is constructed and marked as a prediction result.

[0076] Real-time environmental data related to component operation is captured by environmental sensors deployed in photovoltaic power plants, including light intensity, ambient temperature, air humidity, and wind speed. Based on the preset correlation logic between environmental data and component aging characteristic parameters, the collected real-time environmental data is substituted into the component's exclusive aging characteristic profile, and the range of aging characteristic parameters at each stage in the profile is adjusted. At the same time, the baseline of normal data fluctuation is corrected according to the influence of environmental factors on the component's operating status.

[0077] According to the classification of appearance characteristics and performance operation, the real-time operation data is compared with the same type of parameters corresponding to the corrected normal data fluctuation baseline one by one. The values ​​of the real-time operation data in each parameter dimension are checked to see if they are within the normal fluctuation range of the corrected baseline. If a real-time operation data exceeds the range, the corresponding real-time operation data is marked as suspected fault data, and the specific parameters and deviation of the data exceeding the range are recorded to form a list of suspected fault data, which is marked as a prediction result.

[0078] For example:

[0079] By deploying environmental sensors in the photovoltaic power station, a sampling frequency of 10 seconds per sampling is set to capture four types of data in real time: light intensity, ambient temperature, air humidity, and wind speed, and then transmit them to the data processing center.

[0080] Image correction: Adjust the feature parameter ranges of each stage in the image according to the preset association logic.

[0081] Baseline correction: Adjusting the normal fluctuation range according to the influence of environmental factors.

[0082] Classification and comparison: Real-time operation data is divided into appearance characteristics and performance operation categories, and compared with the same parameters of the corrected baseline one by one.

[0083] List generation: Record the component identifier, timestamp, abnormal parameters, and deviation range of suspected fault data to form a structured list of suspected fault data.

[0084] Cross-validate the list of suspected fault data to distinguish between normal aging fluctuations and actual faults, and output the validation results, which are labeled as secondary prediction results.

[0085] Retrieve the component-specific aging feature profile corresponding to the suspected fault data, the corrected normal data fluctuation baseline, and the historical appearance and performance data of the component under the same environmental conditions. At the same time, extract the corresponding parameter data of the same type of component at the same aging stage and under the same environment. Align and compare the suspected fault data list with the two types of reference data in all dimensions, calculate the differences and overlap between the two types of reference data. Combine the suspected fault data list with the evolution law of aging feature parameters at each stage in the component aging feature profile and the synergistic change relationship between parameters, analyze whether the parameter changes of the suspected fault data are consistent with the natural evolution trend and parameter synergy logic of the current aging stage of the component, and output the analysis results list.

[0086] A three-dimensional judgment standard is established to qualitatively classify suspected fault data. At the same time, the difference parameters, matching degree indicators and collaborative logic involved in the qualitative classification are recorded to form a verification result that includes the judgment basis, the correlation of key parameters and the weight of the judgment dimension, and is labeled as a secondary prediction result.

[0087] For example:

[0088] Historical data of components: retrieve the appearance and performance data of the components corresponding to suspected fault data under the same environmental conditions over the past 3 years.

[0089] Reference data for similar components: Extract parameter data for 30 components of the same model, same years of operation, same aging stage, and same environment.

[0090] Difference Calculation: The Euclidean distance formula is used to calculate the difference between the suspected fault data and the two types of reference data, and the degree of overlap of the differences is also calculated.

[0091] Evolution logic verification: Compare with the component's specific profile to analyze whether the changes in suspected data parameters conform to the natural trend of the current aging stage.

[0092] Three-dimensional judgment criteria: Integrating "difference overlap, evolution logic matching degree, and parameter coordination logic", a comprehensive score of ≥0.7 is judged as a real fault, and <0.7 is judged as normal fluctuation of aging.

[0093] Structured Records: The verification results include judgment conclusions, key difference parameters, matching degree indicators, and collaborative logic analysis, forming a standardized report.

[0094] Collect the results of two predictions and manual inspections, and dynamically update the component aging feature profile and judgment threshold.

