Data mechanism fusion prediction system and method for unbalanced transformer loss of distribution network

Abstract

The invention provides a data mechanism fusion prediction system and method for loss of an unbalanced transformer of a distribution network, and relates to the technical field of operation and loss reduction of the distribution network. Firstly, electric quantity data such as voltage, current, active power and reactive power of a power distribution network are collected to form a matrix to form an original data set, correlation analysis is carried out on data in the original data set by adopting a mutual information method, and dimensionality reduction is carried out by adopting a linear discriminant analysis method. Taking the output after dimension reduction as the input of transformer loss intelligent prediction, calculating the zero-sequence impedance corresponding to the original data set to obtain the total loss of the transformer under the unbalanced load of the distribution network, and training the transformer loss intelligent prediction model by using the output after dimension reduction and the corresponding total loss of the transformer to obtain the transformer loss intelligent prediction model. And finally, the real-time intelligent prediction of the transformer loss under the unbalanced load of the distribution network can be realized by using the real-time monitoring data of the distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network operation and loss reduction, and particularly to a data mechanism fusion prediction system and method for unbalanced transformer losses in a distribution network. Background Art

[0002] Calculation of transformer losses under unbalanced loads in a distribution network has gradually become one of the most important technologies in the field of distribution network operation and loss reduction. The most direct method to effectively reduce distribution network losses is to accurately calculate transformer losses. However, the existing methods for calculating transformer losses in current research still have the problem of low calculation efficiency, and do not consider the analysis and application of losses of multiple transformers in a large distribution network under unbalanced loads. With the development of the modernization and informatization of the power system, a large amount of data is generated in the power consumption of the power system. Calculating transformer losses under unbalanced loads in a distribution network plays a key role in the planning and management of the distribution network.

[0003] However, the current algorithms for calculating transformer losses under unbalanced loads in a distribution network have the following problems: 1. There is no dedicated algorithm for calculating the losses of multiple transformers in a large distribution network under unbalanced loads; 2. Calculation of transformer losses needs to consider the internal structure of the transformer, and the simulation process is relatively complex, requiring a large amount of time; 3. The parameters required for calculating transformer losses cannot be completely obtained from the power parameters monitored in real time in the distribution network, and the calculation efficiency is not high. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a data mechanism fusion prediction system and method for unbalanced transformer losses in a distribution network. Through reasonable algorithms and data preprocessing, the power parameters of the distribution network monitored in real time by the transformer can be directly utilized, and the input of the data mechanism fusion prediction module for transformer losses can be dimensionally reduced, solving the problems that the power parameters monitored in the distribution network cannot be directly utilized in the traditional transformer loss algorithm and the large amount of calculation caused by multiple distribution transformers.

[0005] On the one hand, a data mechanism fusion prediction system for unbalanced transformer losses in a distribution network includes a distribution network power data collection and linear discriminant analysis module, a zero-sequence impedance intelligent prediction module, and a data mechanism fusion prediction module for transformer losses;

[0006] The distribution network power data collection and linear discriminant analysis module uses the mutual information method to analyze the correlation of the collected distribution network voltage, current, active power, reactive power, and neutral line current, and uses the linear discriminant analysis method to perform dimensionality reduction processing on the distribution network voltage, current, active power, reactive power, and neutral line current, and inputs them to the data mechanism fusion prediction module for transformer losses;

[0007] The zero-sequence impedance intelligent prediction module establishes a new mathematical model of the transformer under unbalanced loads in the distribution network, uses an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer, and inputs the predicted zero-sequence impedance of the transformer to the data mechanism fusion prediction module of the transformer loss;

[0008] The data mechanism fusion prediction module of the transformer loss calculates the total loss of the transformer under unbalanced loads in the distribution network by using the input zero-sequence impedance of the transformer, trains the data mechanism fusion prediction module of the transformer loss by using the input after dimensionality reduction by the power grid power data collection and linear discriminant analysis module and the corresponding calculated total loss of the transformer, and performs real-time intelligent prediction of the transformer loss by using the real-time input after dimensionality reduction by the power grid power data collection and linear discriminant analysis module;

