Photovoltaic module I-V feature extraction method based on linear regression

Abstract

The invention discloses a photovoltaic module I-V feature extraction method based on linear regression, and the method comprises the steps: carrying out the preprocessing of the current and voltage data of a photovoltaic module, and generating an equivalent I-V curve with a plurality of voltage data points at equal intervals through a smooth spline method; establishing a judgment criterion of change points according to noise parameters of the I-V curve, identifying the change points generating steps, and decomposing the equivalent I-V curve into a plurality of single-step I-V curves based on the searched change points; linear regression processing is carried out on each single-step curve, only continuous data points with small slope change are adopted in the process, then the selected linear data points are used for correcting a nonlinear area, a complete linear model is obtained, and corresponding I-V feature values are calculated; and finally, carrying out repeatability test on the extracted I-V features by adopting a resampling method. According to the method, the features are extracted from the I-V curve through the methods of smooth spline, linear regression and the like, the repeatability of extraction of each I-V feature is detected, and the method has high reliability and accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of photovoltaic power generation, and in particular relates to a method for extracting IV features of photovoltaic components based on linear regression. Background Art

[0002] Photovoltaic modules will degrade during operation, resulting in system performance degradation and even possible safety accidents. It is very important to effectively diagnose the status of photovoltaic modules; characteristic parameters such as short-circuit current, open-circuit voltage, parallel resistance, series resistance and fill factor are crucial for diagnosing the performance and degradation of photovoltaic modules. Therefore, the extraction method of IV characteristics of photovoltaic modules has important research value.

[0003] Traditional methods for extracting IV characteristics of photovoltaic modules usually obtain these performance characteristics by fitting a diode model to a single IV curve, which involves solving a transcendental equation and requires the use of Lambert W functions to obtain explicit analytical solutions, or using iterative numerical methods, which are time-consuming and require manual settings. As the number of IV curves that need to be analyzed increases, traditional fitting methods become computationally difficult. At the same time, conventional diode models are not always suitable for describing photovoltaic modules. When there is local shadowing in the photovoltaic array or the photovoltaic cells are degraded and damaged, multiple steps will be observed in the IV curve of the photovoltaic module, which does not conform to the diode model. Therefore, it is necessary to design a method for extracting IV characteristics of photovoltaic modules based on linear regression to solve the above problems. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for extracting IV features of photovoltaic modules based on linear regression. Features are extracted from IV curves by smoothing splines and linear regression, and the extracted IV features are tested for repeatability by resampling. The invention aims to solve the problem of extracting IV features of photovoltaic modules, that is, how to achieve efficient and accurate IV feature extraction of photovoltaic modules when the data set is large and the operating conditions are challenging. By combining methods such as smoothing splines and linear regression, the method is expected to significantly improve the efficiency of IV feature extraction of photovoltaic modules, thereby promoting the development and application of renewable energy technology.

[0005] In order to achieve the above technical effects, the technical solution adopted by the present invention is: A photovoltaic module IV feature extraction method based on linear regression, comprising: S1, preprocessing the collected raw data of PV module current and voltage, and using the smoothing spline method to generate an equivalent IV curve with 1000 equally spaced voltage data points; S2, establishing a criterion for determining the change point based on the noise parameters of the IV curve and identifying the change point that produces the step; S3, decomposing the equivalent IV curve into multiple single-step IV curves based on the identified change points; S4, using continuous data points with small slope changes to perform linear regression on each single-step curve, and then using the selected linear data points to correct the nonlinear region, obtain a complete linear model and calculate the corresponding IV eigenvalue; S5, the resampling method was used to test the repeatability of the extracted IV eigenvalues.

[0006] Preferably, in step S1, generating an equivalent IV curve using a smoothing spline method comprises: An equivalent IV curve with 1000 equally spaced voltage observation points was generated using the smoothing spline method on the original IV curve to provide enough data points to calculate the features in the curve and reduce the statistical uncertainty; Let { V i , I i : i = 1, 2, .., n} means n A collection of observations, f( ) To fit the function of the observed data, the smoothing spline is the function f that minimizes equation (1);

[0007] in is a non-negative adjustment parameter that controls the roughness of the smoothing spline.