[0095] Using the unique identifier of the photovoltaic module as an index, the results of the two predictions are classified and aggregated to form a prediction result data set divided by module. Based on the aggregated prediction result data set and manual inspection results, the aging characteristic parameters of the module in the current operating stage and the evolution relationship between parameters are supplemented. The parameter logic in the original profile that does not match the actual aging process is corrected, and the newly formed parameter change patterns and relationships are incorporated into the profile framework. Combined with the updated module-specific aging characteristic profile, the parameter boundary characteristics between normal aging fluctuations and real faults in the verification results are analyzed and verified, and the judgment criteria for normal fluctuations of appearance characterization parameters and performance operation parameters at different stages are adjusted.

[0096] For example:

[0097] The verification results are aggregated using the unique identifier of the component as an index, and the verification results of each component are categorized and aggregated to form a data set divided by component.

[0098] Parameter logic correction: Supplement new aging characteristics in the current operating phase and correct evolution logic that does not match reality.

[0099] Add new relationships: Incorporate newly discovered parameters into the profile framework.

[0100] Adjust the parameter boundaries between normal fluctuations and actual faults in the judgment threshold analysis and verification results, and adjust the judgment criteria for different stages.

[0101] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A machine vision-based intelligent analysis method for photovoltaic power plant operation data, characterized by: The intelligent analysis method includes the following steps: Collect historical aging characteristic data, historical operation data, and real-time operation data of photovoltaic power station components, preprocess the collected data, and construct a standardized dataset; Based on the standardized dataset, extract the aging characteristic parameters of the components, construct the exclusive aging characteristic profile of each photovoltaic module, and draw the baseline of normal data fluctuation. By combining real-time environmental data to correct the aging feature profile and the normal data fluctuation baseline, and by comparing real-time operation data with the corrected baseline, a list of suspected fault data is constructed and marked as a prediction result. Cross-validate the list of suspected fault data to distinguish between normal aging fluctuations and actual faults, and output the validation results, which are labeled as secondary prediction results. Collect the results of two predictions and manual inspections, and dynamically update the component aging feature profile and judgment threshold.

2. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 1, characterized in that: Based on a standardized dataset, aging feature parameters of components are extracted by classification. These aging feature parameters include appearance characterization class and performance operation class. The appearance characterization class is obtained through visual images. Redundant parameters are eliminated using feature-based filtering.

3. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 2, characterized in that: The process of constructing a unique aging characteristic profile for each photovoltaic module is as follows: Using the unique identifier of the photovoltaic module as an index, the filtered aging characteristic parameters are categorized according to the logic of "initial state - operation stage - aging trend" to create a unique aging characteristic profile; By combining the time dimension information in the historical aging feature data of components, the evolution pattern of each aging feature parameter with the runtime is marked, forming a unique aging feature profile that includes parameter baseline value, evolution rate, and stage threshold. Based on the specific aging characteristic profile, the normal fluctuation range of each aging characteristic parameter in different operating stages is statistically analyzed, and abnormal interference data is excluded. Parameter fluctuation baselines are drawn separately for appearance characteristics and performance operation. By combining the variation patterns of parameter fluctuation baselines over time, normal data fluctuation baselines are formed.

4. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 1, characterized in that: Real-time environmental data related to component operation is captured by environmental sensors deployed in photovoltaic power plants, including light intensity, ambient temperature, air humidity, and wind speed. Based on the pre-defined correlation logic between environmental data and component aging characteristic parameters, the collected real-time environmental data is substituted into the component's exclusive aging characteristic profile, the range of aging characteristic parameters at each stage in the profile is adjusted, and the normal data fluctuation baseline is corrected according to the influence of environmental factors on the component's operating status.

5. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 4, characterized in that: According to the classification of appearance characteristics and performance operation, the real-time operation data is compared with the same type of parameters corresponding to the corrected normal data fluctuation baseline one by one. The values ​​of the real-time operation data in each parameter dimension are checked to see if they are within the normal fluctuation range of the corrected baseline. If a real-time operation data exceeds the range, the corresponding real-time operation data is marked as suspected fault data, and the specific parameters and deviation of the data exceeding the range are recorded to form a list of suspected fault data, which is marked as a prediction result.

6. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 1, characterized in that: The system retrieves the component-specific aging feature profile corresponding to the suspected fault data, the corrected normal data fluctuation baseline, and the historical appearance and performance data of the component under the same environmental conditions. Simultaneously, it extracts the corresponding parameter data of similar components at the same aging stage and under the same environment. The suspected fault data list is then compared with the two sets of reference data across all dimensions, calculating the differences and overlap. By comparing the suspected fault data list with the aging feature profile of the component at each stage and the collaborative changes between parameters, the system analyzes whether the parameter changes in the suspected fault data conform to the natural evolution trend and parameter collaboration logic of the component's current aging stage, and outputs a list of analysis results.

7. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 6, characterized in that: A three-dimensional judgment standard is established to qualitatively classify suspected fault data. At the same time, the difference parameters, matching degree indicators and collaborative logic involved in the qualitative classification are recorded to form a verification result that includes the judgment basis, the correlation of key parameters and the weight of the judgment dimension, and is labeled as a secondary prediction result.

8. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 1, characterized in that: Using the unique identifier of the photovoltaic module as an index, the results of the two predictions are classified and aggregated to form a prediction result data set divided by module. Based on the aggregated prediction result data set and manual inspection results, the aging characteristic parameters of the module in the current operating stage and the evolution relationship between parameters are supplemented. The parameter logic in the original profile that does not match the actual aging process is corrected, and the newly formed parameter change patterns and relationships are incorporated into the profile framework. Combined with the updated module-specific aging characteristic profile, the parameter boundary characteristics between normal aging fluctuations and real faults in the verification results are analyzed and verified, and the judgment criteria for normal fluctuations of appearance characterization parameters and performance operation parameters at different stages are adjusted.

9. The intelligent analysis method for photovoltaic power plant operation data based on machine vision according to claim 1, characterized in that: The standardized dataset includes historical aging feature data, historical operation data, and real-time operation data, which are integrated according to a unified data structure. The historical aging characteristic data includes appearance images and operating parameter records of the component at different operating stages; the historical operation data includes the component's past running time, start-stop frequency, and load changes; the real-time operation data includes visual images of the component's current operating status and synchronously generated operating parameter data.

10. A machine vision-based intelligent analysis system for photovoltaic power plant operation data, applied to the machine vision-based intelligent analysis method for photovoltaic power plant operation data as described in any one of claims 1-9, characterized in that: The intelligent analysis system includes a data acquisition and processing module, a feature profile construction module, a fault screening generation module, a cross-validation judgment module, and a profile threshold update module. The data acquisition and processing module is used to collect historical aging characteristic data, historical operation data and real-time operation data of photovoltaic power station components, and after completing the missing data items, integrate them according to a unified structure to construct a standardized dataset. The feature profile construction module is used to extract component aging strongly correlated feature parameters from the standardized dataset, and after filtering out redundant parameters, construct an exclusive aging feature profile with the component's unique identifier as the index, and draw a normal data fluctuation baseline according to the category. The fault screening generation module is used to capture real-time environmental data through environmental sensors, correct the aging characteristic profile and fluctuation baseline, compare real-time operation data with the corrected baseline, mark abnormal data and record relevant information to generate a list of suspected fault data. The cross-validation judgment module is used to retrieve multi-dimensional reference data and compare it with suspected fault data, analyze the matching degree between parameter changes and component aging trends, qualitatively classify data types based on three-dimensional judgment criteria, and output structured verification results. The profile threshold update module is used to collect verification results with the component's unique identifier, supplement and improve the component's aging feature parameters and related logic, update the exclusive aging feature profile, and adjust the normal fluctuation judgment threshold of parameters at different stages. The output of the data acquisition and processing module is connected to the input of the feature profile construction module; the output of the feature profile construction module is connected to the input of the fault initial screening generation module; the output of the fault initial screening generation module is connected to the input of the cross-validation judgment module; and the output of the cross-validation judgment module is connected to the input of the profile threshold update module.

Images