A current step precision detection method for new energy power equipment test

Abstract

The application discloses a current step accurate detection method for new energy power equipment testing, comprising the following steps: S1, a data processing module loads a target current test file, automatically identifies a time column and a current column, performs data type conversion and data cleaning, and adaptively reduces sampling to generate reduced sampling data; S2, a steady state recognition module performs standardization processing on the current column of the reduced sampling data, and adopts an unsupervised clustering algorithm for clustering; the application automatically recognizes a steady state section through the unsupervised clustering, and sets a dynamic proportional threshold to realize step accurate positioning; from data loading, preprocessing, steady state recognition and step positioning to index calculation and visual output, no manual intervention is needed, the consistency and reliability of detection results under different equipment and different working conditions are ensured, accurate data are provided for power equipment performance verification, and the application is especially suitable for large-scale application scenes such as new energy station equipment acceptance and annual detection.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing technology, specifically to a precise current step detection method for testing new energy power equipment. Background Technology

[0002] Current step detection is a technical means to achieve state monitoring, fault diagnosis or control triggering by identifying step changes in current signals. In the performance testing of new energy power equipment, it is mainly used in dynamic response verification, grid connection evaluation and other links. Its detection accuracy can determine the grid connection adaptability and transient adjustment speed of the equipment, and affect the power generation efficiency of the station and the safety of the power grid.

[0003] Existing technologies mainly employ three methods for detection, but these are mostly dominated by manual intervention or rely on a single algorithm adaptation mode: The manual threshold method requires setting a fixed current change threshold based on experience, and then judging the step by comparing the difference with the threshold. While logically simple and suitable for simple testing in low-noise environments, it relies entirely on human experience. Fixed thresholds cannot adapt to different equipment and operating conditions, leading to frequent misjudgments and missed detections. Its efficiency and consistency are insufficient, with errors exceeding 10%, and manual extraction of indicators is required to amplify the error. The simple difference method achieves automatic judgment through first-order difference, sliding window smoothing, and dynamic thresholds, simplifying operation and suitable for detecting large changes in medium to low noise. However, it is prone to misjudging steady-state fluctuations and requires manual assistance in extracting indicators, making full automation impossible. The cluster-assisted detection method uses algorithms such as K-Means to divide steady-state and step segments, eliminating the need for manual threshold setting and making it suitable for standard laboratory testing. However, the number of clusters needs to be manually set, resulting in insufficient adaptability. It is sensitive to noise, leading to large positioning errors, and easily overlooks the continuity of signal timing, causing boundary judgment errors, still requiring manual confirmation of the boundaries.

[0004] Therefore, current step detection of all equipment in a single power station usually requires 3-5 professional technicians to complete it in 1-2 weeks. This involves a large investment of personnel and is time-consuming. In particular, for large-scale testing of energy power stations, the efficiency and accuracy are not high enough. Summary of the Invention

[0005] The purpose of this invention is to provide a precise current step detection method for testing new energy power equipment, which can realize automatic identification of steady-state segment, precise step location, integrated calculation of multiple indicators and standardized visualization output, replace manual testing process, improve testing efficiency and accuracy, and meet the large-scale testing needs of new energy power plants.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for accurate detection of current step changes in new energy power equipment, comprising the following steps: S1. The data processing module loads the target current test file, automatically identifies the time column and current column, performs data type conversion and data cleaning, and adaptively downsamples to generate downsampled data. S2. The steady-state identification module standardizes the current column of the downsampled data, uses an unsupervised clustering algorithm to cluster the data, and selects the effective steady-state segments. S3. The step candidate extraction module extracts candidate step segments based on the effective steady-state segment and expands the range forward and backward to obtain the final candidate step window. S4, the step precise positioning module performs differential smoothing of the raw current data within the candidate step window, sets a dynamic proportional threshold, and performs precise step positioning. S5. The performance calculation module automatically calculates performance indicators, performs anomaly detection and prompts based on the location results. S6. Visualize the results and save them.

[0007] Preferably, in step S1, the data cleaning includes numerical type conversion, missing value removal, and outlier filtering. The time column and current column are forcibly converted to numerical type, rows containing missing values ​​are removed, and current outliers are identified based on the 3σ principle. The data is then completed using linear interpolation.

