Photovoltaic panel and assembly grid-connected test original test data processing method

Abstract

The invention discloses a photovoltaic panel and assembly grid-connected test original test data processing method, relates to the photovoltaic panel and assembly grid-connected test related technical field, and aims to improve the grid-connected test data processing efficiency and accuracy, and the method comprises the following steps: obtaining an original test data stream in real time; inputting the original test data stream into a dynamic verification window for real-time validity verification; inputting the verified data flow into a data stage division model in real time, and identifying and marking time boundaries of a starting stage, a steady-state operation stage, a transition stage and a closing stage; inputting the stage data marked with the time boundary into a self-adaptive anchor point detection model, and identifying a stage starting anchor point, a stage ending anchor point, a power extreme value anchor point and a parameter mutation anchor point through a clustering algorithm; dynamically updating a data stage division model and a self-adaptive anchor point detection model based on an anchor point detection result and a real-time data stream; and outputting an analysis report. According to the invention, the efficiency and accuracy of grid-connected test data processing can be improved.

Description

Technical Field

[0001] This invention relates to the technical field of grid-connected testing of photovoltaic panels and modules. More specifically, this invention relates to a method for processing raw test data during grid-connected testing of photovoltaic panels and modules. Background Technology

[0002] In the industrial production of photovoltaic panels and modules, grid connection testing is a crucial step in verifying the power generation performance of products and ensuring safe grid connection. The raw test data generated directly determines the accuracy of performance evaluation and fault diagnosis. Currently, the grid connection testing process requires the simultaneous collection of electrical parameters such as voltage, current, and power, involving multiple data dimensions and requiring high real-time performance.

[0003] However, existing data processing methods have significant limitations: on the one hand, raw data streams are susceptible to equipment noise and transient interference, resulting in anomalies such as parameter jumps and timestamp gaps, yet lack efficient real-time verification mechanisms, often leading to abnormal data being mixed into subsequent analyses and affecting the reliability of results; on the other hand, photovoltaic systems undergo different states during testing, such as startup, operation, transition, and shutdown, but existing methods mostly rely on fixed thresholds or manual experience to define stages, making it difficult to dynamically adapt to parameter fluctuations and environmental changes, easily leading to misjudgments of stage boundaries. Furthermore, traditional processing methods are mostly post-hoc static analyses, unable to capture key parameter nodes in real time during testing, making it difficult to quickly locate performance inflection points or anomaly causes, ultimately resulting in insufficient value extraction from test data and restricting the efficiency and accuracy of photovoltaic panel grid-connected testing.

[0004] Therefore, it is necessary to design a technical solution that can overcome the above-mentioned defects. Summary of the Invention

[0005] One objective of this invention is to provide a method for processing raw test data during grid-connected testing of photovoltaic panels and modules, which can help improve the efficiency and accuracy of grid-connected test data processing.

[0006] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a method for processing raw test data for grid-connected testing of photovoltaic panels and modules is provided, comprising: S1: acquiring raw test data stream output by a grid-connected testing system in real time, the raw test data stream including timestamps, voltage data, current data, power data, ambient temperature data, and irradiance data; S2: inputting the raw test data stream into a dynamic verification window for real-time validity verification, the dynamic verification window calculating the dynamic threshold range of voltage data and the dynamic threshold range of current data based on the most recent N data points, and performing data continuity verification based on the timestamp sequence; S3: inputting the verified raw test data stream into a data stage partitioning model in real time. The data phase segmentation model integrates the temporal characteristics of voltage, current, power, ambient temperature, and irradiance data to dynamically identify and mark the time boundaries of the startup, steady-state operation, transition, and shutdown phases. S4: The phase data with marked time boundaries is input into the adaptive anchor point detection model. The adaptive anchor point detection model identifies the phase start anchor point, phase end anchor point, power extreme value anchor point, and parameter mutation anchor point based on the electrical parameter change rate curve and environmental parameter change trend for each phase using a clustering algorithm. S5: Based on the anchor point detection results and real-time data stream, the parameters of the data phase segmentation model and the adaptive anchor point detection model are dynamically updated. S6: An analysis report with phase markings and anchor point information is output.

[0007] Further, S2 includes: S21: Based on the sliding window mechanism, maintain a first dynamic window of length N for voltage data and current data respectively; maintain a second dynamic window of length M based on the timestamp sequence, where M≥N; S22: For the first dynamic window, calculate the moving average and moving standard deviation of voltage and current data within the window in real time, and compare the voltage and current values ​​of the current data point with the threshold range respectively. If any parameter exceeds its threshold range, the data point is marked as an instantaneous outlier; S23: For the second dynamic window, calculate the moving average and moving standard deviation of the interval between consecutive timestamps. If the current time interval deviates significantly from the normal interval determined based on the moving average and moving standard deviation, it is determined that the data flow continuity is interrupted and marked as a continuity outlier; S24: If a data point is only marked as an instantaneous outlier, it is replaced with the moving average based on the first dynamic window; if a data point is marked as a continuity outlier, or is marked as both types of outliers simultaneously, the data point and multiple subsequent data points until the time series returns to normal are removed and recorded as invalid data segments.

[0008] Further, S3 includes: S31: Extracting multi-dimensional time-series features from the verified original test data stream, including voltage fluctuation rate, current change trend, power stability coefficient, and the coordinated change features of ambient temperature and irradiance; S32: Inputting the multi-dimensional time-series features into a data stage partitioning model based on the random forest algorithm, and processing the input features in parallel through multiple decision trees to output the stage probability distributions for the startup stage, steady-state operation stage, transition stage, and shutdown stage respectively; S33: Constructing a stage state transition decision mechanism based on the stage probability distribution. When the stage probability distribution of multiple consecutive data points for the same stage continuously exceeds a preset threshold, a stage transition is determined and the corresponding time boundary is recorded; S34: Introducing a sliding verification window at the identified stage transition time boundary, and ensuring the temporal consistency of stage marking through backtracking verification and look-ahead verification; S35: Dynamically adjusting the feature weights in the data stage partitioning model and the preset threshold in the stage state transition decision mechanism based on historical stage partitioning results.

[0009] Furthermore, in S34, based on the initially identified stage transition time boundary, data segments containing K data points are extracted forward and backward respectively, forming a sliding verification window of length 2K. Within the sliding verification window, the consistency of the stage probability distribution output by the data stage partitioning model is checked: the average probability of the first half of the data points being marked as the previous stage and the average probability of the second half of the data points being marked as the next stage are calculated. If both average probability values ​​are higher than the preset consistency threshold, the stage transition time boundary is confirmed to be valid. If either average probability value is lower than the consistency threshold, it is determined to be a stage marking conflict. When a stage marking conflict occurs, the position of the stage transition time boundary is fine-tuned along the time axis, and the verification process is re-executed until a time boundary that meets the consistency verification conditions is found or the maximum number of fine-tuning times is reached.

[0010] Further, S4 includes: S41: For the data of each labeled stage, extract the electrical parameter change feature set, including voltage change rate sequence, current change rate sequence, power change trajectory, and correlation change pattern of ambient temperature and irradiance; S42: Analyze the electrical parameter change feature set using a density-based clustering algorithm, identify dense regions of data points in the feature space, and mark the center point of each dense region as a candidate anchor point; S43: Determine the type of the candidate anchor points: determine the candidate anchor points located at the beginning of the stage and with significant parameter change rate as the stage start point. Anchor points; candidate anchor points located at the end of a stage and with a significant rate of parameter change are identified as stage termination anchor points; candidate anchor points corresponding to local extreme points in the power change trajectory are identified as power extreme point anchor points; candidate anchor points with abrupt changes in the rate of parameter change are identified as parameter abrupt change anchor points; S44: Establish an anchor point verification mechanism, merge adjacent and similar candidate anchor points, perform significance verification on isolated candidate anchor points, and remove candidate anchor points that do not meet the significance threshold; S45: Based on historical anchor point detection results, dynamically adjust the parameter configuration of the clustering algorithm and the threshold setting for significance verification.

[0011] Furthermore, in S42, a relaxed neighborhood radius parameter and minimum point threshold are used in the first round of clustering to identify all possible candidate dense regions in the feature space. In the second round of clustering, for each candidate dense region, the neighborhood radius parameter is automatically adjusted according to the distribution density of data points in the region for refined clustering analysis. Regions with higher distribution density use a smaller neighborhood radius parameter, while regions with lower distribution density use a larger neighborhood radius parameter. A time continuity constraint is introduced, requiring data points within the same dense region to meet the density connectivity condition in the feature space and to be continuous in the time series, with the time interval between adjacent data points not exceeding a preset time threshold. For each final dense region that simultaneously meets the density clustering requirements and the time continuity constraint, the weighted center point of its feature vector is calculated, and the original data point corresponding to the center point is marked as a candidate anchor point. A confidence evaluation system for candidate anchor points is established, calculating a confidence score based on the number of data points in the dense region, the uniformity of the density distribution, and the stability of the time span. Candidate anchor points are then screened based on the confidence score, with only those with scores higher than the preset confidence threshold retained for subsequent processing.