[0009] On the other hand, a data mechanism fusion prediction method for transformer loss under unbalanced distribution network is realized based on the aforementioned data mechanism fusion prediction system for transformer loss under unbalanced distribution network, and includes the following steps:

[0010] Step 1: Collect the power grid power data, form a matrix to constitute the original data set {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}; where the power grid power data includes three-phase voltage, three-phase current, three-phase active power, three-phase reactive power and neutral line current, U A ,U B ,U C are three-phase voltages, I A ,I B ,I C are three-phase currents, θ A ,θ B ,θ C are three-phase power factor angles, I n is the neutral line current, calculate the total loss T of the transformer by using the original data set loss , and analyze the correlation between the original data set {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n} and the total loss T of the transformer loss .

[0011] Step 1.1: Classify and organize the collected three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, and neutral line current data of the distribution network, calculate the power factor angle from the active power and reactive power, and obtain the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n};

[0012] Step 1.2: Set the total index sequence {Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11}, where Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11 are U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n , T loss , select the total transformer loss Z 11 as the reference sequence, the three-phase voltages U A , U B , U C , the three-phase currents I A , I B , I C , the three-phase power factor angles θ A , θ B , θ C , the neutral line current I n , a total of 10 items as the comparison sequences; among them, the total transformer loss is input into the zero-sequence impedance intelligent prediction module by the corresponding distribution network power data to intelligently predict the zero-sequence impedance of the transformer, and finally obtain the total transformer loss T loss ; Mutual information can reflect the amount of information sharing between these two sequences, and the calculation formula is:

[0013]

[0014] Wherein, X and Y respectively represent the reference sequence and the comparison sequence; p(x, y) represents the joint probability distribution, that is, the probability when X = x and Y = y; p(x) represents the marginal probability distribution of the random variable X, that is, the probability when X = x; p(y) represents the marginal probability distribution of the random variable Y, that is, the probability when Y = y; log represents the logarithmic function, using the logarithm with base 2, so the unit of mutual information is bit.

[0015] Step 1.3: Calculate the correlation degree between the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n} of each electric quantity in the distribution network and the total transformer loss T loss :

[0016]

[0017] Among them, I(X; Y) is the mutual information; H(X) and H(Y) are the entropies of X and Y respectively, and the formulas are:

[0018]

[0019] In the formula, H(X) represents the entropy of the sequence X, that is, the uncertainty in X; the larger the entropy, the greater the randomness and uncertainty of the sequence; H(Y) represents the entropy of the sequence Y; I(X; Y) represents the mutual information between X and Y, that is, the amount of information shared by the two sequences; NMI(X; Y) represents the correlation degree between X and Y, which is the normalized mutual information, and the value is between 0 and 1.

[0020] Step 2: Scan the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, and analyze and process the distribution network electric quantity data through the distribution network electric quantity data collection and linear discriminant analysis module, that is, for the purpose of simplifying the dimension of the input variables, reduce the dimension of the original data set.

[0021] Step 2.1: Collect the electrical parameters in the distribution network as the original data set, specifically including: three-phase voltages U A , U B , UC , three-phase current I A , I B , I C , three-phase power factor angle θ A , θ B , θ C , neutral line current I n , before performing linear discriminant analysis, first use unsupervised clustering to generate class labels;

[0022] Step 2.2: Standardize each feature;

[0023] The standardization uses Z-score standardization, and its formula is:

[0024]

[0025] In the formula, x is the original data; μ is the mean of this feature; σ is the standard deviation of this feature; x′ is the standardized data;

[0026] Step 2.3: Calculate the within-class scatter matrix, and its formula is:

[0027]

[0028] In the formula, S W is the within-class scatter matrix, indicating the data distribution within each class; C is the number of classes; X c is the data set belonging to class c; μ c is the mean of class c;

[0029] Calculate the between-class scatter matrix, and its formula is:

[0030]