[0008] Preferably, in step S1, the adaptive downsampling dynamically adjusts the downsampling ratio according to the amount of original data. When the amount of data is greater than 10,000 points, downsampling is performed at a set ratio of 10:1. When the amount of data is less than or equal to 10,000 points, the original data is directly retained.

[0009] Preferably, in step S2, the unsupervised clustering algorithm uses the DBSCAN algorithm to select segments that are not noise clusters and whose duration is greater than or equal to a set threshold as valid steady-state segments, and outputs a data list containing the start time, end time and average current value of each steady-state segment.

[0010] Preferably, in step S3, for adjacent effective steady-state segments, the interval from the end time of the previous steady state to the start time of the subsequent steady state is taken as a candidate step segment, and data within the candidate step window is extracted from the original data.

[0011] Preferably, in step S4, the differential smoothing employs a 3-point center rolling average method, and the dynamic proportional threshold is calculated based on the average current difference between the preceding and following steady-state segments. Step start threshold = Total change × Step threshold ratio; Step end threshold = total change × step threshold ratio; The step positioning logic is as follows: the time corresponding to the first data point that exceeds the start threshold is the step start time, and the time after the last data point that exceeds the end threshold is the step end time. If no threshold is detected, the candidate window boundary is used as the default value or the step end time.

[0012] Preferably, in step S5, the performance index is calculated as follows: Total change = Post-steady-state average current - Pre-steady-state average current; Target response current = pre-steady-state average current + total change × response proportionality coefficient; Response time = Target response current at the corresponding time point - Step start time; Adjustment time = Step end time - Step start time; Overshoot = (peak current during step transition - average current before steady state) / total change × 100%.

[0013] Preferably, in step S5, when the absolute value of the total change is less than 1e-6 amperes, it is determined that there is no effective step, and when there are fewer than 5 step segment data points, it is marked as insufficient data.

[0014] Preferably, in step S6, standardized charts are drawn using matplotlib to achieve visualization output, and the charts support custom sizes, colors, fonts, and legend positions.

[0015] Preferably, the standardized charts include an overall current curve chart and magnified charts of each step. The charts can be saved in PNG format to a specified path, and the chart file name includes the device name and test date information.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention automatically identifies steady-state segments through unsupervised clustering and sets dynamic proportional thresholds to achieve precise step positioning. From data loading, preprocessing, steady-state identification, step positioning to index calculation and visualization output, no manual intervention is required, ensuring the consistency and reliability of test results under different equipment and operating conditions. It provides accurate data for power equipment performance verification, and is especially suitable for large-scale application scenarios such as acceptance and annual testing of new energy power station equipment. 2. This invention supports batch processing from data loading to visualization output, reducing the testing time for a single test report from 2-3 hours to within 10 minutes, reducing manual workload by more than 80%, and improving testing efficiency; 3. This invention can cope with complex environments such as electromagnetic interference and sensor noise through data cleaning, and is suitable for dynamic response characteristic testing of power equipment such as SVG and inverters. It solves the problems of cluster-assisted detection method being sensitive to noise and having incorrect boundary judgment, and is compatible with mainstream data formats such as Excel and CSV, thus improving compatibility. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the present invention; Figure 2This is a schematic diagram of DBSCAN clustering steady-state identification according to the present invention; Figure 3 This is a logic diagram for calculating the performance indicators of this invention; Figure 4 This is a schematic diagram showing the visualization results output of the present invention. Detailed Implementation

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

[0019] Please see Figure 1-4 This invention provides a technical solution: a method for accurate detection of current step changes in new energy power equipment testing, comprising the following steps: S1. The data processing module loads the target current test file, automatically identifies the time column and current column, performs data type conversion and data cleaning, and adaptively downsamples to generate downsampled data. The data processing module supports mainstream power test data formats such as Excel and CSV. It automatically identifies the time and current columns. The default time (s) is for the time column and the default current (A) is for SVG No. 1. If the column name is inconsistent with the default configuration, users can customize the column name through parameters, such as time_col='test time' and current_col='A phase current', to adapt to the test data formats of different devices. Convert the time and current columns to numeric types and remove rows with missing values; use the 3σ principle to identify and interpolate to complete abnormal current values ​​to ensure data continuity; sort the data in ascending order by the time column and reset the data index.