[0008] Preferably, in step S2, establishing a change point determination criterion and identifying a change point that generates a step includes: S201, setting constraint conditions: the change point between the steps is a point where the slope on the left is steeper than the slope on the right, and the slope on the left is a negative value; the difference between the absolute values ​​of the slopes on the left and right sides of the change point is greater than a set threshold; S202, remember β1 is the slope to the left of the change point, β2 is the slope to the right of the change point, a This parameter is determined by the noise of the IV curve. For larger noise, set a larger a ; S203, establishing criteria to identify change points representing multiple steps: |β1|>|β2|; β1<0; |β1|-|β2|>a.

[0009] Preferably, in step S3, decomposing the equivalent IV curve into a plurality of single-step IV curves comprises: In some IV curves, it is observed that n The existence of steps is an important indicator for diagnosing local shadows or component degradation or damage, but it is difficult for conventional diode models to fit this curve. Based on the change point judgment criterion established in step S2, the step generating part is identified. n -1 change point, and then decompose the generated equivalent IV curve into n A single-step IV curve.

[0010] Preferably, in step S4, the steps of performing linear regression processing on each single-step curve, improving the linear model and calculating the corresponding IV characteristic value include: S401, calculates the maximum power by finding the maximum product of current and voltage at each of the 1000 IV data points P mp : P mp = max i (V i × I i ); Where i represents the i-th data point, I represents current, and V represents voltage; S402, perform linear regression on the spline IV curve of 1000 points, and establish an initial linear regression model as shown below: Y = α + β × X + θ; Among them, X is the independent variable, Y is the dependent variable, α and β are coefficients, and θ is the error term; S403, regression is performed on a moving window of 10 consecutive points from left to right, with current as the dependent variable and voltage as the independent variable, and the slope of each 10-point window is stored. The slope changes dramatically when passing through the maximum power point in the IV curve and then stops. The slope changes rapidly when approaching 0 V. Due to poor module connection or recording errors in the IV curve tracking system, it should be removed from the IV curve. Therefore, continuous data points with smaller slope changes are used to exclude the curvature of the low voltage area. The selected linear data points are used to correct the nonlinear area to obtain a more accurate linear model. The linear regression model formula is as follows: I = I sc - (1 / R sh ) × V; Short circuit current I sc is the linear intercept, where V = 0, and the parallel resistance R sh Determined by the slope; S404, on the side where the voltage is higher than the maximum power point, a regression model is established with the voltage as the dependent variable and the current as the independent variable; the current change rate is set to be the difference between the currents of two consecutive data points divided by the current with the lower voltage, a critical value is set, and the data point with the current change rate greater than the critical value is selected; the linear regression model formula is as follows: V = V oc -R sh × I; The open circuit voltage V oc is the linear intercept, where I = 0, and the series resistance Rs is determined by the slope; S405, fill factor FF is defined as the maximum power of a solar cell and V oc and I sc Ratio of products: ; In the formula, is the maximum power; is the short-circuit current; is the open circuit voltage.

[0011] Preferably, in step S5, performing repeatability test on the extracted IV features by using a resampling method comprises: For each iteration, 90% of the data points were randomly selected and the IV feature results were obtained by the extraction method of steps S1 to S4; this was repeated 10,000 times to obtain 10,000 values ​​for each extracted IV feature; the standard deviation SD of all 10,000 iterations was calculated for each IV feature, and repeatability was defined as 100%-SD%.

[0012] The beneficial effects of the present invention are as follows: 1. The present invention generates an equivalent IV curve with equally spaced voltage observation points by smoothing spline method, so that there are enough data points to calculate the characteristics in the curve and reduce statistical uncertainty; 2. The present invention identifies the change point that generates the step by establishing a change point determination criterion, and decomposes the equivalent IV curve into multiple single-step IV curves based on the change point, thereby improving the utilization rate of the original data; 3. The present invention performs linear regression processing on each single-step curve and corrects the curvature problem at low voltage, thereby more accurately defining the IV linear part at low voltage and having a higher accuracy in the feature extraction process; 4. The present invention adopts the resampling method to test the repeatability of the extracted IV features, thereby improving the authenticity of the model accuracy estimation; 5. This method can adapt to the feature extraction of PV modules under different environmental conditions and operating conditions and can be applied to a large amount of IV data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a framework diagram of the photovoltaic module IV feature extraction method based on linear regression in the present invention; Figure 2 It is a standard single-step equivalent IV curve diagram used for feature extraction in the present invention; Figure 3 It is a standard multi-step equivalent IV curve diagram used for feature extraction in the present invention; Figure 4 It is a slope change diagram of linear regression based on the standard single-step equivalent IV curve in the present invention; Figure 5 This is an IV curve diagram showing curvature fluctuation in the low-pressure region of the IV curve due to module connection problems in the present invention.

