System side variable harmonic impedance estimation method based on improved BP neural network

Abstract

The invention discloses a system side variable harmonic impedance estimation method based on an improved BP neural network, and the method comprises the following steps: S1, carrying out the sampling of a system, obtaining the harmonic voltage and harmonic current data at a common connection point, carrying out the segmentation processing of the harmonic voltage and harmonic current, and roughly estimating the harmonic impedance value of each segment through binary linear regression; s2, repeating and continuously arranging the harmonic voltage, the harmonic current and the roughly estimated harmonic impedance value according to an original sequence to generate an extended sequence, and performing filtering and normalization processing; s3, building a BP neural network model combined with an attention mechanism, and improving an output layer loss function of the BP neural network model; s4, the processed harmonic voltage and harmonic current serve as input, the processed harmonic impedance serves as output, a BP neural network model is trained and tested, and a system side variable harmonic impedance estimated value is obtained; according to the method, the improved BP neural network is used for estimating the system side variable harmonic impedance, and the method still has high estimation precision under the condition that background harmonic waves are large.

Description

Technical Field

[0001] The invention relates to the field of novel power systems, and in particular to a system side transformer harmonic impedance estimation method based on an improved BP neural network. Background Art

[0002] With the increasing proportion of wind power generation and photovoltaic power generation in my country, as well as the access of various new energy equipment and traditional electrical equipment, the harmonic impedance on the power grid system side will change slowly. Accurately measuring the harmonic impedance on the system side can ensure the stable operation of the power system and power equipment and improve the quality of electric energy. At present, the harmonic impedance estimation methods are mainly divided into two types: "intervention method" and "non-intervention method". Among them, the intervention method injects appropriate harmonic currents into the power grid through manual operation to generate interference to calculate the harmonic impedance, mainly including the switching operation of capacitor groups, etc., but this method may have a certain impact on the system; "non-intervention method" refers to estimating the harmonic impedance on the system side by measuring the harmonic voltage and current of the common connection point of the system. The main methods include regression method, wave quantity method, independent component method, etc., but the above methods have a high level of harmonic emission and the system side harmonic impedance has a low accuracy when the harmonic impedance changes slowly. To this end, the present invention proposes a system side variable harmonic impedance estimation method based on an improved BP neural network. Summary of the invention

[0003] To achieve the above purpose, the technical solution provided by the present invention is:

[0004] A method for estimating harmonic impedance of a system side transformer based on an improved BP neural network comprises the following steps:

[0005] S1: The system samples the harmonic voltage and harmonic current data at the common connection point, processes the harmonic voltage and harmonic current in sections, and roughly estimates the harmonic impedance value of each section using binary linear regression;

[0006] S2: Repeat and continuously arrange the harmonic voltages, harmonic currents and roughly estimated harmonic impedance values ​​in the original order to generate an extended sequence, filter and normalize them;

[0007] S3: Build a BP neural network model combined with the attention mechanism and improve the loss function of the output layer of the BP neural network model;

[0008] S4: The processed harmonic voltage and harmonic current are used as input, and the processed harmonic impedance is used as output to train and test the BP neural network model to obtain the estimated value of the system side transformer harmonic impedance;

[0009] Furthermore, the specific steps of step S1 are as follows:

[0010] S1-1: The system samples the harmonic voltage and harmonic current at the common connection point and performs segmented processing:

[0011] The complex forms of harmonic voltage and harmonic current are obtained by sampling at the common connection point. A set of data includes a sampled harmonic voltage and the corresponding harmonic current. Depending on the number of sampling points, the number of data groups contained in each segment is different:

[0012]

[0013] In formula (1), D is the number of segments, M is the number of sampling points, and m is the number of data sets contained in each segment;

[0014] S1-2: Use binary linear regression to roughly estimate the system side harmonic impedance:

[0015] V pcc =(I pcc +I s )×Z s =I pcc ×Z s +V s (2)

[0016] V pccr =I pccr ×Z sr -I pcci ×Z si +V sr (3)

[0017] V pcci =I pccr ×Z si +I pcci ×Z sr +V si (4)