[0012] Furthermore, in S3, when a candidate anchor point is located within a preset time window at the beginning of a stage's time boundary, and the rate of change of voltage or current at the candidate anchor point continuously exceeds the static operating threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points after the candidate anchor point show a monotonically increasing trend, then the candidate anchor point is determined to be the stage's starting anchor point. When a candidate anchor point is located within a preset time window at the end of a stage's time boundary, and the rate of change of voltage or current at the candidate anchor point is continuously lower than the static operating threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points before the candidate anchor point show a monotonically decreasing trend, The candidate anchor point is then determined as the stage termination anchor point. First, all local extreme points in the power change trajectory are identified. Then, extreme points that meet the significance threshold are screened through extreme value significance test. Finally, the corresponding candidate anchor points are determined as power extreme value anchor points. Local maxima are marked as maximum power anchor points, and local minima are marked as minimum power anchor points. Parameter mutation anchor points are identified by analyzing the coordinated mutation characteristics of voltage change rate, current change rate, and power change rate. It is required that the change rate of at least two parameters simultaneously exceeds their respective change rate thresholds and maintains high consistency in timestamps.

[0013] Furthermore, S5 includes: S51: Establishing a model performance monitoring mechanism to collect in real time the stage identification accuracy index of the data stage partitioning model and the anchor detection accuracy index of the adaptive anchor detection model; S52: Setting parameter update trigger conditions, when the performance index of any model is lower than the corresponding threshold in multiple consecutive test cycles, the parameter update process is initiated; S53: For the data stage partitioning model, based on historically correctly partitioned sample data, the importance score of each temporal feature in stage partitioning is calculated for the first time, and the splitting weight of the corresponding feature in the random forest algorithm is adjusted according to the score; at the same time, the stage state transition decision is optimized based on the stability evaluation results of stage boundary identification. S54: For the adaptive anchor point detection model, based on historically correctly detected anchor point samples, analyze the contribution of different electrical parameter variation characteristics to anchor point identification, and adjust the parameter configuration of the clustering algorithm accordingly; S55: Establish a parameter update verification process. After completing the parameter adjustment, use an independent verification dataset to test the updated model and compare the performance index changes before and after the parameter update; Execute parameter update decisions based on the verification test results: When the performance index of both models reaches the expected improvement target, confirm and save the new parameter configuration; When the performance of either model does not reach the expected level, start the parameter rollback mechanism to restore the parameter state before the update.

[0014] Furthermore, in S6, the analysis report is divided into a data quality assessment module, a phase division details module, a key event anchor point module, and a system performance statistics module. The data quality assessment module records the integrity indicators of the original test data stream, the distribution statistics of abnormal data points, and the data validity assessment results. The phase division details module displays the start and end time boundaries of each phase in a time series and marks the key electrical parameter characteristics during phase transitions. The key event anchor point module categorizes and lists all detected phase start anchor points, phase end anchor points, power extreme value anchor points, and parameter mutation anchor points, and annotates the time location, parameter value, and confidence score of each anchor point. The system performance statistics module calculates and displays the accuracy indicators of phase identification, the precision indicators of anchor point detection, and the historical effect evaluation of model parameter updates. It generates a time-aligned visualization chart that synchronously displays the original test data curves, phase division regions, and anchor point marker positions, forming a complete test process analysis map. Based on the analysis results, it generates actionable recommendations, including measures to improve data acquisition quality, optimization suggestions for test parameter settings, and early warning information for system maintenance.

[0015] The present invention has at least the following beneficial effects: This invention utilizes a real-time dynamic verification window to validate and handle anomalies in the raw data stream. It accurately removes or corrects invalid data such as parameter jumps and timestamp breaks, significantly improving input data quality and preventing abnormal data from interfering with subsequent analysis, thus providing a reliable data foundation for performance evaluation. This invention integrates multi-dimensional time-series features to dynamically identify test stages, overcoming the limitations of traditional fixed thresholds or manual experience-based divisions. It accurately marks the time boundaries of stages such as startup and steady-state, making the test process state division more closely match actual operating conditions. Through adaptive anchor point detection to capture key parameter nodes, it can quickly locate core events such as power extrema and parameter mutations, facilitating timely identification of performance inflection points or anomaly causes. The dynamic parameter update mechanism of this invention allows the system to continuously adapt to different test scenarios, ensuring a stable improvement in processing accuracy. The final analysis report output by this invention, containing stage and anchor point information, provides a clear basis for product performance evaluation, fault diagnosis, and test process optimization, comprehensively improving the efficiency and accuracy of grid-connected test data processing.

[0016] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0017] Figure 1 This is a flowchart of one embodiment of this application. Detailed Implementation

[0018] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0019] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0020] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0021] like Figure 1As shown, embodiments of this application provide a method for processing raw test data for grid-connected testing of photovoltaic panels and modules, including: S1: Real-time acquisition of raw test data stream output by the grid-connected testing system, the raw test data stream including timestamps, voltage data, current data, power data, ambient temperature data, and irradiance data; S2: Inputting the raw test data stream into a dynamic verification window for real-time validity verification, the dynamic verification window calculates the dynamic threshold range of voltage data and the dynamic threshold range of current data based on the most recent N data points, and performs data continuity verification based on the timestamp sequence; S3: Real-time inputting the verified raw test data stream into a data stage partitioning model, the data stage partitioning model integrating voltage... The system dynamically identifies and marks the time boundaries of the startup, steady-state operation, transition, and shutdown phases based on the temporal characteristics of data, including current, power, ambient temperature, and irradiance. S4: The stage data with marked time boundaries are input into the adaptive anchor point detection model. The adaptive anchor point detection model identifies the stage start anchor point, stage end anchor point, power extreme value anchor point, and parameter mutation anchor point based on the electrical parameter change rate curve and environmental parameter change trend for each stage using a clustering algorithm. S5: Based on the anchor point detection results and real-time data stream, the parameters of the data stage division model and the adaptive anchor point detection model are dynamically updated. S6: An analysis report with stage markings and anchor point information is output.

[0022] For example, this photovoltaic panel and module production line involves a method for processing raw test data for grid connection testing. This method is applied to the grid connection performance testing stage before photovoltaic modules leave the factory, aiming to accurately assess the grid connection adaptability of modules through systematic processing of test data. Specifically, step S1 involves real-time acquisition of the raw test data stream output by the grid connection testing system. This grid connection testing system can be the test data acquisition system built into the Huawei SUN2000 series grid-connected inverter or the GoodWe photovoltaic grid connection testing device. The timestamps in the raw data stream can use UTC time or Beijing time accurate to milliseconds to mark the acquisition time of each set of data. The voltage data is the DC or AC voltage value at the grid connection end of the photovoltaic panel and module, with a value range of 0-1000V or 0-1500V, used to reflect the voltage output characteristics of the module. The current data is the DC or AC current value in the grid connection circuit, with a value range of 0-50A or 0-10A. 0A reflects the current output capability of the module; power data is the real-time active power output value of the module, which is the core parameter for evaluating the module's power generation performance; ambient temperature data is collected by a temperature sensor deployed in the test environment. This temperature sensor can be an SHT30 high-precision temperature and humidity sensor or a DS18B20 digital temperature sensor, with a value range of -40℃ to 85℃, used to correct for the impact of ambient temperature on module performance; irradiance data is collected by a solar radiometer in the test area. This solar radiometer can be a TBQ-2 total radiation meter or an MS-80 portable radiometer, with a value range of 0-2000W / ㎡, reflecting the effect of light intensity on module output.

[0023] Step S2 inputs the raw test data stream into the dynamic verification window for real-time validity verification. The dynamic verification window is an analysis window that slides based on a preset number of data points or time length. It calculates the dynamic threshold range of voltage and current data based on the most recent N data points. The value of N can be 50 or 100. The dynamic threshold range is calculated as follows: first, the moving average of the most recent N data points is calculated, and then the moving standard deviation is calculated. The dynamic threshold range of voltage data is set to the moving average ± 2 times the moving standard deviation or the moving average ± 3 times the moving standard deviation. The dynamic threshold range of current data uses the same calculation logic. At the same time, data continuity is checked based on the timestamp sequence. Specifically, the interval between two adjacent timestamps is calculated. If the interval exceeds 100ms or 200ms, it is determined that there is a continuity problem in the data.