[0014] Figure 6 It is a graph of the IV curve after the curvature fluctuation in the low-pressure area of ​​the IV curve due to the module connection problem in the present invention is repaired. DETAILED DESCRIPTION

[0015] Embodiment 1: like Figure 1 As shown, a photovoltaic module IV feature extraction method based on linear regression includes: S1, preprocessing the collected raw data of PV module current and voltage, and using the smoothing spline method to generate an equivalent IV curve with 1000 equally spaced voltage data points; S2, establishing a criterion for determining the change point based on the noise parameters of the IV curve and identifying the change point that produces the step; S3, decomposing the equivalent IV curve into multiple single-step IV curves based on the identified change points; S4, using continuous data points with small slope changes to perform linear regression on each single-step curve, and then using the selected linear data points to correct the nonlinear region, obtain a complete linear model and calculate the corresponding IV eigenvalue; S5, the resampling method was used to test the repeatability of the extracted IV eigenvalues.

[0016] Preferably, in step S1, generating an equivalent IV curve using a smoothing spline method comprises: An equivalent IV curve with 1000 equally spaced voltage observation points was generated using the smoothing spline method on the original IV curve to provide enough data points to calculate the features in the curve and reduce the statistical uncertainty; Let { V i , I i : i = 1, 2, .., n} means nA collection of observations, f( ) To fit the function of the observed data, the smoothing spline is the function f that minimizes equation (1);

[0017] in is a non-negative adjustment parameter that controls the roughness of the smoothing spline.

[0018] Preferably, in step S2, establishing a change point determination criterion and identifying a change point that generates a step includes: S201, setting constraint conditions: the change point between the steps is a point where the slope on the left is steeper than the slope on the right, and the slope on the left is a negative value; the difference between the absolute values ​​of the slopes on the left and right sides of the change point is greater than a set threshold; S202, remember β1 is the slope to the left of the change point, β2 is the slope to the right of the change point, a This parameter is determined by the noise of the IV curve. For larger noise, set a larger a ; S203, establishing criteria to identify change points representing multiple steps: |β1|>|β2|; β1<0; |β1|-|β2|>a.

[0019] Preferably, in step S3, decomposing the equivalent IV curve into a plurality of single-step IV curves comprises: In some IV curves, it is observed that n The existence of steps is an important indicator for diagnosing local shadows or component degradation or damage, but it is difficult for conventional diode models to fit this curve. Based on the change point judgment criterion established in step S2, the step generating part is identified. n -1 change point, and then decompose the generated equivalent IV curve into n A single-step IV curve.

[0020] Preferably, in step S4, the steps of performing linear regression processing on each single-step curve, improving the linear model and calculating the corresponding IV characteristic value include: S401, calculates the maximum power by finding the maximum product of current and voltage at each of the 1000 IV data points P mp : P mp = max i (V i × I i ); Where i represents the i-th data point, I represents current, and V represents voltage; S402, perform linear regression on the spline IV curve of 1000 points, and establish an initial linear regression model as shown below: Y = α + β × X + θ; Among them, X is the independent variable, Y is the dependent variable, α and β are coefficients, and θ is the error term; S403, regression is performed on a moving window of 10 consecutive points from left to right, with current as the dependent variable and voltage as the independent variable, and the slope of each 10-point window is stored. The slope changes dramatically when passing through the maximum power point in the IV curve and then stops. The slope changes rapidly when approaching 0 V. Due to poor module connection or recording errors in the IV curve tracking system, it should be removed from the IV curve. Therefore, continuous data points with smaller slope changes are used to exclude the curvature of the low voltage area. The selected linear data points are used to correct the nonlinear area to obtain a more accurate linear model. The linear regression model formula is as follows: I = I sc - (1 / R sh ) × V; Short circuit current I sc is the linear intercept, where V = 0, and the parallel resistance R sh Determined by the slope; S404, on the side where the voltage is higher than the maximum power point, a regression model is established with the voltage as the dependent variable and the current as the independent variable; the current change rate is set to be the difference between the currents of two consecutive data points divided by the current with the lower voltage, a critical value is set, and the data point with the current change rate greater than the critical value is selected; the linear regression model formula is as follows: V = V oc -R sh × I; The open circuit voltage V oc is the linear intercept, where I = 0, and the series resistance Rs is determined by the slope; S405, fill factor FF is defined as the maximum power of a solar cell and V oc and I sc Ratio of products: ; In the formula, is the maximum power; is the short-circuit current; is the open circuit voltage.