[0031] In the formula, S B is the between-class scatter matrix, indicating the distribution difference between classes; N c is the number of samples in class c; μ is the mean of the original data set;

[0032] Step 2.4: Solve the generalized eigenvalue problem:

[0033]

[0034] In the formula, W is the corresponding eigenvector matrix; λ is the eigenvalue; select the first d eigenvectors with the largest eigenvalues to construct the dimensionality reduction projection matrix W d ;

[0035] Step 2.5: Project the original data set through the eigenvector matrix W d to obtain the dimensionality-reduced data set X d :

[0036] X d = XW d

[0037] where X is the original data after standardization; W d is the optimal projection matrix; X d is the data after dimensionality reduction;

[0038] Step 2.6: Apply dimensionality reduction analysis to the power consumption data of the distribution network. Input the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, where each sample data corresponds to a label of a different category. Output the data set after dimensionality reduction. Through dimensionality reduction by linear discriminant analysis, the original 10-dimensional data set can be projected into a lower-dimensional space.

[0039] Step 3: Use the original power consumption data set of the distribution network {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n} to establish a new mathematical model of the transformer under unbalanced load in the zero-sequence impedance intelligent prediction module, and use an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer; the mathematical model of the transformer under unbalanced load in the distribution network is the zero-sequence impedance intelligent prediction model, which is a mathematical modeling for the transformer and is used to predict the zero-sequence impedance of the transformer;

[0040] Step 3.1: Determine the short-circuit impedance parameters of the transformer according to the model of the transformer.

[0041] Short-circuit resistance R T :

[0042]

[0043] Short-circuit reactance X T :

[0044]

[0045] where ΔP k is the short-circuit loss of the transformer, ΔU t* is the percentage of short - circuit voltage of the transformer, and U n is the rated voltage of the transformer, and S n is the rated apparent power of the transformer;

[0046] Step 3.2: Using the mathematical model of the transformer under unbalanced distribution network load, determine the expression of the neutral point voltage:

[0047]

[0048] In the formula, R n is the zero - sequence resistance, X n is the zero - sequence reactance, I nr is the real part of the neutral - line current, and I ni is the imaginary part of the neutral - line current.

[0049] Step 3.3: Expand the established complex - number equation system into a real - part and imaginary - part equation system. This equation system has a total of six variables.

[0050] E r = f(E, R n , X n , α a , α b , α c )

[0051] E i = f(E, R n , X n , α a , α b , α c )

[0052] Among them, E is the equivalent voltage source on the load side of the transformer, E r is the real part of the equivalent voltage source, and E i is the imaginary part of the equivalent voltage source, R n is the zero - sequence resistance, X n is the zero - sequence reactance, and α a , α b , α c are respectively the angles between the three - phase measurement point voltages and the equivalent voltage source.

[0053] Step 3.4: Adopt an optimization algorithm to intelligently predict the zero - sequence impedance of the transformer;

[0054] First, collect the historical operation data and sensor information of the transformer for data pre - processing. Then, select an optimization algorithm, adjust the parameters of the zero - sequence impedance intelligent prediction model by optimizing the objective function, and continuously update the model parameters through iteration until the error reaches the set threshold or the maximum number of iterations; finally, evaluate the model through cross - validation and error analysis;

[0055] Step 4: Use the predicted zero-sequence impedance of the transformer to calculate the additional loss of the transformer under unbalanced distribution network loads, and the total loss of the transformer can be obtained by adding other losses of the transformer.

[0056] Step 4.1: Determine the no-load loss P0 of the transformer according to the model of the transformer.

[0057] Step 4.2: Use the original power data set of the distribution network input to determine the load loss P k :

[0058]

[0059] Wherein, R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively, k is the transformation ratio of the transformer, I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively.

[0060] Step 4.3: Use the zero-sequence impedance of the transformer predicted by intelligence to calculate the additional loss P f0 of the transformer under unbalanced distribution network loads, and determine the total loss P of the transformer.