[0024] Step S3 inputs the verified raw test data stream into the data phase segmentation model in real time. This data phase segmentation model is an algorithm model that integrates multi-dimensional time-series features, including voltage fluctuation rate, current change trend, power stability coefficient, and the coordinated change features of ambient temperature and irradiance. The voltage fluctuation rate is (current voltage value - previous voltage value) / previous voltage value × 100%. The current change trend is determined by linearly fitting the slope of the first 10 or 20 current data points. The power stability coefficient is the ratio of the difference between the maximum and minimum power values ​​within a certain time period to the average value. The coordinated change features of ambient temperature and irradiance are obtained by calculating their Pearson correlation coefficient. The model dynamically identifies and marks the time boundaries of the startup phase, steady-state operation phase, transition phase, and shutdown phase through these features. The startup phase is the time period from when the component is powered on until the power rises to 80% of the rated power. The steady-state operation phase is the time period when the power fluctuation amplitude is ≤5% or ≤8%. The transition phase is the time period when the power switches from one steady-state range to another. The shutdown phase is the time period when the power drops from 20% of the rated power to 0.

[0025] Step S4 inputs the stage data with marked time boundaries into the adaptive anchor point detection model. This model analyzes the electrical parameter change rate curves and environmental parameter change trends for each stage. The electrical parameter change rate curves include voltage change rate curves and current change rate curves. The voltage change rate is (current voltage value - previous voltage value) / time interval. The current change rate is calculated using the same logic. The environmental parameter change trend is determined by the mean change of ambient temperature and irradiance within a sliding window. The model identifies key anchor points using a clustering algorithm, which can be either DBSCAN density clustering or K-means clustering. The identified stage start anchor point is the parameter data point corresponding to the start of the stage, the stage end anchor point is the parameter data point corresponding to the end of the stage, the power extreme value anchor point is the parameter data point corresponding to the maximum or minimum power value within the stage, and the parameter mutation anchor point is the parameter data point where the voltage, current, or power change rate exceeds a preset threshold. The voltage change rate mutation threshold can be 10V / s or 15V / s, and the current change rate mutation threshold can be 5A / s or 8A / s.

[0026] Step S5 dynamically updates the parameters of the data phase segmentation model and the adaptive anchor point detection model based on the anchor point detection results and the real-time data stream. Specifically, the latest detected anchor point data and the real-time acquired data stream are used as samples to adjust the weights of each temporal feature in the data phase segmentation model, such as increasing the weight of the power stability coefficient in steady-state phase identification. At the same time, the neighborhood radius parameter of the clustering algorithm in the adaptive anchor point detection model is updated. The neighborhood radius can be adjusted from 0.5 to 0.8 or from 1.0 to 1.2.

[0027] Step S6 outputs an analysis report with stage markers and anchor point information. The report includes the start and end times of each stage within the test period, the timestamp corresponding to each anchor point, parameter values ​​and types, and the validity evaluation results of the original data. The report can be in PDF or Excel format for easy archiving and analysis later.

[0028] In existing technologies, the processing of raw data from grid-connected photovoltaic panels and modules largely relies on manual operation. Staff manually filter raw data using Excel spreadsheets, judging anomalies based on experience, lacking a systematic real-time verification mechanism. This makes it difficult to accurately identify abnormal data such as voltage jumps and timestamp gaps. Phase division also relies on manually setting fixed time nodes, such as defining 0-5 minutes as the startup phase and 5-20 minutes as the steady-state phase, which cannot adapt to the performance differences of different modules. Anchor point identification requires manual inspection of data curve markers one by one, which is inefficient and prone to omissions. This embodiment achieves real-time data verification through a dynamic verification window, more accurately removing abnormal data and ensuring data validity compared to existing technologies. A multi-feature fusion phase division model automatically identifies phase boundaries, avoiding division errors caused by manual experience. An adaptive anchor point detection model automatically locates key anchor points, improving detection efficiency. Simultaneously, a dynamic model parameter update mechanism is introduced, allowing the processing method to continuously adapt to different test scenarios. This solves the problems of high manual dependence, low processing accuracy, and poor efficiency in existing technologies, providing a more reliable technical solution for photovoltaic module grid-connected test data processing.

[0029] In another embodiment, S2 includes: S21: Based on a sliding window mechanism, maintain a first dynamic window of length N for voltage data and current data respectively; maintain a second dynamic window of length M based on the timestamp sequence, where M≥N; S22: For the first dynamic window, calculate the moving average and moving standard deviation of voltage data and current data within the window in real time, and compare the voltage value and current value of the current data point with the threshold range respectively. If any parameter exceeds its threshold range, the data point is marked as an instantaneous outlier; S23: For the second dynamic window, calculate the moving average and moving standard deviation of the interval between consecutive timestamps. If the current time interval deviates significantly from the normal interval determined based on the moving average and moving standard deviation, it is determined that the data flow continuity is interrupted and marked as a continuity outlier; S24: If a data point is only marked as an instantaneous outlier, it is replaced with the moving average based on the first dynamic window; if a data point is marked as a continuity outlier, or is marked as both types of outliers simultaneously, the data point and multiple subsequent data points until the time series returns to normal are removed and recorded as invalid data segments.

[0030] For example, step S2 specifically includes the following: Step S21 maintains a first dynamic window of length N for voltage data and current data respectively based on the sliding window mechanism. The sliding window mechanism means that the window moves forward point by point as the data acquisition progresses, and the earliest data point is removed when a new data point is added. The value of N can be 30 or 60. At the same time, a second dynamic window of length M is maintained based on the timestamp sequence. The value of M must satisfy M≥N, and can be 50 or 80. This window is used to monitor the temporal continuity of data acquisition. Step S22: For the first dynamic window, calculate the moving average and moving standard deviation of the voltage and current data within the window in real time. The moving average is obtained by summing all data points within the window and dividing by the number of data points. The moving standard deviation is obtained by summing the squares of the differences between each data point and the moving average, dividing by the number of data points, and then taking the square root. After calculation, compare the current voltage and current values ​​with the corresponding threshold ranges. The threshold range for voltage data is the moving average ± 2 times the moving standard deviation, and the threshold range for current data is the moving average ± 2.5 times the moving standard deviation. If the current voltage or current value exceeds the corresponding threshold range, mark the data point as a transient anomaly. Step S23: For the second dynamic window, calculate the moving average and moving standard deviation of the interval between consecutive timestamps, using the same calculation logic as the first dynamic window. If the current time interval significantly deviates from the normal range determined based on the moving average and moving standard deviation (the normal range is set to the moving average ± 3 times the moving standard deviation), and the current time interval continues to exceed 200ms or 300ms, then a data flow continuity interruption is determined, and the data is marked as a continuity anomaly. Step S24 performs different processing based on the outlier type. If a data point is only marked as an instantaneous outlier, it is replaced with a moving average based on the first dynamic window to ensure the continuity of the data series. If a data point is marked as a continuous outlier, or is marked as both an instantaneous and continuous outlier, the data point and all subsequent data points until the time series returns to normal are removed. The criterion for the time series to return to normal is that the time intervals of 5 or 8 consecutive data points are within the normal range. At the same time, the removed data segment is recorded as an invalid data segment, and the start and end timestamps of the invalid data segment are marked.

[0031] In existing technologies, anomaly handling in photovoltaic grid-connected test data often employs fixed threshold filtering, such as setting a fixed voltage threshold of 0-1200V and a fixed current threshold of 0-60A. Data exceeding these ranges is discarded. This approach fails to adapt to dynamic parameter changes during testing and is prone to misinterpreting normal fluctuations. Furthermore, for data continuity issues, simply deleting data points with anomalies at single time intervals without considering the correlation of subsequent data results in gaps in the data sequence. This embodiment uses dual dynamic windows to verify both parameter values ​​and time continuity. The moving threshold of the first dynamic window adapts to dynamic parameter changes, reducing misjudgments. The second dynamic window handles continuity anomalies by covering subsequent related data, ensuring the integrity of the data sequence. Simultaneously, differentiated processing methods are used for different anomaly types, improving the accuracy of anomaly handling and the effectiveness of the data sequence compared to existing technologies, providing a more reliable data foundation for subsequent data phase division and anchor point detection.

[0032] In another embodiment, S3 includes: S31: Extracting multi-dimensional time-series features from the verified original test data stream, including voltage fluctuation rate, current change trend, power stability coefficient, and the coordinated change features of ambient temperature and irradiance; S32: Inputting the multi-dimensional time-series features into a data stage partitioning model based on a random forest algorithm, processing the input features in parallel through multiple decision trees, and outputting the stage probability distributions for the startup stage, steady-state operation stage, transition stage, and shutdown stage respectively; S33: Constructing a stage state transition decision mechanism based on the stage probability distribution, determining that a stage transition has occurred and recording the corresponding time boundary when the stage probability distribution of multiple consecutive data points for the same stage continuously exceeds a preset threshold; S34: Introducing a sliding verification window at the identified stage transition time boundary, ensuring the temporal consistency of stage markings through backtracking verification and look-ahead verification; S35: Dynamically adjusting the feature weights in the data stage partitioning model and the preset threshold in the stage state transition decision mechanism based on historical stage partitioning results.

