A method for analyzing time-varying data delay

Abstract

The present invention provides a method for analyzing time-varying data delay, comprising the following steps: step S1, data preprocessing; step S2, analyzing the coarse delay d0 by cross-correlation; step S3, compensating the coarse delay d0 by coarse phase rotation to obtain a first delay d1; step S4, calculating the phase difference weighted correction ω, and then obtaining the phase P d With amplitude P u ; Step S5, fitting P by least squares method d With P u To estimate the second delay d2; Step S6, calculate the total delay d according to the first delay and the second delay, d = d1 × f s +d2; Step S7, multiple delay analysis, and median filtering. The method of analyzing the time-varying data delay of the present invention can improve the accuracy of delay estimation, eliminate the asynchrony between variables, improve the correlation between variables, and thus accurately predict the location of the fault.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a method for analyzing time-varying data delay in a large wind farm. Background Art

[0002] With the continuous development of modern industry, process technology has become more and more complex, and people pay more and more attention to product quality and safe production. If enterprises want to ensure the safe and smooth operation of the process, improve product quality, and maximize the benefits of the enterprise, they need various types of detection instruments to quickly and accurately obtain a large amount of variable information closely related to the process and product. However, due to technical or economic reasons, these key variables cannot be measured by detection instruments in most cases, or the measurement is time-consuming and has a large lag, so that product quality cannot be adjusted in time, resulting in unqualified product quality, huge economic losses, and the inability to implement automatic control and optimization strategies, resulting in deterioration of process characteristics.

[0003] There is generally a significant time lag between input and output variables in the production process. For example, the transmission delay and volume delay caused by material flow, and the pure delay caused by signal transmission, these delays may cause the causal relationship between the auxiliary variables and the dominant variables measured at the same time to change. At present, in the field of correlation analysis, the main technical indicators used are Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient. These are all to characterize the correlation through the size of the elements of the two sets of data, that is, the nature of increasing and decreasing together. In the field of classification and clustering, in order to make up for the shortcomings of the above correlation, distance and direction are also introduced into the characterization of correlation. Commonly used indicators include Euclidean distance, angle cosine, etc. Although these methods perform well in the classification and clustering process, they still have limitations. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for analyzing time-varying data delay, accurately analyzing the delay, eliminating asynchrony between variables, improving the correlation between variables, and then accurately predicting the location of the fault.

[0005] The present invention is implemented as follows: A method for analyzing time-varying data delay comprises the following steps:

[0006] Step S1, data preprocessing;

[0007] Step S2, analyzing the coarse delay d0 by cross-correlation;

[0008] Step S3, compensating the coarse delay d0 by coarse phase rotation to obtain a first delay d1;

[0009] Step S4: Calculate the phase difference weighted correction ω, and then obtain the phase P d With amplitude Pu ;

[0010] Step S5: Fitting P by least squares method d With P u To estimate the second delay d2;

[0011] Step S6: Calculate the total delay d according to the first delay and the second delay, d = d1 × f s +d2;

[0012] Step S7: perform multiple delay analyses and median filtering.

[0013] Furthermore, the time-varying data is a data pair with associated periodic changes in the frequency domain in the wind turbine equipment, and the data pair includes the winding temperature at the generator input end and the winding temperature at the generator output end, the gearbox speed and the drive end bearing temperature, the wind speed and the gearbox speed, the gearbox speed and the generator speed or the port voltage and the winding temperature.

[0014] Furthermore, the step S1 specifically includes 0-1 normalization and mean filtering processing.

[0015] Furthermore, the step S2 specifically includes: taking 2N groups of data, calculating their cross-correlation values ​​using a cross-correlation function, and taking the time corresponding to the peak value in the cross-correlation value as the coarse delay.

[0016] Furthermore, the step S3 is specifically as follows: using a unit vector with a phase of 0 as a test vector, the phase of the feedback vector of the test vector is the introduced phase rotation, taking the conjugate of the feedback vector, obtaining a phase rotation compensation value d'0, and then obtaining a delay d1 = d0 + d'0.

[0017] Furthermore, the step S4 is specifically as follows:

[0018] Step S41, with sampling frequency f s Continuously sample 2N signals to form a sequence;

[0019] Step S42: Calculate the phase difference weighted correction ω using the following formula:

[0020]

[0021] Among them, x(i) is the spectrum sequence of the first N points of the sequence to form the new sequence, x(j) is the spectrum sequence of the last N points of the sequence to form the new sequence, τ i represents the peak normalization correction of x(i), τ j represents the peak normalization correction of x(j);

[0022] Step S43: Update the correction phase P according to the phase difference weighted correction ω d With amplitude Pu , as follows:

[0023]

[0024]

[0025] Furthermore, the calculation formula of the delay d2 in step S5 is specifically:

[0026] d2=P u tan(P d ) / tan 2 (1 / P d ).

[0027] Furthermore, in the multiple delay analyses, the analysis window size is 10%-20% of the data length.