[0021] Preferably, in step S5, performing repeatability test on the extracted IV features by using a resampling method comprises: For each iteration, 90% of the data points were randomly selected and the IV feature results were obtained by the extraction method of steps S1 to S4; this was repeated 10,000 times to obtain 10,000 values ​​for each extracted IV feature; the standard deviation SD of all 10,000 iterations was calculated for each IV feature, and repeatability was defined as 100%-SD%.

[0022] Embodiment 2: This embodiment provides a specific method for extracting IV characteristics of a photovoltaic module, including the following steps: Step 1: Preprocess the collected raw data of PV module current and voltage. In most PV module IV curves, the observation points are evenly spaced in voltage; however, when the voltage is close to V oc When the current decays exponentially, the pseudo-linear region is close to V oc There are very few points, which may be in the estimation V oc and R s In addition, some IV tracers P mp More data points were obtained near the ideal operating point to more accurately determine the ideal operating point. I sc and V oc Fewer points are obtained in the pseudo linear region near the original IV curve. Therefore, a smoothing spline method is used on the original IV curve to generate an equivalent IV curve with 1000 equally spaced voltage observation points to provide enough data points to calculate the features in the curve and reduce the statistical uncertainty, such as Figure 2 As shown, Figure 2 It is a standard single-step IV equivalent curve and five IV characteristics: I sc , R sh , P mp , V oc and R s ; make{ V i ，I i :i = 1, 2, .., n}express n A collection of observations, f ( ) To fit the function of the observed data, the smoothing spline is the function that minimizes the following equation f ;

[0023] in is a non-negative adjustment parameter that controls the roughness of the smoothing spline.

[0024] Step 2: Establish the judgment criteria of the change point according to the noise parameters of the IV curve and identify the change point that produces the step. The specific steps include: The change point between the steps is the point where the slope on the left is steeper than the slope on the right, and the slope on the left is negative; in addition, the difference in the absolute value of the slope on the left and right sides of the point should be large enough; β1 is the slope to the left of the change point, β2 is the slope to the right of the change point, a This parameter is determined by the noise of the IV curve. For larger noise, set a larger a ; The following criteria are established to identify change points representing multiple steps: |β1|>|β2|; β1<0; |β1|−|β2|>a.

[0025] Step 3: Decompose the equivalent IV curve into multiple single-step IV curves based on the found change points, such as Figure 3 As shown, Figure 3 This is an example of a multi-step IV equivalent curve. The specific steps include: In some IV curves, it is observed that n The existence of steps is an important indicator for diagnosing local shadows or component degradation or damage, but it is difficult for conventional diode models to fit this curve. Based on the change point judgment criterion established in step 2, the step-generating n- 1 change point, and then decompose the generated equivalent IV curve into n A single-step IV curve.