[0033] For example, step S3 specifically includes the following: Step S31 extracts multi-dimensional time-series features from the verified original test data stream. The multi-dimensional time-series features include voltage fluctuation rate, current change trend, power stability coefficient, and the coordinated change features of ambient temperature and irradiance. Voltage fluctuation rate is calculated as (current voltage - previous voltage) / previous voltage × 100%, calculated once every 100ms or 200ms. Current change trend is obtained by performing linear regression on 15 or 25 consecutive current data points, with the slope of the regression line representing the change trend. Power stability coefficient is the difference between the maximum and minimum power values ​​within a 1-minute or 2-minute time period, divided by the average power value within that time period, with the result rounded to two decimal places. The coordinated change features of ambient temperature and irradiance are obtained by calculating the Pearson correlation coefficient between the two within a sliding window. The sliding window length can be 10 or 15 data points, and the correlation coefficient ranges from -1 to 1, reflecting the degree of correlation between the two.

[0034] Step S32 inputs multi-dimensional temporal features into a data stage partitioning model based on a random forest algorithm. This random forest algorithm contains 50 or 100 decision trees, each constructed based on a randomly selected subset of features and a subset of data. After inputting features, each decision tree determines the stage to which the data belongs and outputs the probability values ​​of the start-up stage, steady-state operation stage, transition stage, and shutdown stage. Then, a voting mechanism is used to determine the stage probability distribution of the data point, that is, the proportion of votes obtained by each stage to the total number of decision trees.

[0035] Step S33 constructs a stage state transition decision mechanism based on the stage probability distribution. The preset stage probability threshold is 80% or 85%. When the stage probability distribution of 10 or 15 consecutive data points for the same stage continuously exceeds the threshold, it is determined that a stage transition has occurred, and the timestamp of the current data point is recorded as the time boundary of the stage transition.

[0036] Step S34 introduces a sliding verification window at the identified stage transition time boundary. The sliding verification window uses the initially identified stage transition time boundary as a reference and cuts out data segments containing K data points forward and backward respectively. The value of K can be 20 or 30, which together form a sliding verification window with a length of 2K. Within the window, the consistency of the stage probability distribution output by the data stage segmentation model is checked. The average probability of the data points in the first half of the window being marked as the previous stage and the average probability of the data points in the second half of the window being marked as the next stage are calculated. If both average probability values ​​are higher than the consistency threshold of 85% or 90%, the stage transition time boundary is confirmed to be valid.

[0037] Step S35 dynamically adjusts the feature weights in the data stage segmentation model and the preset thresholds in the stage state transition decision mechanism based on the historical stage segmentation results. Specifically, it calculates the contribution of each time-series feature in the historical data when correctly segmenting the stage. For example, if the power stability coefficient has a high contribution in the steady-state stage identification, its weight in the model is increased. At the same time, based on the accuracy of the historical stage transition judgment, if the accuracy is less than 90%, the stage probability threshold is adjusted from 80% to 85% to improve the accuracy of stage segmentation.

[0038] In existing technologies, photovoltaic grid-connected test data segmentation often employs a fixed-time segmentation method, such as dividing the entire test process into three stages: startup, steady state, and shutdown, or based on preset fixed parameter thresholds, such as determining the steady state stage when the power reaches 500W. This approach fails to adapt to differences in component performance and variations in the test environment, resulting in significant deviations in stage boundaries. This embodiment extracts multi-dimensional time-series features to comprehensively reflect the dynamic changes in parameters during the test process. The random forest algorithm-based model integrates the judgments of multiple decision trees, improving the reliability of stage identification. A sliding verification window further validates stage boundaries, reducing misjudgments. Simultaneously, by dynamically adjusting model parameters based on historical data, the segmentation effect is continuously optimized. Compared to existing technologies, this approach solves the problems of rigid and low-precision stage segmentation, providing more accurate stage data for subsequent anchor point detection.

[0039] In another embodiment, in S34, based on the initially identified stage transition time boundary, data segments containing K data points are extracted forward and backward respectively, forming a sliding verification window of length 2K. Within the sliding verification window, the consistency of the stage probability distribution output by the data stage division model is checked: the average probability of the first half of the data points being marked as the previous stage and the average probability of the second half of the data points being marked as the next stage are calculated. If both average probability values ​​are higher than a preset consistency threshold, the stage transition time boundary is confirmed to be valid. If either average probability value is lower than the consistency threshold, it is determined to be a stage marking conflict. When a stage marking conflict occurs, the position of the stage transition time boundary is fine-tuned along the time axis, and the verification process is re-executed until a time boundary that meets the consistency verification conditions is found or the maximum number of fine-tuning times is reached.

[0040] For example, in step S34, sliding verification is performed based on the initially identified stage transition time boundary. First, the length of the sliding verification window is determined, and data segments containing K data points are extracted forward and backward respectively. The value of K can be 15 or 25, and the two together constitute a sliding verification window of length 2K. The window covers the key data before and after the stage transition, ensuring the comprehensiveness of the verification. Within the sliding verification window, the consistency of the stage probability distribution output by the data stage division model is checked. Specifically, the window is divided into a first half and a second half. The first half corresponds to the historical data before the stage transition, and the second half corresponds to the new data after the stage transition. The probability value of each data point in the first half of the window being marked as the previous stage is calculated, summed, and divided by the number of data points to obtain the average probability of the previous stage. The same logic is used to calculate the average probability value of the data points in the second half being marked as the next stage. Set a consistency threshold of 88% or 92%. If the mean probability of both the previous and subsequent stages is higher than this threshold, the initially identified stage transition time boundary is confirmed to be valid and requires no adjustment. If either mean probability is lower than the consistency threshold, a stage marker conflict is identified. In this case, the position of the stage transition time boundary is fine-tuned along the time axis, with each fine-tuning moving the boundary by 1 or 2 data points. After fine-tuning, the sliding verification window is re-captured, and the mean probability calculation and consistency judgment process are executed. Set a maximum fine-tuning limit of 5 or 8 times. If a time boundary that meets the consistency verification conditions is found within the maximum number of fine-tunings, the adjusted time boundary is confirmed as the final stage transition time boundary. If the conditions are not met even after reaching the maximum number of fine-tunings, the uncertainty of the stage transition is recorded, and the stage transition characteristics of historical similar data are used for auxiliary judgment to ensure the integrity of the stage division.

[0041] In existing technologies, once the stage boundaries of photovoltaic grid-connected test data are determined, there is a lack of effective verification mechanisms. Reliance is placed solely on the initial output of the model. If the model is misjudged due to momentary disturbances, the stage boundary deviation will directly affect subsequent analysis and cannot be corrected in a timely manner. This embodiment comprehensively verifies the rationality of the boundaries by covering data before and after stage transitions through a sliding verification window; it quantifies the verification results through probability mean calculation to avoid subjective judgment bias; and it balances verification accuracy and efficiency with a fine-tuning mechanism for conflict situations and a maximum number of iterations. Compared to existing technologies, this improves the reliability of stage boundaries, ensuring the accuracy of subsequent anchor point detection.

[0042] In another embodiment, S4 includes: S41: For the data of each labeled stage, extract electrical parameter change feature set, including voltage change rate sequence, current change rate sequence, power change trajectory, and correlation change pattern of ambient temperature and irradiance; S42: Analyze the electrical parameter change feature set using a density-based clustering algorithm to identify dense regions of data points in the feature space, and mark the center point of each dense region as a candidate anchor point; S43: Determine the type of the candidate anchor points: determine the candidate anchor points located at the beginning of the stage and with significant parameter change rate as the stage. Starting anchor point; candidate anchor points located at the end of a stage and with significant parameter change rate are identified as stage end anchor points; candidate anchor points corresponding to local extreme points in the power change trajectory are identified as power extreme point anchor points; candidate anchor points with abrupt parameter change rate are identified as parameter abrupt change anchor points; S44: Establish an anchor point verification mechanism, merge adjacent and similar candidate anchor points, perform significance verification on isolated candidate anchor points, and remove candidate anchor points that do not meet the significance threshold; S45: Based on historical anchor point detection results, dynamically adjust the parameter configuration of the clustering algorithm and the threshold setting for significance verification.