[0020] S2. The steady-state identification module standardizes the current column of the downsampled data, uses an unsupervised clustering algorithm to cluster the data, and selects the effective steady-state segments. S3. The step candidate extraction module extracts candidate step segments based on the effective steady-state segment and expands the range forward and backward to obtain the final candidate step window. S4, the step precise positioning module performs differential smoothing of the raw current data within the candidate step window, sets a dynamic proportional threshold, and performs precise step positioning. S5. The performance calculation module automatically calculates performance indicators, performs anomaly detection and prompts based on the location results. S6. Visualize the results and save them.

[0021] In step S1, the data cleaning includes numerical type conversion, missing value removal, and outlier filtering. The time column and current column are forcibly converted to numerical type, rows with missing values ​​are removed, and current outliers are identified based on the 3σ principle. The data is completed using linear interpolation to ensure data continuity. The data is sorted in ascending order by the time column, and the data index is reset. In step S1, adaptive downsampling dynamically adjusts the downsampling ratio based on the amount of original data. When the amount of data is greater than 10,000 points, downsampling is performed at a set ratio of 10:1 to reduce the amount of clustering calculation. When the amount of data is less than or equal to 10,000 points, the original data is directly retained to avoid the loss of key features. The downsampled data is only used for steady-state identification, while the original data is completely retained for subsequent step positioning and index calculation. The preprocessed original data, downsampled data, and basic data information are output. The basic data information includes the number of original data points, the number of downsampled data points, the average sampling interval, and the data time range.

[0022] In step S2, the current column of the downsampled data is standardized to have a mean of 0 and a variance of 1, eliminating the influence of dimensions and improving clustering accuracy. The unsupervised clustering algorithm used is the DBSCAN algorithm, which does not require pre-specifying the number of clusters and can adapt to changes in the number of steady-state segments under different equipment and operating conditions. After calling the DBSCAN algorithm, the standardized current data is input. The clustering parameter neighborhood radius eps defaults to 0.05, where the neighborhood radius eps can be adjusted according to the fluctuation amplitude of the current signal. The minimum number of cluster samples min_samples defaults to 10 to adapt to the normal sampling rate of 10. The frequency range is 00Hz to 10000Hz. The minimum steady-state duration, min_steady_duration, defaults to 0.01 seconds to ensure that short-term fluctuations are not misjudged as steady state. The clustering label for each data point is obtained. The clustering results are traversed, and segments with non-noise clusters and a duration greater than or equal to a set threshold are selected as valid steady-state segments. That is, continuous data segments with clustering labels ≠ -1 and no noise are extracted. The time span of each segment is calculated, and segments with a time span ≥ 0.01 seconds are retained as valid steady-state segments. The output is a data list containing the start time, end time, and average current value of each steady-state segment.

[0023] like Figure 2 As shown, the DBSCAN clustering classification effect on downsampled current data is intuitively demonstrated: the horizontal axis is time (seconds) and the vertical axis is current (A). Different colors mark different clusters. Segments that are not noise clusters and have a duration ≥ a set threshold are marked as steady-state segments. Noise clusters and short-duration clusters are judged as non-steady-state segments. The automatic division logic of steady-state segments is clear.

[0024] In step S3, adjacent steady-state segments are traversed in chronological order. For adjacent valid steady-state segments, the interval from the end time of the previous steady state to the start time of the subsequent steady state is taken as a candidate step segment to ensure complete coverage of the step transition process. Data within the candidate step window is extracted from the original data. To avoid truncating the transition data before and after the step, the initial candidate range is expanded forward and backward. The expansion time can be configured by parameters, with a default forward expansion of 0.005 seconds and a backward expansion of 0.005 seconds. The expanded window is taken as the final candidate step window. All data within this window, including time and current columns, are extracted from the original data for subsequent precise positioning and serve as the candidate step data corresponding to the adjacent steady-state segment. If the number of candidate data points is less than 5, it is marked as an invalid candidate and the subsequent processing is skipped.

[0025] Output a list of candidate step segments, including the pre-steady-state information, post-steady-state information, window start time, window end time, and candidate data for each candidate; In step S4, differential smoothing uses the 3-point center rolling average method, and the dynamic proportional threshold is calculated based on the average current difference between the preceding and following steady-state segments: step start threshold = total change × step threshold ratio, with the default step threshold ratio being 5%; The step end threshold = total change × step threshold ratio. The step threshold ratio defaults to 2% to ensure that the threshold and step amplitude are adaptively matched, avoiding missed detection of small step amplitudes and false detection of large step amplitudes. The step positioning logic is based on the smoothed differential signal. The time corresponding to the first data point that exceeds the start threshold is the step start time, and the time after the last data point that exceeds the end threshold is the step end time. When no threshold exceeding point is detected, the candidate window boundary is used as the default value or the step end time to ensure the robustness of the positioning logic.

