CN-121980425-A - Automatic ladder searching and processing method for grid-connected test data of photovoltaic inverter

CN121980425ACN 121980425 ACN121980425 ACN 121980425ACN-121980425-A

Abstract

The invention discloses an automatic ladder searching and processing method for grid-connected test data of a photovoltaic inverter, which belongs to the technical field of photovoltaic power generation system test, and comprises the steps of collecting and preprocessing time sequence data through a power grid simulator and a power analyzer; the method comprises the steps of inputting preprocessed data into a pre-trained convolutional neural network model, automatically outputting the coordinates of a starting point and an ending point of step change, obtaining a trigger signal of a power grid simulator, performing time compensation alignment on the trigger signal and multi-channel data, extracting a steady-state interval of each step platform based on the aligned data and the step point coordinates, calculating a platform value, response time and overshoot, and finally filling the result into a standard test report template to automatically generate a test report. The invention solves the problems of low efficiency, strong subjectivity and easy error caused by relying on manual analysis in the prior art, and remarkably improves the accuracy, consistency and efficiency of the test through full-automatic data processing and machine learning identification.

Inventors

Hu pin
Bao Yuanqiong
LI HUAWEI
ZHENG KAI
LIN XIUYI
LU HUI
Tan huamei
LI JUNHAO
Xi Zhouhui
CHEN ZHIHAO
LENG YINGYING

Assignees

通标标准技术服务有限公司

Dates

Publication Date: 20260505
Application Date: 20260330

Claims (10)