[0043] For example, step S41 extracts an electrical parameter change feature set for the data of each marked stage. This feature set covers the voltage change rate sequence, current change rate sequence, power change trajectory, and the correlation change pattern of ambient temperature and irradiance. The voltage change rate sequence is calculated as (current voltage - previous voltage) / 0.1s, and a value is recorded every 0.1s or 0.2s to form sequence data. The calculation logic of the current change rate sequence is consistent with that of the voltage. The power change trajectory is formed by sorting all power data points in the stage by timestamp and fitting them to form a continuous curve. The correlation change pattern of ambient temperature and irradiance is determined by calculating the ratio of their change amplitudes within the same time window. The length of the time window can be 1 minute or 1.5 minutes. Step S42 uses a density-based clustering algorithm to analyze the electrical parameter variation feature set. The clustering algorithm is the DBSCAN algorithm, with the neighborhood radius parameter set to 0.6 or 0.9 and the minimum number of points threshold set to 5 or 8. The algorithm identifies dense regions of data points in the feature space. The criteria for determining a dense region are that the number of data points in the region reaches the minimum number of points threshold, and any data point in the region can be connected to other data points in the region through neighboring data points. After identification, the center point of each dense region is marked as a candidate anchor point. The center point is determined by calculating the average value of the feature values ​​of all data points in the region. Step S43 determines the type of candidate anchor points. If a candidate anchor point is located within a preset time window at the start of a phase (5 seconds or 8 seconds before or after the start of the phase), and the voltage change rate at the anchor point exceeds the static operating threshold of 3V / s or 5V / s for 3 or 5 consecutive times, while the power values ​​of the 10 or 15 data points following the anchor point show a monotonically increasing trend, then the candidate anchor point is determined to be a phase start anchor point. If a candidate anchor point is located within a preset time window at the end of a phase (5 seconds or 8 seconds before or after the end of the phase), and the voltage change rate at the anchor point is below the static operating threshold of 1V / s or 2V / s for 3 or 5 consecutive times, while the power values ​​of the 10 or 15 data points before the anchor point show a monotonically decreasing trend, then it is determined to be a phase end anchor point. First, through... The extreme points of the power change trajectory are calculated by differentiation, and then the extreme value significance is tested by t-test with a significance level set at 0.05 or 0.01. Extreme points that meet the significance threshold are selected and the corresponding candidate anchor points are determined as power extreme value anchor points. Local maxima are marked as maximum power anchor points, and local minima are marked as minimum power anchor points. The values ​​of voltage change rate, current change rate, and power change rate are calculated. If the change rate of at least two parameters simultaneously exceeds their respective change rate thresholds (voltage change rate threshold is 8V / s or 12V / s, current change rate threshold is 4A / s or 6A / s, and power change rate threshold is 50W / s or 80W / s), and the difference in change timestamps is ≤100ms or ≤150ms, maintaining a high degree of consistency, then it is determined as a parameter abrupt change anchor point.Step S44 establishes an anchor point verification mechanism, merging adjacent candidate anchor points of the same type. The adjacent criteria are that the time interval between the two anchor points is ≤200ms or ≤300ms. After merging, the average of the feature values ​​of the two anchor points is used as the final anchor point feature value. For isolated candidate anchor points, significance verification is performed. An isolated anchor point refers to an anchor point that has no other candidate anchor points of the same type among the surrounding 5 or 8 data points. Significance verification is performed by calculating the difference between the feature values ​​of the anchor point and the surrounding data points. The difference threshold is set to 5% or 8%. If the difference is lower than the threshold, the candidate anchor point is removed. Candidate anchor points that do not meet the significance threshold are also removed. Step S45 dynamically adjusts the parameter configuration of the clustering algorithm and the threshold setting of significance verification based on the historical anchor point detection results. It statistically analyzes the accuracy of different types of anchor point detection in historical data. If the accuracy of anchor point detection at the beginning of the stage is less than 90%, the neighborhood radius of the DBSCAN algorithm is adjusted from 0.6 to 0.7. If the false detection rate of power extreme value anchor points is higher than 5%, the difference threshold of significance verification is adjusted from 8% to 6% to optimize the anchor point detection effect.

[0044] In existing technologies, anchor point identification for photovoltaic grid-connected test data largely relies on manual observation of data curve markings, such as manually searching for power peak points as extreme anchor points. This is not only inefficient but also susceptible to subjective factors, leading to missed or misjudged anchor points. Some methods use fixed threshold detection, such as marking a sudden change anchor point as the voltage change rate exceeding 10V / s, which cannot adapt to parameter characteristics at different stages and has low detection accuracy. This embodiment extracts multi-dimensional electrical parameter change features to comprehensively characterize the anchor point's attribute attributes; density-based clustering algorithms can accurately identify feature-dense regions and reduce interference from irrelevant data; multi-condition constraints and verification mechanisms for anchor point type determination improve the accuracy of anchor point classification; and parameter adjustments based on historical data enable the detection method to continuously adapt to different scenarios. Compared with existing technologies, this solves the problems of low efficiency and poor accuracy in anchor point detection, providing key anchor point information for subsequent model updates and report output.

[0045] In another embodiment, in S42, a relaxed neighborhood radius parameter and minimum point threshold are used in the first round of clustering to identify all possible candidate dense regions in the feature space. In the second round of clustering, for each candidate dense region, the neighborhood radius parameter is automatically adjusted according to the distribution density of data points in the region to perform refined clustering analysis. Regions with higher distribution density use a smaller neighborhood radius parameter, and regions with lower distribution density use a larger neighborhood radius parameter. A time continuity constraint is introduced, where data points in the same dense region satisfy the density connection condition in the feature space and have continuity in the time series, and the time interval between adjacent data points must not exceed a preset time threshold. For each final dense region that simultaneously meets the density clustering requirements and the time continuity constraint, the weighted center point of its feature vector is calculated, and the original data point corresponding to the center point is marked as a candidate anchor point. A confidence evaluation system for candidate anchor points is established, and a confidence score is calculated based on the number of data points in the dense region, the uniformity of the density distribution, and the stability of the time span. Candidate anchor points are screened based on the confidence score, and only candidate anchor points with scores higher than the preset confidence threshold are retained for subsequent processing.

[0046] For example, in step S42, a two-step clustering method is used to analyze the electrical parameter variation feature set. The first step is preliminary clustering, using a relaxed neighborhood radius parameter and minimum point threshold. The neighborhood radius parameter is set to 1.2 or 1.5, and the minimum point threshold is set to 3 or 5. This setting identifies all possible candidate dense regions in the feature space, avoiding the omission of potential anchor point feature regions. The second step is refined clustering. For each candidate dense region identified in the first step, the distribution density of data points in the region is calculated. The distribution density is the number of data points in the region divided by the volume of the region in the feature space. The neighborhood radius parameter is automatically adjusted according to the distribution density. If the distribution density is high, such as the number of data points per unit feature space volume ≥ 10, a smaller neighborhood radius parameter is used, set to 0.4 or 0.6; if the distribution density is low, such as the number of data points per unit feature space volume < 5, a larger neighborhood radius parameter is used, set to 0.8 or 1.0. Refined clustering analysis is achieved through differentiated parameter configuration. A temporal continuity constraint is introduced, requiring data points within the same dense region to satisfy the density connectivity condition in the feature space. This means any two data points within the region can be connected through other data points within the region, and the data points must be continuous in time. The time interval between adjacent data points must not exceed 150ms or 250ms. If the time interval exceeds this range, the region is divided into different dense regions. For each final dense region that simultaneously satisfies the density clustering requirement and the temporal continuity constraint, a weighted centroid of its feature vector is calculated. The weight is determined based on the density of data points within the region; points with higher density have greater weights. Specifically, the weight is the proportion of the number of data points in the neighboring area of ​​that point to the total number of data points in the region. The original test data point corresponding to this weighted centroid is marked as a candidate anchor point, ensuring that the candidate anchor point represents the core features of the region. A confidence assessment system for candidate anchor points was established, calculating confidence scores from three dimensions: First, the number of data points in the densely populated area; the more data points, the higher the score, with a maximum score of 30 points. ≥20 data points = 30 points, 10-19 = 20 points, 5-9 = 10 points, and <5 = 5 points. Second, the uniformity of the density distribution, determined by calculating the standard deviation of the data point density within the area; the smaller the standard deviation, the higher the score, with a maximum score of 30 points. ≤0.2 standard deviation = 30 points, 0.2-0... 0.5 scores 20 points, 0.5-1.0 scores 10 points, and >1.0 scores 5 points; the third is the stability of the time span, which is the time difference between the first and last data points in the region. The closer the time difference is to the preset reasonable range (such as 1-3s), the higher the score, with a full score of 40 points. A time difference within 1-3s scores 40 points, 0.5-1s or 3-5s scores 25 points, <0.5s or 5-10s scores 10 points, and >10s scores 5 points; the three scores are added together to form the total confidence score, with a full score of 100 points.Candidate anchors are screened based on confidence scores. A pre-set confidence threshold of 60 or 70 points is used. Only candidate anchors with scores higher than the threshold are retained for further processing. Candidate anchors with scores lower than the threshold are considered non-critical anchors and are removed.