[0018] In formulas (2)-(4), V pcc with I pcc are the sampled harmonic voltage and harmonic current, I s is the equivalent harmonic current source on the system side, Z s is the system side harmonic impedance, V s is the equivalent harmonic voltage source on the system side, V pccr With V pcci is the harmonic voltage V pcc The real and imaginary parts, V sr With V si is the system side equivalent harmonic voltage source V s The real and imaginary parts, I pccr with I pcci is the harmonic current I pcc The real and imaginary parts, Z sr With Z si is the system side harmonic impedance Z s The real and imaginary parts of

[0019]

[0020] In formula (5), T represents the transpose of the matrix. The rough estimated harmonic impedance value is obtained by solving formula (5); X1, X2, Y1, and Y2 are the solution matrices based on the real and imaginary parts of the harmonic voltage and the real and imaginary parts of the harmonic current. The specific expression is:

[0021]

[0022] Furthermore, the specific steps of step S2 are as follows:

[0023] S2-1: Repeat and continuously arrange the harmonic voltage, harmonic current and roughly estimated harmonic impedance values ​​in the original order of the data segments to generate an extended sequence and filter processing:

[0024] The sampled harmonic voltage and harmonic current data and the roughly estimated harmonic impedance values ​​are repeated ten times in the original order and sorted in the order of front to back. Then, each data sequence is processed by moving average filtering. The moving average filtering is averaged through the window data. The filtering output is:

[0025]

[0026] In formula (7), y is the sequence after moving average filtering, x is the original sequence, j is the sliding window position, and P is the sliding window size;

[0027] S2-2: Normalize the harmonic voltage and harmonic current obtained after repeated filtering and the roughly estimated harmonic impedance value:

[0028]

[0029] In formula (8), L i is the normalized value, X i is the i-th element in the sequence X, X max is the maximum value of the X sequence, X min is the minimum value of the X sequence.

[0030] Furthermore, the specific steps of step S3 are as follows:

[0031] S3-1: Improve the loss function to optimize the estimation results at the actual impedance mutation point:

[0032]

[0033] In formula (9), mseloss is the mean square error, Y is the output impedance value predicted by the BP neural network model, N is the total number of samples, L is the actual harmonic impedance value, diff is the absolute error between the predicted impedance value and the actual impedance value, m is the set threshold, and the value of the absolute error diff greater than the set threshold m is accumulated to obtain the total loss penalty, k is the weight for the penalty, and loss is the improved loss function;

[0034] S3-2: Build a BP neural network model combined with attention mechanism:

[0035] Build a BP neural network structure combined with an attention mechanism, including an input layer, an attention mechanism layer, two hidden layers, and an output layer with an improved loss function; the input layer corresponds to two feature inputs, the number of neurons in the two hidden layers is variable, an attention mechanism can strengthen feature selection, optimize weight distribution and improve training efficiency, and the output layer corresponds to a feature output; the adam optimization algorithm is used to calculate the exponential moving average of the first-order moment and the second-order moment of the gradient at the same time, and the calculation results are corrected for deviations to avoid the gradient estimation biased towards zero in the early stage of training. The implementation process of the adam optimization algorithm is as follows:

[0036]

[0037] In formula (10), q is the number of iterations, g is q is the gradient when the number of iterations is q, m is the gradient first-order moment estimate, v is the gradient second-order moment estimate, θ is the weight, m q-1 、m q 、m q+1 and v q-1 、v q 、v q+1 With θ q-1 ,θ q ,θ q+1 Respectively represent the gradient first-order moment estimate, gradient second-order moment estimate, and weight of the last iteration, this iteration, and the next iteration. is the weight θ when the number of iterations is q-1 q-1 The gradient of β1 is the exponential decay rate of the first-order moment estimate, and β2 is the exponential decay rate of the second-order moment estimate. represents the qth power operation of the exponential decay rate β1 controlling the first-order moment estimate and the exponential decay rate β2 controlling the second-order moment estimate, α is the learning rate, μ is a very small constant to prevent the denominator from being zero, and ⊙ is the vector product operation.

[0038] Furthermore, the specific steps of step S4 are as follows:

[0039] The processed real / imaginary parts of harmonic voltage and harmonic current are used as input features, and the roughly estimated real / imaginary parts of harmonic impedance are used as output features. 90% of the data are used for BP neural network training, and the remaining 10% is used to test the real / imaginary parts of harmonic impedance on the system side.