1. The automatic ladder searching and processing method for the grid-connected test data of the photovoltaic inverter is characterized by comprising the following steps of: s1, acquiring time sequence data of grid-connected testing of a photovoltaic inverter through a power grid simulator and a power analyzer, wherein the time sequence data comprises voltage channel data, current channel data and power channel data, performing low-pass filtering processing on the time sequence data, unifying time stamps of all channel data, and generating preprocessed multichannel data; S2, inputting the preprocessed multichannel data into a pre-trained convolutional neural network model, training the convolutional neural network model by using historical ladder test data, and outputting coordinates of a starting point and an ending point of the ladder change; s3, acquiring a trigger signal output by the power grid simulator, detecting the rising edge of the trigger signal and recording a first time stamp of the trigger signal, detecting the starting point of voltage change in the preprocessed multichannel data and recording a second time stamp, calculating the time difference value between the first time stamp and the second time stamp, and performing time compensation alignment on the trigger signal and the preprocessed multichannel data based on the time difference value; S4, extracting a steady-state interval of each step platform based on the multi-channel data after time compensation alignment and the starting point and ending point coordinates of step change, calculating a voltage average value, a current average value and a power average value of the steady-state interval as platform values, and calculating response time and overshoot according to time points corresponding to the starting point coordinates of step change and the platform values; And S5, filling the response time, the overshoot and the platform value into a standard test report template to generate a test report document.
2. The method for automatically searching and processing the step of the grid-connected test data of the photovoltaic inverter according to claim 1, wherein the step S1 specifically comprises the steps of: s101, applying a step-type power grid disturbance test signal to a photovoltaic inverter through a power grid simulator, synchronously acquiring response data of the photovoltaic inverter through a power analyzer at a sampling rate not lower than 10kHz, and acquiring original voltage channel data, original current channel data and original power channel data, wherein the power analyzer and the power grid simulator adopt hardware trigger signals to realize acquisition and starting synchronization; S102, respectively carrying out Hanning window-based finite length unit impulse response low-pass filtering treatment on original voltage channel data, original current channel data and original power channel data, wherein the cut-off frequency of a filter is dynamically adjusted according to the switching frequency of a photovoltaic inverter, and particularly, the cut-off frequency is set to be 0.4 times of the switching frequency and not lower than 2kHz; S103, stamping microsecond-level precision time stamps for the filtered voltage channel data, the current channel data and the power channel data by adopting a time stamp synchronization method based on a precision clock source, wherein the precision clock source adopts a global positioning system clock signal or a high-precision constant-temperature crystal oscillator clock signal; S104, aligning the time-stamped voltage channel data, the current channel data and the power channel data according to time sequences to generate preprocessed multi-channel data, and processing time deviation which is smaller than one sampling period and exists between the channel data by adopting a nearest neighbor interpolation algorithm based on the time stamp when the preprocessed multi-channel data are aligned.
3. The method for automatically searching and processing the step of the grid-connected test data of the photovoltaic inverter according to claim 1, wherein the step S5 specifically comprises: The method comprises the steps of reading a predefined standard test report template file, wherein the standard test report template file adopts an extensible markup language format and comprises a test item number, test time, a device model fixed field and a dynamic data filling position mark, and establishing a mapping relation between response time, overshoot and a platform value and a dynamic data filling position in a template, wherein the response time corresponds to dynamic response characteristics in a test item, the overshoot corresponds to transient characteristics, and the platform value corresponds to steady-state characteristics; The method comprises the steps of adopting a data verification method based on rules, checking verification conditions, normalizing data passing through all the verification conditions according to an international unit system format, reserving three effective digits and adding corresponding unit symbols, and finally filling the normalized data into corresponding positions of a template file by adopting a document object model interface to generate a portable document format test report document comprising complete test data, test waveform screenshot and a data table, and recording generation time, an operator identifier and test environment parameter information in document metadata.
4. The method for automated ladder finding and processing of photovoltaic inverter grid-tie test data according to claim 1, wherein in step S2, The convolutional neural network model adopts an encoder-decoder architecture, wherein an encoder consists of five convolutional layers and is used for extracting multi-scale characteristics, a decoder consists of three transposition convolutional layers and is used for generating time sequence output with the same length as input data, historical ladder test data comprise voltage step, frequency step and power step samples with different amplitudes, gaussian white noise and impulse noise used for simulating an actual test environment are added to each sample, the convolutional neural network model is trained by using a weighted loss function combining a focus loss function and a mean square error loss, the weight coefficient of the focus loss function is set to be 0.7, the weight coefficient of the mean square error loss is set to be 0.3, the convolutional neural network model is output into two time sequences with the same length as the input data, the two time sequences respectively represent ladder starting point probability distribution and ending point probability distribution, and the starting point and ending point coordinates of final ladder change are determined by searching peak points exceeding a 0.8 threshold in the probability distribution.
5. The method for automated step-finding and processing of photovoltaic inverter grid-tie test data according to claim 4, further comprising, after determining and outputting the coordinates of the start point and the end point of the step change, the steps of: The method comprises the steps of carrying out time sequence sequencing on all starting point and ending point coordinates output by a convolutional neural network model, then calculating time intervals between adjacent starting points and ending points, and eliminating coordinate pairs with the time intervals smaller than 10 milliseconds; And (3) performing time constraint optimization on the rest coordinate pairs, pairing each starting point with the nearest ending point, controlling the time interval between the starting point and the ending point in the standard duration range of the step test, and outputting the verified and optimized final coordinate sequence of the starting point and the ending point.
6. The method for automatically searching and processing the step of the grid-connected test data of the photovoltaic inverter according to claim 1, wherein the step S4 specifically comprises: Defining a position from 50 milliseconds after a step starting point to 20 milliseconds before a step ending point as a section to be analyzed for each step platform, calculating a moving standard deviation of voltage data in the section to be analyzed by adopting a sliding window method, wherein the window length is 10 milliseconds, the step length is 1 millisecond, judging that the region enters a steady state when continuous 5 moving standard deviation values are smaller than 0.5 percent of a voltage rated value, expanding the region backwards to 20 milliseconds before the ending point and forwards to 50 milliseconds after the starting point, finally determining the region as a steady state section, respectively calculating an arithmetic average value of the voltage channel data in the steady state section as a voltage average value, an arithmetic average value of the current channel data as a current average value and an arithmetic average value of the power channel data as a power average value.
7. The method for automatically searching and processing the step of the grid-connected test data of the photovoltaic inverter according to claim 6, wherein after calculating the voltage average value, the current average value and the power average value in the steady-state interval, performing the verification of the validity of the platform data, comprising: Calculating the correlation coefficient among the voltage channel data, the current channel data and the power channel data, and when the correlation coefficient of the voltage and the current is lower than 0.95 or the correlation coefficient of the voltage and the power is lower than 0.98, judging that the data consistency among the channels is insufficient and starting a data re-acquisition flow; and for the verified platform data, a global extremum searching method is adopted when the overshoot is calculated, and a voltage maximum deviation value is searched within a time range of 200 milliseconds after a step starting point, wherein the percentage of the difference value between the maximum deviation value and the steady-state voltage average value is the overshoot.
8. The method for automatically searching and processing the grid-connected test data of the photovoltaic inverter according to claim 7, wherein the response time is calculated according to a time point and a platform value corresponding to the coordinates of the starting point of the step change, and the method comprises the following steps: Defining a steady-state error band as a positive and negative 1% range of a steady-state voltage average value, scanning voltage data backwards by taking 1 millisecond as a step length from a step starting point time, searching the time when a voltage value enters the steady-state error band for the first time as initial entering time, continuing scanning from the initial entering time point, judging the initial entering time as effective response time if the voltage data in the subsequent continuous 15 milliseconds are all kept in the steady-state error band, restarting scanning from a jump-out point if the voltage value jumps out of the error band in the scanning process, until a time segment meeting the condition that the voltage value is continuously kept in the error band for 15 milliseconds is found, marking the starting point of the time segment as final response time, and calculating the time difference between the final response time and the step starting point time as a response time result.
9. The method for automatically searching and processing the step of the grid-connected test data of the photovoltaic inverter according to claim 1, wherein the step S3 specifically comprises: The method comprises the steps of adopting a Schmitt trigger circuit to shape a trigger signal output by a power grid simulator, eliminating burrs and oscillations in the trigger signal, detecting rising edges of the shaped trigger signal through a digital logic circuit, recording the accurate time when the rising edges occur as a first time stamp, simultaneously adopting a sliding window variance detection algorithm to calculate variance values of data in a window with 20 sampling points before and after each sampling point in voltage channel data of preprocessed multichannel data, judging as a voltage change starting point when variance values of continuous 5 sampling points exceed a set threshold value, recording the time corresponding to the starting point as a second time stamp, calculating a difference value between the first time stamp and the second time stamp to obtain system transmission delay time, and applying the obtained system transmission delay time as compensation quantity to time stamp correction of all channel data.
10. The automated step lookup and processing method of photovoltaic inverter grid-tie test data of claim 9, further comprising the step of time compensation accuracy verification after time compensation alignment of the trigger signal and the multi-channel data: Selecting a plurality of step change points in the test process, repeatedly executing a time difference value calculation process to obtain a plurality of time difference value samples, calculating the average value and standard deviation of the time difference values, starting a dynamic compensation adjustment mechanism when the standard deviation exceeds 50 microseconds, fitting the change trend of the time difference value by adopting a least square method, establishing a linear correction model of time compensation quantity, applying the corrected time compensation quantity to the time stamp correction of subsequent test data, and simultaneously recording the adjustment quantity and the adjustment time of each time of time compensation to form a time compensation precision log.