[0047] Using a single clustering parameter, such as a fixed neighborhood radius of 1.0, cannot adapt to feature regions of varying densities, leading to inaccurate identification of dense regions. Furthermore, the lack of consideration for temporal continuity easily results in similar feature points from different time periods being grouped into the same region, and the absence of confidence assessment for candidate anchor points makes it impossible to distinguish the importance of anchor points. This embodiment, through differentiated parameter configuration in a two-step clustering approach, accurately identifies feature regions of varying densities; temporal continuity constraints ensure the temporal correlation of data within the region; weighted centroid calculation makes candidate anchor points more representative; and the confidence assessment system effectively filters key anchor points. Compared to existing technologies, this improves the quality and accuracy of candidate anchor points, laying a solid foundation for subsequent anchor point type determination.

[0048] In another embodiment, in S3, when a candidate anchor point is located within a preset time window at the beginning of a stage time boundary, and the rate of change of voltage or current at the candidate anchor point continuously exceeds the static operating threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points after the candidate anchor point show a monotonically increasing trend, then the candidate anchor point is determined to be a stage start anchor point; when a candidate anchor point is located within a preset time window at the end of a stage time boundary, and the rate of change of voltage or current at the candidate anchor point is continuously lower than the static operating threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points before the candidate anchor point show a monotonically decreasing trend, then the candidate anchor point is determined to be a stage start anchor point. If the potential is identified, the candidate anchor point is determined to be the stage termination anchor point. First, all local extreme points in the power change trajectory are identified. Then, extreme points that meet the significance threshold are screened through extreme value significance test. Finally, the corresponding candidate anchor points are determined as power extreme value anchor points, where local maxima are marked as maximum power anchor points and local minima are marked as minimum power anchor points. Parameter mutation anchor points are identified by analyzing the coordinated mutation characteristics of voltage change rate, current change rate, and power change rate. It is required that the change rate of at least two parameters simultaneously exceeds their respective change rate thresholds and maintains high consistency in timestamps.

[0049] For example, in step S3, the type determination of candidate anchor points adopts a multi-condition joint constraint method. For the stage start anchor point, a preset time window at the start of the stage time boundary is first determined. This window is a time period extending 2s or 3s forward and 3s or 5s backward from the stage start time. If the candidate anchor point is located within this window, the change of electrical parameters is further checked. The voltage change rate and current change rate at the candidate anchor point are calculated. The voltage change rate threshold is set to 2V / s or 4V / s, and the current change rate threshold is set to 1A / s or 2A / s. If the voltage change rate or current change rate exceeds the static working threshold of the corresponding parameter 4 or 6 times consecutively, and the power value of the 8 or 12 consecutive data points after the candidate anchor point shows a monotonically increasing trend, that is, the power value of the next data point is always greater than the power value of the previous data point, then the candidate anchor point is determined to be the stage start anchor point. For the stage termination anchor point, a preset time window for the stage time boundary termination end is first determined. The window is a time period extending 3s or 5s before the stage termination time and 2s or 3s after the stage termination time. If the candidate anchor point is located within this window, the voltage change rate and current change rate at the anchor point are calculated. The static working threshold for the voltage change rate is set to 0.5V / s or 1V / s, and the static working threshold for the current change rate is set to 0.2A / s or 0.5A / s. If the voltage change rate or current change rate is lower than the corresponding threshold for 4 or 6 consecutive times, and the power values ​​of the 8 or 12 consecutive data points before the candidate anchor point show a monotonically decreasing trend, that is, the power value of the subsequent data point is always less than the power value of the previous data point, then the candidate anchor point is determined to be the stage termination anchor point. For power extreme value anchor points, firstly, all local extreme points in the power change trajectory are identified by differentiation. The first derivative of the power change trajectory function is calculated, and points where the derivative changes from positive to negative are local maxima, and points where the derivative changes from negative to positive are local minima. Then, the extreme value significance is tested using a t-test, with a significance level of 0.05 or 0.01. The difference between the extreme point and the power values ​​of the surrounding 10 or 15 data points is calculated. If the difference passes the t-test, i.e., P < 0.05 or P < 0.01, the extreme points that meet the significance threshold are selected. Finally, the corresponding candidate anchor points are determined as power extreme value anchor points, with local maxima marked as maximum power anchor points and local minima marked as minimum power anchor points. For parameter mutation anchor points, the co-mutation characteristics of voltage change rate, current change rate, and power change rate are analyzed for identification. The mutation thresholds for voltage change rate are set to 6V / s or 9V / s, current change rate is set to 3A / s or 5A / s, and power change rate is set to 40W / s or 60W / s. If the change rates of at least two parameters simultaneously exceed their respective change rate thresholds, and the timestamp difference between these parameter changes is ≤80ms or ≤120ms, maintaining a high degree of temporal consistency, then the corresponding candidate anchor point is determined to be a parameter mutation anchor point.

[0050] In existing technologies, anchor point type determination for photovoltaic grid-connected test data often relies on a single condition, such as determining the maximum power anchor point solely based on the maximum power value, without considering the significance of extreme values, easily misclassifying minor fluctuations as extreme values. Stage start and end anchor points are determined only based on time location, without considering parameter change trends, leading to misclassification. Parameter mutation anchor points only detect a single parameter, ignoring the coordinated changes of multiple parameters, resulting in frequent missed detections. This embodiment determines anchor point types through multi-condition joint constraints. Stage start and end anchor points combine time windows, parameter change rates, and power trends; power extreme value anchor points introduce significance testing; and parameter mutation anchor points focus on the coordination of multiple parameters. Compared to existing technologies, this improves the accuracy of anchor point type determination, avoids misclassification and missed detection, and provides accurate anchor point classification data for subsequent model updates.

[0051] In another embodiment, S5 includes: S51: establishing a model performance monitoring mechanism to collect in real time the stage identification accuracy index of the data stage partitioning model and the anchor detection accuracy index of the adaptive anchor detection model; S52: setting parameter update trigger conditions, and initiating the parameter update process when the performance index of any model is lower than the corresponding threshold in multiple consecutive test cycles; S53: for the data stage partitioning model, based on historically correctly partitioned sample data, calculating the importance score of each temporal feature in stage partitioning for the first time, and adjusting the splitting weight of the corresponding feature in the random forest algorithm according to the score; and simultaneously optimizing the stage state transition based on the stability evaluation results of stage boundary identification. S54: For the adaptive anchor point detection model, based on historically correctly detected anchor point samples, analyze the contribution of different electrical parameter variation characteristics to anchor point identification, and adjust the parameter configuration of the clustering algorithm accordingly; S55: Establish a parameter update verification process. After completing the parameter adjustment, use an independent verification dataset to test the updated model and compare the performance index changes before and after the parameter update; Execute parameter update decisions based on the verification test results: When the performance index of both models reaches the expected improvement target, confirm and save the new parameter configuration; When the performance of either model does not reach the expected level, start the parameter rollback mechanism to restore the parameter state before the update.

[0052] For example, step S51 establishes a model performance monitoring mechanism to collect performance indicators of two core models in real time: one is the stage identification accuracy indicator of the data stage segmentation model, calculated as the ratio of the number of correctly segmented stage data points to the total number of data points; the other is the anchor point detection accuracy indicator of the adaptive anchor point detection model, calculated as the ratio of the number of correctly detected anchor points to the total number of anchor points. The performance indicator collection frequency is set to once every 5 minutes or every 10 minutes to ensure timely monitoring of the model's operating status. Step S52 sets parameter update trigger conditions, setting the stage identification accuracy threshold of the data stage segmentation model to 92% or 95%, and the anchor point detection accuracy threshold of the adaptive anchor point detection model to 90% or 93%. If the performance indicator of any model falls below the corresponding threshold within 3 or 5 consecutive test cycles, the parameter update process is initiated. The test cycle is consistent with the performance indicator collection cycle, which is one cycle every 5 minutes or every 10 minutes.