[0040] Compared with the prior art, the principles and advantages of this solution are as follows:

[0041] The present invention discloses a method for estimating harmonic impedance of a system side transformer based on an improved BP neural network, comprising the following steps: first, the system samples to obtain harmonic voltage and harmonic current data at a common connection point, performs segmented processing on the harmonic voltage and harmonic current, and roughly estimates the harmonic impedance value of each segment by using binary linear regression; second, the harmonic voltage and harmonic current and the roughly estimated harmonic impedance value are repeated and continuously arranged in the original order to generate an extended sequence, and then filtered and normalized; then, a BP neural network model combined with an attention mechanism is built, and the output layer loss function of the BP neural network model is improved; finally, the processed harmonic voltage and harmonic current are used as input, and the processed harmonic impedance is used as output, and the BP neural network model is trained and tested to obtain the estimated value of the harmonic impedance of the system side transformer; the present invention uses the improved BP neural network to estimate the harmonic impedance of the system side transformer, and still has a high estimation accuracy under the condition of large background harmonics. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of a method for estimating harmonic impedance of a system side transformer based on an improved BP neural network in an embodiment of the present invention;

[0043] Figure 2 It is a Norton equivalent circuit diagram in an embodiment of the present invention;

[0044] Figure 3 is the estimation result of the real part of the system-side variable harmonic impedance when the harmonic emission level k=0.1 in the embodiment of the present invention;

[0045] Figure 4 is the estimation result of the imaginary part of the system-side variable harmonic impedance when the harmonic emission level k=0.1 in the embodiment of the present invention;

[0046] Figure 5 is the root mean square error RMSE of the real part estimation of the system side variable harmonic impedance under different harmonic emission levels k between the method of the present invention and the binary linear regression method in the embodiment of the present invention;

[0047] Figure 6 It is the root mean square error RMSE of the imaginary part estimation of the system side variable harmonic impedance under different harmonic emission levels k between the method of the present invention and the binary linear regression method in the embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with specific embodiments:

[0049] Figure 1 The flowchart of the system side transformer harmonic impedance estimation method based on the improved BP neural network is shown, which includes the following steps:

[0050] S1: The system samples the harmonic voltage and harmonic current data at the common connection point, processes the harmonic voltage and harmonic current in sections, and roughly estimates the harmonic impedance value of each section using binary linear regression;

[0051] The specific steps of step S1 are as follows:

[0052] S1-1: The system samples the harmonic voltage and harmonic current at the common connection point and performs segmented processing:

[0053] The complex forms of harmonic voltage and harmonic current are obtained by sampling at the common connection point. A set of data includes a sampled harmonic voltage and the corresponding harmonic current. Depending on the number of sampling points, the number of data groups contained in each segment is different:

[0054]

[0055] In formula (11), D is the number of segments, M is the number of sampling points, and m is the number of data groups contained in each segment;

[0056] S1-2: Use binary linear regression to roughly estimate the system side harmonic impedance:

[0057] V pcc =(I pcc +I s )×Z s =I pcc ×Z s +V s (12)

[0058] V pccr =I pccr ×Z sr -I pcci ×Z si +V sr (13)

[0059] V pcci =I pccr ×Z si +I pcci ×Z sr +V si (14)

[0060] In formulas (12)-(14), V pcc with I pcc are the sampled harmonic voltage and harmonic current, Is is the equivalent harmonic current source on the system side, Z s is the system side harmonic impedance, V s is the equivalent harmonic voltage source on the system side, V pccr With V pcci is the harmonic voltage V pcc The real and imaginary parts, V sr With V si is the system side equivalent harmonic voltage source V s The real and imaginary parts, I pccr with I pcci is the harmonic current I pcc The real and imaginary parts, Z sr With Z si is the system side harmonic impedance Z s The real and imaginary parts of

[0061]

[0062] In formula (15), T represents the transpose of the matrix. The roughly estimated harmonic impedance value is obtained by solving formula (15); X1, X2, Y1, and Y2 are the solution matrices based on the real and imaginary parts of the harmonic voltage and the real and imaginary parts of the harmonic current. The specific expression is:

[0063]

[0064] The following step S2 is to repeat and continuously arrange the harmonic voltage and harmonic current and the roughly estimated harmonic impedance value in the original order to generate an extended sequence, filter and normalize them. The specific process is as follows:

[0065] S2-1: Repeat and continuously arrange the harmonic voltage, harmonic current and roughly estimated harmonic impedance values ​​in the original order of the data segments to generate an extended sequence and filter processing:

[0066] The sampled harmonic voltage and harmonic current data and the roughly estimated harmonic impedance values ​​are repeated ten times in the original order and sorted in the order of front to back. Then, each data sequence is processed by moving average filtering. The moving average filtering is averaged through the window data. The filtering output is:

[0067]

[0068] In formula (17), y is the sequence after moving average filtering, x is the original sequence, j is the sliding window position, and P is the sliding window size;

[0069] S2-2: Normalize the harmonic voltage and harmonic current obtained after repeated filtering and the roughly estimated harmonic impedance value:

[0070]

[0071] In formula (18), L i is the normalized value, X i is the i-th element in the sequence X, X max is the maximum value of the X sequence, X min is the minimum value of the X sequence.

[0072] The following step S3 is to build a BP neural network model combined with the attention mechanism and improve the output layer loss function of the BP neural network model. The specific process is as follows:

[0073] S3-1: Improve the loss function to optimize the estimation results at the actual impedance mutation point:

[0074]

[0075] In formula (19), mseloss is the mean square error, Y is the output impedance value predicted by the BP neural network model, N is the total number of samples, L is the actual harmonic impedance value, diff is the absolute error between the predicted impedance value and the actual impedance value, m is the set threshold, and the value of the absolute error diff greater than the set threshold m is accumulated to obtain the total loss penalty, k is the weight for the penalty, and loss is the improved loss function;

[0076] S3-2: Build a BP neural network model combined with attention mechanism:

[0077] Build a BP neural network structure combined with an attention mechanism, including an input layer, an attention mechanism layer, two hidden layers, and an output layer with an improved loss function; the input layer corresponds to two feature inputs, the number of neurons in the two hidden layers is variable, an attention mechanism can strengthen feature selection, optimize weight distribution and improve training efficiency, and the output layer corresponds to a feature output; the adam optimization algorithm is used to calculate the exponential moving average of the first-order moment and the second-order moment of the gradient at the same time, and the calculation results are corrected for deviations to avoid the gradient estimation biased towards zero in the early stage of training. The implementation process of the adam optimization algorithm is as follows:

[0078]

[0079] In formula (20), q is the number of iterations, g q is the gradient when the number of iterations is q, m is the gradient first-order moment estimate, v is the gradient second-order moment estimate, θ is the weight, m q-1 、m q 、m q+1 and v q-1 、v q 、v q+1 With θ q-1 ,θ q ,θq+1 Respectively represent the gradient first-order moment estimate, gradient second-order moment estimate, and weight of the last iteration, this iteration, and the next iteration. is the weight θ when the number of iterations is q-1 q-1 The gradient of β1 is the exponential decay rate of the first-order moment estimate, and β2 is the exponential decay rate of the second-order moment estimate. represents the qth power operation of the exponential decay rate β1 controlling the first-order moment estimate and the exponential decay rate β2 controlling the second-order moment estimate, α is the learning rate, μ is a very small constant to prevent the denominator from being zero, and ⊙ is the vector product operation.

[0080] The following step S4 is to use the processed harmonic voltage and harmonic current as input, and the processed harmonic impedance as output, train and test the BP neural network model, and obtain the estimated value of the system side transformer harmonic impedance. The specific process is as follows:

[0081] The processed real / imaginary parts of harmonic voltage and harmonic current are used as input features, and the roughly estimated real / imaginary parts of harmonic impedance are used as output features. 90% of the data are used for BP neural network training, and the remaining 10% is used to test the real / imaginary parts of harmonic impedance on the system side.