[0026] In step S5, the performance indicators are calculated as follows: Total change = Post-steady-state average current - Pre-steady-state average current; Target response current = pre-steady-state average current + total change × response proportionality coefficient; Response time = Target response current at the corresponding time point - Step start time; Adjustment time = Step end time - Step start time; Overshoot = (peak current during step transition - average current before steady state) / total change × 100%, with the maximum value for rising step transitions and the minimum value for falling step transitions.

[0027] In step S5, when the absolute value of the total change is less than 1e-6 amperes, it is determined that there is no effective step, and the index calculation is skipped. When there are fewer than 5 step segment data points, it is marked as insufficient data, and the user is prompted to supplement the data to ensure the validity of the calculation results.

[0028] like Figure 3As shown, the calculation path for performance indicators is as follows: Based on the average current value before and after the steady-state period, calculate the total change; Combined with the response ratio, the target current value is determined by default at 90%, and the time point closest to the target value in the original data is matched to obtain the response time; The adjustment time is calculated by the time difference between the start and end of the step; The overshoot is calculated based on the current peak value and the total change within the step segment.

[0029] The precise positioning of the step jump is achieved through the following steps: Based on the pre-steady-state average current value prev_mean and the post-steady-state average current value next_mean corresponding to the candidate step segment, the total change total_change = next_mean - prev_mean is calculated; if abs(total_change) < 1e-6, it is determined that there is no valid step, and the candidate is skipped. The first-order difference (diff = abs(current[i] - current[i-1])) is calculated for the current column of the candidate data, and the difference sequence is smoothed by the 3-point center rolling average method to obtain diff_smoothed; The step start threshold is set to abs(total_change) × step_threshold_ratio, with a default value of 5%. The step end threshold is set to abs(total_change) × settle_threshold_ratio, with a default value of 2%. Iterating through diff_smoothed, the time corresponding to the first data point that satisfies diff_smoothed ≥ the start threshold is set as step_start; the time following the last data point that satisfies diff_smoothed ≥ the end threshold is set as step_end. If no threshold exceedance is detected, step_start = window start time, and step_end = window end time.

[0030] Output a list of precisely located step segments, including step start time, end time, duration, total change, step direction, and original step segment data; Target response current = prev_mean + total_change × response_ratio, default 90%. In the original data of the step segment, calculate the absolute difference between the current value of each data point and the target response current. Select the optimal data point according to the priority of the smallest difference and the earliest time. Response time = time of the data point - step_start in seconds. It needs to be converted to milliseconds for display. Output a list of performance indicators, including the response time, settling time, overshoot, total change and step direction of each step segment. In step S6, standardized charts are drawn using matplotlib to achieve visualization output. The charts support custom sizes, colors, fonts, and legend positions. The axis labels include physical units: time: s, current: A. Key data labels retain 2-3 decimal places, conforming to the format requirements of power test reports.

[0031] Plot the original current curve and label all effective steady-state segments, such as... Figure 4 As shown, this is a step detection of current data. The colored shading represents the effective steady-state segment, the red shading represents the step segment with a defined position, and the blue dashed line represents the response time point. Add axis labels, legend, and title, and save it as "overall_precise_steps.png". The standardized charts include an overall current curve chart and enlarged charts of each step segment. The charts can be saved in PNG format to a specified path. The chart file name includes the device name and test date information, which facilitates file management and report citation. For each precisely located step segment, draw a separate magnified curve, labeling key information such as the mean value before steady state, the mean value after steady state, the total change, the start / end time of the step, the response time, the settling time, and the overshoot, and save it as "step_XXX_precise.png", where XXX is the step number; The Excel format generates a summary table of performance indicators, including step number, steady-state mean before and after, total change, step direction, response time, settling time, overshoot, and other information, which facilitates report preparation and data archiving.

[0032] Output standardized visualization charts, performance index summary tables, and test report summaries. The test report summaries include basic information about the test data, the number of step segments, and statistical results for each index.