[0026] Step 4: Perform linear regression processing on each single-step curve, use continuous data points with small slope changes, and then use the selected linear data points to correct the nonlinear region, obtain a complete linear model and calculate the corresponding IV characteristic value. The specific steps include: S401, calculates the maximum power by finding the maximum product of current and voltage at each of the 1000 IV data points P mp , the formula is as follows: P mp = max i (V i × I i ); S402, perform linear regression on the 1000-point spline IV curve to establish an initial linear regression model as shown below: Y = α + β × X + θ; in, X is the independent variable, Y is the dependent variable, α and β is the coefficient, θ is the error term; S403, perform regression on a moving window of 10 consecutive points from left to right, with current as the dependent variable and voltage as the independent variable, and store the slope of each 10-point window. Figure 4 The slope change diagram of the linear regression based on the standard single-step equivalent IV curve is given by Figure 4 It can be seen that when passing through the maximum power point of the IV curve, the slope changes dramatically, so when the regression reaches the maximum power point, it stops immediately; sometimes due to poor module connection or IV curve tracking system recording errors, the slope changes rapidly when approaching 0 V. Figure 5 As shown, the points in this section should be removed from the IV curve, and the continuous data points with smaller slope changes should be retained. Then, the nonlinear region should be corrected using the selected linear data points to obtain a more accurate linear model. Figure 6 is the corrected IV curve; The linear regression model formula is as follows: I = I sc - (1 / R sh ) × V; Short circuit current I sc is the linear intercept, where V = 0, parallel resistance R sh estimated by the slope; S404, on the side where the voltage is higher than the maximum power point, establish a regression model with the voltage as the dependent variable and the current as the independent variable; let the current change rate be the difference between the currents of two consecutive data points divided by the current with the lower voltage, set a critical value, and select the data point whose current change rate is greater than the critical value; The linear regression model formula is as follows: V = V oc -R sh × I; Open circuit voltage V oc is the linear intercept, where I = 0, series resistance Rs estimated by the slope; S405, fill factor FF Defined as the maximum power of the solar cell andV oc and I sc The ratio of products is given by the following formula: ; Where: is the maximum power; is the short-circuit current; is the open circuit voltage; Step 5: For each iteration, randomly select 90% of the data points and apply the extraction method to obtain the IV feature results; repeat 10,000 times to obtain 10,000 values ​​for each extracted IV feature. Calculate the standard deviation SD of all 10,000 iterations for each IV feature, and repeatability is defined as 100%-SD%.

[0027] The IV curve generated by the single diode battery model is subjected to IV feature extraction based on linear regression to verify the IV feature extraction method proposed in this paper: The single diode cell model assumes that the dark current can be expressed using the diode ideality factor n The current-voltage relationship is described by a modified single exponential dependence. ; Where: V and I are the terminal voltage in volts and the current in amperes, respectively, I ph ≈ I sc is the photogenerated current, I 0 is the diode reverse saturation current, V th is the thermal voltage.

[0028] The following is an implicit transcendental equation for I and V , which cannot be solved explicitly using general elementary functions in general; therefore, I and V The exact explicit analytical solution of can be expressed in terms of the Lambert W function, defined as the solution to the equation W (x) exp [ W(x) ] = x as follows: ; ; in: ; WRepresents the Lambert W function. To illustrate the robustness of the IV feature extraction method proposed in this method, an IV curve with 1000 observations was generated based on a single diode battery model. Then, the IV parameters were calculated using this method, including I sc , V oc , R sh and R s Let N c is the number of units, included in the Vth unit in the above formula, Temp is the temperature; set N c = 60, Temp = 25°C, Voc = 40.20 V, n = 1.5, Isc = 8A, Rsh = 600Ω, Rs = 0.48Ω, the IV curve is generated by the above formula, where V is the voltage from 0 to V oc For the I value, random noise is added, which obeys a normal distribution with a mean of zero and different SDs, as shown in Table 1: Table 1: Random noise follows a normal distribution with zero mean and different SDs;

[0029] Table 1 generates IV curves with noise levels between 0 and 0.02 A, which is a typical noise level for real-world IV curves; Table 1 shows the percentage error and repeatability of the four IV parameters extracted using the IV curves generated using the diode model at different noise levels. I sc and V oc The percentage error is low, which shows that the feature extraction method proposed in this method is I sc and V oc It performs well in feature estimation. R sh As the noise level increases, the percentage error also increases, and when the noise is 0.015 A on the simulation curve, the accuracy is significantly reduced. R s , because the extracted value is higher than the set value, the percentage error exceeds 61%. However, all the extracted values ​​are calculated with a repeatability greater than 99.9%. Therefore, the data-driven IV curve feature extraction method is a robust, practical, and easy-to-implement IV curve feature extraction method.

[0030] In summary, the photovoltaic module IV feature extraction method based on linear regression proposed in the present invention can adapt to the application requirements of actual scenarios, correct the low voltage curvature problem, and realize efficient and reliable photovoltaic module IV feature extraction when the data set is large.