[0053] Step S53 involves adjusting parameters for the data stage segmentation model based on historically correctly segmented sample data. First, a feature importance assessment algorithm, such as the Gini coefficient importance assessment method built into random forest, is used to calculate the importance score of each time-series feature in stage segmentation. The higher the score, the greater the contribution of the feature to stage segmentation. The split weight of the corresponding feature in the random forest algorithm is adjusted according to the score. For example, if the voltage fluctuation rate score is higher than other features, its split weight is adjusted from 0.2 to 0.3. At the same time, the probability threshold in the stage state transition decision mechanism is adjusted according to the stability assessment results of stage boundary identification. The stability assessment is determined by calculating the time deviation values ​​of 10 or 20 consecutive stage boundaries. If the average deviation value is >1s, the probability threshold is adjusted from 82% to 85%. Step S54, for the adaptive anchor point detection model, analyzes the direction of parameter adjustment based on historically correctly detected anchor point samples, and statistically analyzes the contribution of different electrical parameter change characteristics to anchor point identification. For example, if the power change trajectory contributes significantly to the identification of power extreme value anchor points, the weight of the power change trajectory feature in the clustering algorithm is adjusted from 0.25 to 0.35. At the same time, the parameter configuration of the clustering algorithm is adjusted according to the false detection rate of anchor point detection. If the false detection rate of anchor points with parameter mutations is >3%, the minimum number of points threshold of the DBSCAN algorithm is adjusted from 5 to 6. Step S55 establishes a parameter update verification process. After completing the parameter adjustment, the updated model is tested using an independent verification dataset. The independent verification dataset consists of photovoltaic grid-connected test data that did not participate in model training, and the data volume is no less than 20% of the total data volume. After the test, the performance indicators before and after the parameter update are compared. If the stage recognition accuracy of the data stage segmentation model improves by ≥2% and the anchor detection accuracy of the adaptive anchor detection model improves by ≥1.5%, and the performance indicators of both models meet the expected improvement targets, then the new parameter configuration is confirmed and saved. If the performance indicators of either model do not meet the expectations, such as the accuracy of the data stage segmentation model not improving or even decreasing, then the parameter rollback mechanism is activated to restore the parameter state before the update, and the problem points of this parameter adjustment are recorded for subsequent optimization.

[0054] In existing technologies, photovoltaic grid-connected test data processing models mostly operate with fixed parameters. Once the model is trained, the parameters are not adjusted. As the types of test components increase or the test environment changes, the model performance gradually declines, yet timely optimization is not possible. While some models do update parameters, they lack scientific triggering conditions and verification mechanisms, resulting in significant performance fluctuations after updates. This embodiment addresses this by real-time monitoring of model performance indicators to accurately determine update timing; adjusting parameters based on historical samples and contribution analysis to ensure reasonable adjustment directions; employing an independent verification process that compares datasets and performance to guarantee update effectiveness; and implementing a rollback mechanism to prevent performance degradation. Compared to existing technologies, this approach achieves dynamic and scientific optimization of model parameters, enabling the model to continuously adapt to changes in test scenarios and maintain good processing performance.

[0055] In another embodiment, in S6, the analysis report is divided into a data quality assessment module, a phase division details module, a key event anchor point module, and a system performance statistics module. The data quality assessment module records the integrity indicators of the original test data stream, the distribution statistics of abnormal data points, and the data validity assessment results. The phase division details module displays the start and end time boundaries of each phase in a time series and marks the key electrical parameter characteristics during phase transitions. The key event anchor point module lists all detected phase start anchor points, phase end anchor points, power extreme value anchor points, and parameter mutation anchor points, and annotates the time location, parameter value, and confidence score of each anchor point. The system performance statistics module calculates and displays the accuracy indicators of phase identification, the precision indicators of anchor point detection, and the historical effect evaluation of model parameter updates. A time-aligned visualization chart is generated, synchronously displaying the original test data curves, phase division regions, and anchor point marker positions to form a complete test process analysis map. Based on the analysis results, executable suggestions are generated, including measures to improve data acquisition quality, optimization suggestions for test parameter settings, and early warning information for system maintenance.

[0056] For example, in step S6, the analysis report is modularly designed, divided into four core modules: a data quality assessment module, a phase division details module, a key event anchor point module, and a system performance statistics module. Each module has a clear function and the data is interconnected, forming a complete analysis system. In the data quality assessment module, the integrity index of the original test data stream, the distribution statistics of abnormal data points, and the data validity assessment results are recorded. The integrity index is the ratio of the actual number of data points collected to the theoretical number of data points that should be collected. The theoretical number of data points that should be collected is calculated based on the test duration and collection frequency. For example, if the test duration is 30 minutes and the collection frequency is 0.1s / time, the theoretical number of data points is 18,000. The distribution statistics of abnormal data points include the number of instantaneous and continuous abnormal points, the timestamp distribution of each abnormal point, and the start and end times and duration statistics of invalid data segments. The data validity assessment results are represented by the ratio of the number of valid data points to the total number of collected data points, and the main factors affecting data validity, such as equipment noise or signal interference, are also marked. The phase division details module displays the start and end time boundaries of each phase in a time sequence, accurate to the millisecond level. It also marks key electrical parameter characteristics during phase transitions, such as power, voltage, and current values ​​when transitioning from the startup phase to the steady-state phase, as well as ambient temperature and irradiance. The duration of each phase is calculated, for example, the startup phase lasts 2 minutes and 30 seconds, and the steady-state phase lasts 20 minutes, providing a basis for evaluating component performance at each stage. The critical event anchor point module lists all detected phase start anchor points, phase end anchor points, power extreme value anchor points, and parameter mutation anchor points. Each anchor point is annotated with specific information, including time location (timestamp accurate to milliseconds), parameter values ​​(voltage, current, power, ambient temperature, and irradiance values ​​at the corresponding moment), and confidence score (calculated according to the confidence assessment system in claim 6). The correlation between anchor points is also marked, such as whether a parameter mutation anchor point triggers a phase transition, helping users quickly locate critical events and their impact.

[0057] The system performance statistics module calculates and displays the accuracy index of stage identification, the precision index of anchor point detection, and the historical performance evaluation of model parameter updates. The accuracy index of stage identification is the proportion of correctly classified stage data points to the total number of data points, and the precision index of anchor point detection is the proportion of correctly detected anchor points to the total number of detected anchor points. The historical performance evaluation of model parameter updates includes the time of each parameter update, a comparison of performance indicators before and after the update, and the stable runtime after the update, providing historical reference for model optimization. During report generation, a time-aligned visualization chart is created. The X-axis of the chart is the time axis, accurate to the second, while the Y-axis simultaneously displays the numerical changes of voltage, current, and power. Different colored curves represent voltage, current, and power curves, respectively. Different colored shading areas are used in the chart to mark the division range of each stage, such as blue shading for the startup stage and green shading for the steady-state stage. Anchor points are marked with different shaped markers, such as circular markers for stage start anchor points and triangular markers for power extreme value anchor points. The original test data curves, stage division areas, and anchor point marker positions are displayed synchronously to form a complete test process analysis map. Based on the analysis results, actionable recommendations are generated. Data acquisition quality improvement measures include, based on noise issues identified in the data quality assessment, suggesting the replacement of sensors with higher precision ones or adjusting the installation location of the acquisition equipment. Test parameter setting optimization recommendations include, based on the parameter characteristics of each stage in the stage division details, suggesting adjustments to the voltage rise rate during startup or the power monitoring threshold during steady-state operation. System maintenance early warning information includes, based on frequently occurring parameter mutation anchor points in the critical event anchor points, prompting checks on the contact status of the grid-connected circuit or the operational stability of the test equipment, ensuring that the recommendations have practical operational value.

[0058] In existing technologies, photovoltaic grid-connected test analysis reports are mostly simple data listings, such as only displaying raw voltage and current data tables, lacking structured module design, and key information is buried in a large amount of data; the visualization level is low, mostly showing single parameter curves, which cannot intuitively present stage and anchor point information; the suggestions are too general, such as "optimize data acquisition", lacking actionability. This embodiment, through modular report design, makes information clearly classified and easy to look up; time-aligned visualization graphs intuitively display key information throughout the testing process; targeted and actionable suggestions provide guidance for actual operation. Compared with existing technologies, this improves the practicality and readability of the report, and can better support photovoltaic module performance evaluation and test process optimization.

[0059] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for processing raw test data from grid-connected testing of photovoltaic panels and modules, characterized in that, include: S1: Real-time acquisition of raw test data stream output by the grid-connected test system. The raw test data stream includes timestamps, voltage data, current data, power data, ambient temperature data, and irradiance data. S2: Input the raw test data stream into the dynamic verification window for real-time validity verification. The dynamic verification window calculates the dynamic threshold range of voltage data and the dynamic threshold range of current data based on the most recent N data points, and performs data continuity verification based on the timestamp sequence. S3: The verified raw test data stream is input into the data phase segmentation model in real time. The data phase segmentation model integrates the temporal characteristics of voltage data, current data, power data, ambient temperature data, and irradiance data to dynamically identify and mark the time boundaries of the startup phase, steady-state operation phase, transition phase, and shutdown phase. S4: Input the stage data with marked time boundaries into the adaptive anchor point detection model. The adaptive anchor point detection model identifies the stage start anchor point, stage end anchor point, power extreme value anchor point, and parameter mutation anchor point based on the electrical parameter change rate curve and environmental parameter change trend of each stage through a clustering algorithm. S5: Based on the anchor point detection results and real-time data stream, dynamically update the parameters of the data stage partitioning model and the adaptive anchor point detection model; S6: Outputs an analysis report with stage markers and anchor point information.

2. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 1, characterized in that, S2 include: S21: Based on the sliding window mechanism, maintain a first dynamic window of length N for voltage data and current data respectively; maintain a second dynamic window of length M based on the timestamp sequence, where M≥N; S22: For the first dynamic window, calculate the moving average and moving standard deviation of the voltage and current data within the window in real time, and compare the voltage and current values ​​of the current data point with the threshold range respectively. If any parameter exceeds its threshold range, mark the data point as an instantaneous outlier. S23: For the second dynamic window, calculate the moving average and moving standard deviation of the interval between consecutive timestamps. If the current time interval deviates significantly from the normal range determined based on the moving average and moving standard deviation, it is determined that the data flow continuity has been interrupted and marked as a continuity anomaly. S24: If a data point is marked only as an instantaneous outlier, it is replaced with a moving average based on the first dynamic window; if a data point is marked as a continuous outlier, or is marked as both outliers, the data point and all subsequent data points until the time series returns to normal are removed and recorded as invalid data segments.

3. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 1, characterized in that, S3 include: S31: Extract multi-dimensional time-series features from the validated raw test data stream. These features include voltage fluctuation rate, current variation trend, power stability coefficient, and the coordinated variation characteristics of ambient temperature and irradiance. S32: Input multi-dimensional temporal features into a data stage partitioning model based on the random forest algorithm, process the input features in parallel through multiple decision trees, and output the stage probability distributions of the startup stage, steady-state operation stage, transition stage, and shutdown stage respectively. S33: Construct a stage state transition decision mechanism based on stage probability distribution. When the stage probability distribution of multiple consecutive data points for the same stage continuously exceeds the preset threshold, determine that a stage transition has occurred and record the corresponding time boundary. S34: Introduce a sliding verification window at the time boundary of the identified stage transition to ensure the temporal consistency of stage markings through backtracking verification and look-ahead verification; S35: Based on the historical stage division results, dynamically adjust the feature weights in the data stage division model and the preset thresholds in the stage state transition decision mechanism.

4. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 3, characterized in that, In S34, based on the initially identified stage transition time boundary, data segments containing K data points are extracted forward and backward respectively, forming a sliding verification window of length 2K. Within the sliding verification window, the consistency of the stage probability distribution output by the data stage partitioning model is checked: the mean probability of the first half of the data points being marked as the previous stage and the mean probability of the second half of the data points being marked as the next stage are calculated. If both mean probability values ​​are higher than the preset consistency threshold, the stage transition time boundary is confirmed to be valid. If either mean probability value is lower than the consistency threshold, it is determined to be a stage marking conflict. When a stage marking conflict occurs, the position of the stage transition time boundary is fine-tuned along the time axis, and the verification process is re-executed until a time boundary that meets the consistency verification conditions is found or the maximum number of fine-tuning times is reached.

5. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 3, characterized in that, S4 include: S41: For the data of each labeled stage, extract the electrical parameter change feature set, including voltage change rate sequence, current change rate sequence, power change trajectory, and the correlation change pattern of ambient temperature and irradiance; S42: A density-based clustering algorithm is used to analyze the electrical parameter variation feature set, identify dense regions of data points in the feature space, and mark the center point of each dense region as a candidate anchor point; S43: Determine the type of candidate anchor points: determine the candidate anchor points located at the beginning of the stage and with a significant rate of parameter change as stage start anchor points; determine the candidate anchor points located at the end of the stage and with a significant rate of parameter change as stage end anchor points; determine the candidate anchor points corresponding to local extreme points in the power change trajectory as power extreme point anchor points; determine the candidate anchor points with abrupt changes in the rate of parameter change as parameter abrupt change anchor points. S44: Establish an anchor point verification mechanism to merge adjacent and identical candidate anchor points, verify the significance of isolated candidate anchor points, and remove candidate anchor points that do not meet the significance threshold. S45: Based on historical anchor point detection results, dynamically adjust the parameter configuration of the clustering algorithm and the threshold setting for significance verification.

6. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 5, characterized in that, In S42, a relaxed neighborhood radius parameter and minimum point threshold are used in the first round of clustering to identify all possible candidate dense regions in the feature space. In the second round of clustering, for each candidate dense region, the neighborhood radius parameter is automatically adjusted according to the distribution density of data points in the region for refined clustering analysis. Regions with higher distribution density use a smaller neighborhood radius parameter, while regions with lower distribution density use a larger neighborhood radius parameter. A time continuity constraint is introduced, requiring data points within the same dense region to satisfy the density connection condition in the feature space and to be continuous in the time series, with the time interval between adjacent data points not exceeding a preset time threshold. For each final dense region that simultaneously meets the density clustering requirements and the time continuity constraint, the weighted center point of its feature vector is calculated, and the original data point corresponding to the center point is marked as a candidate anchor point. A confidence evaluation system for candidate anchor points is established, calculating a confidence score based on the number of data points in the dense region, the uniformity of the density distribution, and the stability of the time span. Candidate anchor points are then screened based on the confidence score, with only those with scores higher than the preset confidence threshold retained for subsequent processing.

7. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 6, characterized in that, In S3, when a candidate anchor point is located within a preset time window at the beginning of the stage time boundary, the rate of change of voltage or current at the candidate anchor point continuously exceeds the static working threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points after the candidate anchor point show a monotonically increasing trend, then the candidate anchor point is determined to be the stage start anchor point. When a candidate anchor point is located within a preset time window at the end of a phase time boundary, and the rate of change of voltage or current at the candidate anchor point is continuously lower than the static working threshold of the corresponding parameter, and the power values ​​of multiple consecutive data points before the candidate anchor point show a monotonically decreasing trend, then the candidate anchor point is determined to be the phase termination anchor point. First, all local extreme points in the power change trajectory are identified. Then, extreme points that meet the significance threshold are screened out through extreme value significance test. Finally, the corresponding candidate anchor points are determined as power extreme value anchor points. Local maximum points are marked as maximum power anchor points, and local minimum points are marked as minimum power anchor points. By analyzing the coordinated abrupt change characteristics of voltage change rate, current change rate, and power change rate, parameter abrupt change anchor points are identified. It is required that the change rate of at least two parameters simultaneously exceeds their respective change rate thresholds and maintains a high degree of consistency on the timestamp.

8. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 6, characterized in that, S5 include: S51: Establish a model performance monitoring mechanism to collect data in real time, including the stage recognition accuracy index of the stage segmentation model and the anchor detection accuracy index of the adaptive anchor detection model. S52: Set the parameter update trigger condition. When the performance index of any model is lower than the corresponding threshold in multiple consecutive test cycles, the parameter update process will be started. S53: For the data stage partitioning model, based on historically correctly partitioned sample data, the importance score of each time series feature in the stage partitioning is calculated for the first time, and the split weight of the corresponding feature in the random forest algorithm is adjusted according to the score; at the same time, the probability threshold in the stage state transition decision mechanism is optimized according to the stability evaluation results of stage boundary identification. S54: For the adaptive anchor point detection model, based on historically correctly detected anchor point samples, analyze the contribution of different electrical parameter variation characteristics to anchor point identification, and adjust the parameter configuration of the clustering algorithm accordingly. S55: Establish a parameter update verification process. After completing the parameter adjustment, use an independent verification dataset to test the updated model and compare the changes in performance indicators before and after the parameter update. Execute parameter update decisions based on the verification test results: when the performance indicators of both models reach the expected improvement target, confirm and save the new parameter configuration; when the performance of either model does not reach the expected level, activate the parameter rollback mechanism to restore the parameter state before the update.

9. The method for processing raw test data for grid-connected testing of photovoltaic panels and modules as described in claim 6, characterized in that, In S6, the analysis report is divided into a data quality assessment module, a phase division details module, a key event anchor point module, and a system performance statistics module. In the data quality assessment module, the integrity indicators of the original test data stream, the distribution statistics of abnormal data points, and the data validity assessment results are recorded. In the phase division details module, the start and end time boundaries of each phase are displayed in time series, and the key electrical parameter characteristics during phase transitions are marked. In the critical event anchor point module, all detected stage start anchor points, stage end anchor points, power extreme value anchor points, and parameter mutation anchor points are listed in categories, with annotations for the time location, parameter value, and confidence score of each anchor point. In the system performance statistics module, the accuracy index of stage identification, the precision index of anchor point detection, and the historical effect evaluation of model parameter updates are calculated and displayed. A visualization chart with time alignment is generated, synchronously displaying the original test data curves, stage division regions, and anchor point marker positions to form a complete test process analysis map. Based on the analysis results, actionable recommendations are generated, including measures to improve data acquisition quality, suggestions for optimizing test parameter settings, and early warning information for system maintenance.

Images