[0033] The current step detection technology based on DBSCAN proposed in this invention can realize a closed-loop process across the entire chain, including data input, preprocessing, steady-state identification, candidate step extraction, precise positioning, performance calculation, and visualization output. After preprocessing, the raw current data is simultaneously distributed to the steady-state identification module and the raw data cache pool. The steady-state identification result triggers candidate step extraction, which is combined with the raw data to complete precise positioning. Finally, the detection result is output through the performance calculation and visualization module, forming an integrated data-analysis-result process.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A precise current step detection method for testing new energy power equipment, characterized in that, Includes the following steps: S1. The data processing module loads the target current test file, automatically identifies the time column and current column, performs data type conversion and data cleaning, and adaptively downsamples to generate downsampled data. S2. The steady-state identification module standardizes the current column of the downsampled data, uses an unsupervised clustering algorithm to cluster the data, and selects the effective steady-state segments. S3. The step candidate extraction module extracts candidate step segments based on the effective steady-state segment and expands the range forward and backward to obtain the final candidate step window. S4, the step precise positioning module performs differential smoothing of the raw current data within the candidate step window, sets a dynamic proportional threshold, and performs precise step positioning. S5. The performance calculation module automatically calculates performance indicators, performs anomaly detection and prompts based on the location results. S6. Visualize the results and save them.

2. The method for accurate detection of current step changes in new energy power equipment according to claim 1, characterized in that: In step S1, the data cleaning includes numerical type conversion, missing value removal, and outlier filtering. The time column and current column are forcibly converted to numerical type, rows containing missing values ​​are removed, and current outliers are identified based on the 3σ principle. The data is then completed using linear interpolation.

3. The method for accurate detection of current step changes in new energy power equipment according to claim 2, characterized in that: In step S1, adaptive downsampling dynamically adjusts the downsampling ratio according to the amount of original data. When the amount of data is greater than 10,000 points, downsampling is performed at a set ratio of 10:

1. When the amount of data is less than or equal to 10,000 points, the original data is retained directly.

4. The method for accurate detection of current step changes in new energy power equipment according to claim 3, characterized in that: In step S2, the unsupervised clustering algorithm uses the DBSCAN algorithm to select segments that are not noisy clusters and whose duration is greater than or equal to a set threshold as valid steady-state segments, and outputs a data list containing the start time, end time and average current value of each steady-state segment.

5. The method for accurate detection of current step changes in new energy power equipment according to claim 4, characterized in that: In step S3, for adjacent effective steady-state segments, the interval from the end time of the previous steady state to the start time of the subsequent steady state is taken as a candidate step segment, and data within the candidate step window is extracted from the original data.

6. The method for accurate detection of current step changes in new energy power equipment according to claim 1, characterized in that: In step S4, differential smoothing employs a 3-point center rolling average method, and the dynamic proportional threshold is calculated based on the average current difference between the preceding and following steady-state segments. Step start threshold = Total change × Step threshold ratio; Step end threshold = total change × step threshold ratio; The step positioning logic is as follows: the time corresponding to the first data point that exceeds the start threshold is the step start time, and the time after the last data point that exceeds the end threshold is the step end time. If no threshold is detected, the candidate window boundary is used as the default value or the step end time.

7. The method for accurate detection of current step changes in new energy power equipment according to claim 1, characterized in that: In step S5, the performance indicators are calculated as follows: Total change = Post-steady-state average current - Pre-steady-state average current; Target response current = pre-steady-state average current + total change × response proportionality coefficient; Response time = Target response current at the corresponding time point - Step start time; Adjustment time = Step end time - Step start time; Overshoot = (peak current during step transition - average current before steady state) / total change × 100%.

8. The method for accurate detection of current step changes in new energy power equipment according to claim 7, characterized in that: In step S5, when the absolute value of the total change is less than 1e-6 amperes, it is determined that there is no effective step, and when there are fewer than 5 step segment data points, it is marked as insufficient data.

9. The method for accurate detection of current step changes in new energy power equipment according to claim 1, characterized in that: In step S6, standardized charts are drawn using matplotlib to achieve visualization output. The charts support custom sizes, colors, fonts, and legend positions.

10. A method for accurate detection of current step changes in new energy power equipment according to claim 9, characterized in that: The standardized charts include an overall current curve chart and magnified charts of each step. The charts can be saved in PNG format to a specified path, and the chart file name includes the device name and test date information.

Images