[0028] The present invention has the following advantages: analyzing from the frequency domain perspective, combining multiple delay analyses to analyze the correlation between variables to estimate the delay, eliminating the asynchrony between variables, improving the correlation between variables, and improving the stability of data results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0030] Figure 1 The present invention is a flowchart of a method for analyzing time-varying data delay.

[0031] Figure 2 FIG. 1 is a schematic diagram of the time delay between the temperature of the generator input winding and the temperature of the generator output winding in a specific embodiment of the present invention, wherein the horizontal axis unit is seconds.

[0032] Figure 3 FIG. 1 is a schematic diagram of the time delay probability between the port voltage and the winding temperature in a specific embodiment of the present invention, wherein the horizontal axis unit is seconds.

[0033] Figure 4 FIG. 1 is a schematic diagram of the probability of time delay between wind speed and gearbox speed in a specific embodiment of the present invention, wherein the horizontal axis unit is seconds.

[0034] Figure 5 FIG. 1 is a schematic diagram of the time delay probability between the gearbox speed and the generator speed in a specific embodiment of the present invention, wherein the horizontal axis unit is seconds.

[0035] Figure 6 FIG. 1 is a schematic diagram of the time delay probability between the gearbox speed and the drive end bearing temperature in a specific embodiment of the present invention, wherein the horizontal axis unit is seconds. DETAILED DESCRIPTION

[0036] like Figure 1As shown, a method for analyzing time-varying data delay of the present invention comprises the following steps:

[0037] Step S1, data preprocessing;

[0038] Step S2, analyzing the coarse delay d0 by cross-correlation;

[0039] Step S3, compensating the coarse delay d0 by coarse phase rotation to obtain a first delay d1;

[0040] Step S4: Calculate the phase difference weighted correction ω, and then obtain the phase P d With amplitude P u ;

[0041] Step S5: Fitting P by least squares method d With P u To estimate the second delay d2;

[0042] Step S6: Calculate the total delay d according to the first delay and the second delay, d = d1 × f s +d2;

[0043] Step S7, multiple delay analysis and median filtering. Multiple delay analysis can make the data results more stable. By median filtering the delay data, the excessive error delay can be smoothed out. For example, we can take the data within 12 hours for median filtering according to the specific situation, and finally we can get the delay between dynamic variables.

[0044] In this embodiment, the time-varying data is a data pair with associated periodic changes in the frequency domain in the wind turbine device, and the data pair includes the generator input winding temperature and the generator output winding temperature, the gearbox speed and the drive end bearing temperature, the wind speed and the gearbox speed, the gearbox speed and the generator speed, or the port voltage and the winding temperature. In fact, the data pair includes more than just these, as long as there are some statistical characteristics associated in the frequency domain, they can be used as the same type of data, that is, the above-mentioned data pair.

[0045] In this embodiment, step S1 specifically includes 0-1 standardization and mean filtering. The mean filtering sampling time is 3 to 7 minutes, and generally 5 minutes is sufficient. Taking a group of wind turbines used in a large wind farm as an example, in the mechanical system of the wind turbine, due to the large amount of wind power data, the instantaneous variable has too many oscillations. When performing mean filtering, 60 data points are taken, and the sampling time is 5 minutes, then the frequency is 5s, which has the least impact on the curve.

[0046] In this embodiment, step S2 specifically includes: taking 2N groups of data, calculating their cross-correlation values ​​using a cross-correlation function, and taking the time corresponding to the peak value in the cross-correlation value as the coarse delay.

[0047] For example, the temperature transmitted to the bearing is defined as x(k), k = 1, 2, 3, ..., N, and the temperature outside the cabin is defined as y(k), k = 1, 2, 3, ..., N. The cross-correlation between x(k) and y(k) is calculated. The cross-correlation function describes the correlation between the values ​​of the two signals x(k) and y(k) at any time. The cross-correlation function is expressed by the following formula:

[0048]

[0049] Among them, τ is a variable. In principle, it is equivalent to keeping the x(k) signal fixed and moving the y(k) signal on the time axis, and calculating the R value under different values ​​of τ in turn. If a certain correlation function R τ (k) If there is a peak at τ = τ0, the time corresponding to the peak is taken as the coarse delay. After the coarse correction, the phase should no longer wrap around, and the linear regression will correctly determine its slope.

[0050] In this embodiment, step S3 is specifically as follows: using a unit vector with a phase of 0 as a test vector, the phase of the feedback vector of the test vector is the introduced phase rotation, taking a conjugate of the feedback vector, obtaining a phase rotation compensation value d'0, and then obtaining a delay d1 = d0 + d'0. For example, in a wind turbine device, the wind turbine blades are directly connected to the generator and the transmission is coaxial, then the phase rotation compensation value can be ignored, and d'0 = 0 is taken at this time.