Description

Automatic ladder searching and processing method for grid-connected test data of photovoltaic inverter Technical Field The invention relates to the technical field of photovoltaic power generation system testing, in particular to an automatic ladder searching and processing method for grid-connected test data of a photovoltaic inverter. Background The photovoltaic inverter is used as core equipment of a photovoltaic power generation system, and the advantages and disadvantages of grid connection performance of the photovoltaic inverter are directly related to safe and stable operation of a power grid. To evaluate the response characteristics of a photovoltaic inverter in the event of grid anomalies, industry standards specify a series of test projects, with step testing being one of the key projects. The step test applies a sudden change signal of voltage or frequency to the photovoltaic inverter through a power grid simulator, simulates the instantaneous disturbance of a power grid, and records the response waveform of the inverter through a power analyzer. From the response waveforms, the testers need to accurately identify the starting and ending moments of each step change, calculate steady-state parameters (such as voltage, current and power platform values) and dynamic performance indexes (such as response time and overshoot) of the inverter according to the starting and ending moments, and finally form a test report meeting the specification. Currently, there is a general reliance on manual handling in performing such tests and analyses. The testing process is generally as follows, firstly, an operator controls a power grid simulator to send out step disturbance, synchronously starts a power analyzer to record data, then, a large amount of collected time sequence data are exported to general data processing software (such as MATLAB, excel and the like), and finally, an experienced engineer manually marks the starting point and the ending point of step change through naked eyes to observe a waveform chart, and a section which looks stable is selected to calculate parameters. This traditional manual analysis method exposes a number of problems in practical applications. First, there is significant subjectivity in manually identifying the step change points. Different engineers may have different judgments about the beginning and ending moments of a step in the same waveform, and even the same engineer's judgments about the same set of data at different times may deviate. This subjectivity directly results in poor consistency and repeatability of the test results, and it is difficult to meet the high requirements of standardized tests for accuracy and objectivity. Second, the overall process is extremely inefficient. The test of the photovoltaic inverter often needs to be completed into tens or even hundreds of step test points, each test point relates to the processing of massive high-sampling-rate data, the manual analysis is time-consuming and tedious, and the period of product research and development and authentication test is severely restricted. Furthermore, manual operation is extremely prone to errors. When characteristic points are found from complex waveforms possibly containing noise, human eyes are easy to fatigue, tiny changes can be omitted or the noise can be misjudged as an effective signal, and particularly under the condition of low signal-to-noise ratio, the accuracy is difficult to guarantee. While some automated test software exists on the market, its core algorithm relies mostly on simple fixed threshold or slope decisions to detect step changes. When the method faces the common complex situations in actual tests, such as high-frequency noise, signal oscillation and non-ideal step edges which are introduced by the switching frequency of the inverter, the robustness is often insufficient, and false detection or omission detection is easy to generate. In addition, inherent signal transmission and processing delay exists between equipment such as a power grid simulator, a power analyzer and the like in the test system, if accurate time synchronization compensation is not carried out on the equipment, the system error exists in dynamic parameters such as response time and the like of subsequent calculation, and the accuracy of a test result is affected. Accordingly, there is a strong need in the art for an automated solution that overcomes the above-mentioned drawbacks. The ideal method can automatically, objectively and accurately identify the step change, intelligently process the time synchronization problem in the test system and efficiently complete the whole flow from data to report, thereby remarkably improving the reliability, efficiency and standardization level of the grid-connected test of the photovoltaic inverter. Disclosure of Invention The invention solves the problems of low efficiency, strong subjectivity and easy error caused by relying on manual analysis in the prio