Claims

1. A method for extracting IV features of photovoltaic modules based on linear regression, characterized in that: include: S1, preprocessing the collected raw data of PV module current and voltage, and using the smoothing spline method to generate an equivalent IV curve with multiple equally spaced voltage data points; S2, establishing a criterion for determining the change point based on the noise parameters of the IV curve and identifying the change point that produces the step; S3, decomposing the equivalent IV curve into multiple single-step IV curves based on the identified change points; S4, using continuous data points with small slope changes to perform linear regression on each single-step curve, and then using the selected linear data points to correct the nonlinear region, obtain a complete linear model and calculate the corresponding IV eigenvalue; S5, the resampling method was used to test the repeatability of the extracted IV eigenvalues.

2. A photovoltaic module IV feature extraction method based on linear regression according to claim 1, characterized in that: In step S1, generating an equivalent IV curve using a smoothing spline method includes: Using the smoothing spline method on the original IV curve, an equivalent IV curve with multiple equally spaced voltage observation points is generated to provide enough data points to calculate the features in the curve and reduce the statistical uncertainty; Let {Vi, Ii: i = 1, 2, .., n} denote n A collection of observations, f( ) To fit the function of the observed data, the smoothing spline is the function f that minimizes equation (1); in is a non-negative adjustment parameter that controls the roughness of the smoothing spline.

3. A photovoltaic module IV feature extraction method based on linear regression according to claim 1, characterized in that: In step S2, establishing a change point determination criterion and identifying the change point that generates the step includes: S201, setting constraint conditions: the change point between the steps is a point where the slope on the left is steeper than the slope on the right, and the slope on the left is a negative value; the difference between the absolute values ​​of the slopes on the left and right sides of the change point is greater than a set threshold; S202, remember β1 is the slope to the left of the change point, β2 is the slope to the right of the change point, a is a parameter determined by the noise of the IV curve; S203, establishing criteria to identify change points representing multiple steps: |β1| > |β2|； β1 < 0； |β1|-|β2| >a。 4. A photovoltaic module IV feature extraction method based on linear regression according to claim 3, characterized in that: In step S3, decomposing the equivalent IV curve into a plurality of single-step IV curves includes: Based on the change point determination criteria established in step S2, the step-generating n -1 change point, and then decompose the generated equivalent IV curve into n A single-step IV curve.

5. The method for extracting photovoltaic module IV features based on linear regression according to claim 1, characterized in that: In step S4, the steps of performing linear regression processing on each single-step curve, improving the linear model and calculating the corresponding IV characteristic value include: S401, calculates the maximum power by finding the maximum product of the current and voltage at each point in the plurality of IV data points P mp : P mp = max i (V i × I i ); Where i represents the i-th data point, I represents current, and V represents voltage; S402, performing linear regression on the spline IV curves of multiple points, and establishing an initial linear regression model as shown below: Y = α + β × X + θ; Among them, X is the independent variable, Y is the dependent variable, α and β are coefficients, and θ is the error term; S403, regressing on a moving window of m consecutive points from left to right, with current as the dependent variable and voltage as the independent variable, and storing the slope of each m-point window, the slope changes dramatically when passing through the maximum power point of the IV curve, and then stops; using the selected linear data points, the nonlinear area is corrected to obtain a more accurate linear model; the linear regression model formula is as follows: I = I sc - (1 / R sh ) × V; Short circuit current I sc is the linear intercept, where V = 0, and the parallel resistance R sh Determined by the slope; S404, on the side where the voltage is higher than the maximum power point, a regression model is established with the voltage as the dependent variable and the current as the independent variable; the current change rate is set to be the difference between the currents of two consecutive data points divided by the current with the lower voltage, a critical value is set, and the data point with the current change rate greater than the critical value is selected; the linear regression model formula is as follows: V = V oc - R sh × I; The open circuit voltage V oc is the linear intercept, where I = 0, and the series resistance Rs is determined by the slope; S405, fill factor FF is defined as the maximum power of a solar cell and V oc and I sc Ratio of products: ； In the formula, is the maximum power; is the short-circuit current; is the open circuit voltage.

6. The method for extracting IV features of photovoltaic modules based on linear regression according to claim 1, characterized in that: In step S5, the resampling method is used to perform repeatability test on the extracted IV features, including: For each iteration, 90% of the data points are randomly selected and the IV feature results are obtained by the extraction method of steps S1 to S4; this is repeated N times to obtain N values ​​for each extracted IV feature; the standard deviation SD of all N iterations is calculated for each IV feature, and repeatability is defined as 100%-SD%.

Images