[0082] In order to verify the effectiveness of the system side transformer harmonic impedance estimation method based on the improved BP neural network, a Figure 2 Norton equivalent circuit, simulation parameters are set as follows: User side harmonic current source I c =40+j60A, real part superimposed with 10% random interference, imaginary part superimposed with 15% random interference; user side harmonic impedance Z c =50+j100Ω, the real part is superimposed with 10% random interference, and the imaginary part is superimposed with 5% random interference; the system side harmonic current source I s The amplitude is the user side harmonic current source I c The amplitude is k times, the harmonic current source I s Amplitude adds 10% random interference; system side harmonic current source I s The phase angle is 56°, and 5% random interference is added; the system side harmonic impedance Z s Adjustments are made in different sampling intervals to simulate sudden and continuous changes in impedance values. In the first 400 sampling points, the system-side harmonic impedance Z is set s =20+j15Ω, set the system side harmonic impedance Z within the sampling points of 401 to 1200 s =Z c / m, m is between 2.981 and 8.944, and the system side harmonic impedance Z is set within the sampling points of 1201 to 1600 s =10+j7.5Ω, during the whole process, the system side harmonic impedance Z sA 2.5% random interference is added to the real part, and a 5% random interference is added to the imaginary part, and a total number of samples M=1600 is obtained by sampling.

[0083] Figure 3 and Figure 4 They are the real and imaginary estimation results of the system side harmonic impedance when the harmonic emission level k=0.1. Through calculation, the real impedance estimation error is about 7.19%, and the imaginary impedance estimation error is about 1.76%. It can be seen that the harmonic impedance estimation results are more accurate and the overall fit is higher.

[0084] In order to further verify that the method of the present invention has a high estimation accuracy, the estimation results of the method of the present invention are compared with the estimation results of the binary linear regression method at different harmonic emission levels k. Figure 5 and Figure 6 As shown in the figure, with the increase of the harmonic emission level, the root mean square error (RMSE) of the real and imaginary part estimation results of the harmonic impedance increases significantly when the binary linear regression method is used; while the real part root mean square error (RMSE) of the harmonic impedance estimated by the method of the present invention is always less than 1.5, and the imaginary part root mean square error (RMSE) is always less than 0.9, which are much smaller than the root mean square error (RMSE) of the binary linear regression method at the same harmonic emission level. This indicates that the method of the present invention still has high estimation accuracy when the harmonic impedance emission level is high and the harmonic impedance on the system side changes.

[0085] The embodiments described above are only preferred embodiments of the present invention and are not intended to limit the scope of implementation of the present invention. Therefore, all changes made according to the shape and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A system side transformer harmonic impedance estimation method based on improved BP neural network, characterized in that: The following steps are involved: S1: The system samples the harmonic voltage and harmonic current data at the common connection point, processes the harmonic voltage and harmonic current in sections, and roughly estimates the harmonic impedance value of each section using binary linear regression; S2: Repeat and continuously arrange the harmonic voltages, harmonic currents and roughly estimated harmonic impedance values ​​in the original order to generate an extended sequence, filter and normalize them; S3: Build a BP neural network model combined with the attention mechanism and improve the loss function of the output layer of the BP neural network model; S4: The processed harmonic voltage and harmonic current are used as input, and the processed harmonic impedance is used as output to train and test the BP neural network model to obtain the estimated value of the system side transformer harmonic impedance; The specific steps of step S1 are as follows: S1-1: The system samples the harmonic voltage and harmonic current at the common connection point and performs segmented processing: The complex forms of harmonic voltage and harmonic current are obtained by sampling at the common connection point. A set of data includes a sampled harmonic voltage and the corresponding harmonic current. Depending on the number of sampling points, the number of data groups contained in each segment is different: In formula (1), D is the number of segments, M is the number of sampling points, and m is the number of data sets contained in each segment; S1-2: Use binary linear regression to roughly estimate the system side harmonic impedance: V pcc =(I pcc +I s )×Z s =I pcc ×Z s +V s (2) V pccr =I pccr ×Z sr -I pcci ×Z si +V sr (3) V pcci =I pccr ×Z si +I pcci ×Z sr +V si (4) In formulas (2)-(4), V pcc with I pcc are the sampled harmonic voltage and harmonic current, I s is the equivalent harmonic current source on the system side, Z s is the system side harmonic impedance, V s is the equivalent harmonic voltage source on the system side, V pccr With V pcci is the harmonic voltage V pcc The real and imaginary parts, V sr With V si is the system side equivalent harmonic voltage source V s The real and imaginary parts, I pccr with I pcci is the harmonic current I pcc The real and imaginary parts, Z sr With Z si is the system side harmonic impedance Z s The real and imaginary parts of In formula (5), T represents the transpose of the matrix. The rough estimated harmonic impedance value is obtained by solving formula (5); X1, X2, Y1, and Y2 are the solution matrices based on the real and imaginary parts of the harmonic voltage and the real and imaginary parts of the harmonic current. The specific expression is:

2. According to the method for estimating harmonic impedance of system side transformer based on improved BP neural network in claim 1, it is characterized in that: The specific steps of step S2 are as follows: S2-1: Repeat and continuously arrange the harmonic voltage, harmonic current and roughly estimated harmonic impedance values ​​in the original order of the data segments to generate an extended sequence and filter processing: The sampled harmonic voltage and harmonic current data and the roughly estimated harmonic impedance values ​​are repeated ten times in the original order and sorted in the order of front to back. Then, each data sequence is processed by moving average filtering. The moving average filtering is averaged through the window data. The filtering output is: In formula (7), y is the sequence after moving average filtering, x is the original sequence, j is the sliding window position, and P is the sliding window size; S2-2: Normalize the harmonic voltage and harmonic current obtained after repeated filtering and the roughly estimated harmonic impedance value: In formula (8), L i is the normalized value, X i is the i-th element in the sequence X, X max is the maximum value of the X sequence, X min is the minimum value of the X sequence.

3. The system side transformer harmonic impedance estimation method based on improved BP neural network according to claim 2 is characterized in that: The specific steps of step S3 are as follows: S3-1: Improve the loss function to optimize the estimation results at the actual impedance mutation point: In formula (9), mseloss is the mean square error, Y is the output impedance value predicted by the BP neural network model, N is the total number of samples, L is the actual harmonic impedance value, diff is the absolute error between the predicted impedance value and the actual impedance value, m is the set threshold, and the value of the absolute error diff greater than the set threshold m is accumulated to obtain the total loss penalty, k is the weight for the penalty, and loss is the improved loss function; S3-2: Build a BP neural network model combined with attention mechanism: Build a BP neural network structure combined with an attention mechanism, including an input layer, an attention mechanism layer, two hidden layers, and an output layer with an improved loss function; the input layer corresponds to two feature inputs, the number of neurons in the two hidden layers is variable, an attention mechanism can strengthen feature selection, optimize weight distribution and improve training efficiency, and the output layer corresponds to a feature output; the adam optimization algorithm is used to calculate the exponential moving average of the first-order moment and the second-order moment of the gradient at the same time, and the calculation results are corrected for deviations to avoid the gradient estimation biased towards zero in the early stage of training. The implementation process of the adam optimization algorithm is as follows: In formula (10), q is the number of iterations, g is q is the gradient when the number of iterations is q, m is the gradient first-order moment estimate, v is the gradient second-order moment estimate, θ is the weight, m q-1 、m q 、m q+1 and v q-1 、v q 、v q+1 With θ q-1 ,θ q ,θ q+1 Respectively represent the gradient first-order moment estimate, gradient second-order moment estimate, and weight of the last iteration, this iteration, and the next iteration. is the weight θ when the number of iterations is q-1 q-1 The gradient of β1 is the exponential decay rate of the first-order moment estimate, and β2 is the exponential decay rate of the second-order moment estimate. represents the qth power operation of the exponential decay rate β1 controlling the first-order moment estimate and the exponential decay rate β2 controlling the second-order moment estimate, α is the learning rate, μ is a very small constant to prevent the denominator from being zero, and ⊙ is the vector product operation.

4. The system side transformer harmonic impedance estimation method based on improved BP neural network according to claim 3 is characterized in that: The specific steps of step S4 are as follows: The processed real / imaginary parts of harmonic voltage and harmonic current are used as input features, and the roughly estimated real / imaginary parts of harmonic impedance are used as output features. 90% of the data are used for BP neural network training, and the remaining 10% is used to test the real / imaginary parts of harmonic impedance on the system side.

Images