[0051] In this embodiment, step S4 is specifically as follows:

[0052] Step S41, with sampling frequency f s Continuously sample 2N signals to form a sequence;

[0053] Step S42: Calculate the phase difference weighted correction ω using the following formula:

[0054]

[0055] Among them, x(i) is the spectrum sequence of the first N points of the sequence to form the new sequence, x(j) is the spectrum sequence of the last N points of the sequence to form the new sequence, τ i represents the peak normalization correction of x(i), τ j represents the peak normalization correction of x(j);

[0056] Step S43, calculate the phase P according to the phase difference weighted correction ω d With amplitude P u , as follows:

[0057]

[0058]

[0059] By performing weighted correction on the phase difference, the problem of uneven error distribution in the entire spectrum interval can be smoothed out, further improving the correction accuracy.

[0060] In this embodiment, the calculation formula of the delay d2 in step S5 is specifically:

[0061] d2=P u tan(P d ) / tan 2 (1 / P d ).

[0062] The least squares solution is found by projecting the phase onto a basis formed by a constant term and a linear phase term, the latter corresponding to unit delay. When the phase has been weighted, the same weight region corresponds to the basis. The resulting coefficient of the linear phase term is the residual delay in samples, which is summed with the coarse delay to give the final delay estimate.

[0063] In this embodiment, in the multiple delay analyses, the analysis window size is 10%-20% of the data length. By using a sliding window to perform multiple delay analyses, the dynamic delay becomes a static delay, which facilitates the stable calculation of the results. In particular, the size of the sliding window is about 10%-20% of the data. In this example, the wind power data generates about 10,000 data per day, and we can use 20 minutes as the window size.

[0064] According to the data obtained in this example, after analyzing them using the above delay analysis method, the following results can be obtained: Figure 2 As shown, there is essentially no time delay between the generator input winding temperature (i.e., winding temperature 1) and the generator output winding temperature (i.e., winding temperature 2). Figure 3 As shown in Figure 2, the time delay between the terminal voltage and the winding temperature is approximately two minutes. Figure 4 As shown in Figure 2, the time delay between wind speed and gearbox speed is about one minute. Figure 5 As shown in Figure 1, there is essentially zero time delay between the gearbox speed and the generator speed. Figure 6 As shown, the time delay between the gearbox speed and the drive end bearing temperature is approximately 4.5 minutes.

[0065] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for analyzing time-varying data delay, characterized in that: The steps include: Step S1, data preprocessing; Step S2: Analyze the coarse delay by cross-correlation function ; Step S3: Coarse delay by coarse phase rotation Compensation for the first delay ; Step S4: Calculate phase difference weighted correction , and then get the phase With amplitude ; Step S5: Fitting by least squares method and To estimate the second delay ; Step S6: Calculate the total delay based on the first delay and the second delay d , ;in, refers to the sampling frequency; Step S7: perform multiple delay analyses and median filtering.

2. A method for analyzing time-varying data delay according to claim 1, characterized in that: The time-varying data is a pair of data with associated periodic changes in the frequency domain in the wind turbine equipment, and the data pair includes the generator input winding temperature and the generator output winding temperature, the gearbox speed and the drive end bearing temperature, the wind speed and the gearbox speed, the gearbox speed and the generator speed or the port voltage and the winding temperature.

3. The method for analyzing time-varying data delay according to claim 1, characterized in that: The step S1 specifically includes 0-1 normalization and mean filtering processing.

4. The method for analyzing time-varying data delay according to claim 1, characterized in that: The step S2 specifically includes: taking 2N groups of data, calculating their cross-correlation values ​​using a cross-correlation function, and taking the time corresponding to the peak value in the cross-correlation value as the coarse delay.

5. The method for analyzing time-varying data delay according to claim 1, characterized in that: The step S3 specifically includes: using a unit vector with a phase of 0 as a test vector, the phase of the feedback vector of the test vector is the introduced phase coarse rotation, taking a conjugate of the feedback vector, and obtaining a phase coarse rotation compensation value , and thus get the delay .

6. The method for analyzing time-varying data delay according to claim 1, characterized in that: The step S4 is specifically as follows: Step S41, with sampling frequency Continuously sample 2N signals to form a sequence; Step S42: Calculate the phase difference weighted correction using the following formula: : ,i=1,2,…,N,j=N+1,…,2N; Among them, x(i) is the first The points form the spectrum sequence of the new sequence, and x(j) is the sequence after points constitute the spectrum sequence of the new sequence, τ i represents the peak normalization correction of x(i), τ j represents the peak normalization correction of x(j); Step S43: Weighted correction based on phase difference Update correction phase With amplitude , as follows: ； 。 7. The method for analyzing time-varying data delay according to claim 1, characterized in that: The calculation formula of the delay d2 in step S5 is specifically: 。 8. The method for analyzing time-varying data delay according to claim 1, characterized in that: In the multiple delay analyses, the analysis window size is 10%-20% of the data length.

Images