[0061] Zero-sequence loss of the transformer under unbalanced distribution network loads:

[0062] P f0 =(3I0) 2 R0

[0063] Additional copper loss of the transformer under unbalanced distribution network loads:

[0064]

[0065] Total loss of the transformer:

[0066] P = P0 + P k + P′ k + P f0

[0067] Wherein, I0 is the zero-sequence current of the transformer, R0 is the zero-sequence resistance of the transformer, I A1 , I B1 and I C1 are the phase currents of the primary winding of the transformer respectively, I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively, and R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively.

[0068] Step 5: Use the calculated total transformer losses as the output part of the training samples for the data mechanism fusion prediction module of transformer losses.

[0069] Step 6: Use the input after dimensionality reduction by the power grid electricity data collection and linear discriminant analysis module as the input part of the training samples for the data mechanism fusion prediction module of transformer losses, and train the data mechanism fusion prediction module of transformer losses.

[0070] Step 7: Use the real-time input after dimensionality reduction by the power grid electricity data collection and linear discriminant analysis module to perform real-time intelligent prediction of transformer losses under unbalanced loads in the power grid.

[0071] Step 7.1: Obtain the label values and preliminary characteristic variables of the data model through the mechanism modeling of the unbalanced transformer in the power grid to guide the construction of the data model. The preliminary characteristic variables are used as the input part of the model, all of which are relevant operating parameters adopted in the model construction process and specific data can be collected through the power grid monitoring device. The label values of the model are used as the output part to guide the training of the model.

[0072] Step 7.2: Randomly divide the data set into a training set and a test set according to a ratio, set parameters, and establish a transformer loss prediction model under unbalanced conditions in the power grid based on XGBoost.

[0073] The tree model used by XGBoost is the CART regression tree model, which is represented by the following formula:

[0074]

[0075] In the formula: x i is the sample feature, f t (x i ) is the prediction of the x i sample using the t-th tree; q and K are the regression tree structure and number; is the predicted transformer loss value of the i-th sample; F is the set of all CART regression trees; f t (x) = ω q(x) , ω q(x) represents the weight of each leaf node, and q(x) represents the serial number of the output leaf node.

[0076] The iterative process of the transformer loss prediction model is as follows:

[0077]

[0078] In the formula: is the regression value of the i-th sample after t iterations; f t (x i ) is the newly added tree model function; is the regression value after the previous iteration of the i-th sample.

[0079] Step 7.3: Determine the objective function of the transformer loss prediction model as follows:

[0080]

[0081] In the formula: represents the residual between the true value and the predicted value, that is, the loss function; Ω(f t ) is the regularization term of the objective function, which determines the complexity of the tree and serves to prevent the model from overfitting; T is the number of leaf nodes; represents the square of the modulus of the leaf node weight; γ is the L1 regularization penalty term used to control the number of leaf nodes; λ is the L2 regularization penalty term to ensure that the leaf node scores are not too large; C is a constant term.

[0082] Step 7.4: To obtain the minimized objective function, perform a second-order Taylor expansion on the objective function at f t = 0 and remove the constant term to obtain the loss function of the transformer loss prediction model as:

[0083]

[0084] In the formula: respectively represent the first-order derivative and the second-order derivative of the loss function with respect to the current model; I j = {i|q(x i ) = j}, which represents the sample set on the leaf node with serial number j.

[0085] Step 7.5: Apply the transformer loss prediction model to predict the actual distribution network transformer loss value, compare with the prediction performance of multiple models, and finally calculate the model error and analyze the accuracy through the test set.

[0086] The beneficial effects produced by adopting the above technical solutions are as follows:

[0087] The present invention provides a data mechanism fusion prediction system and method for unbalanced transformer losses in a distribution network, having the following beneficial effects:

[0088] 1. Through reasonable algorithms and data preprocessing, the dimensionality of the large amount of distribution network power data can be reduced, and the transformer losses under unbalanced distribution network loads can be predicted in real time and intelligently, solving the problem that traditional transformer loss algorithms are not applicable to large distribution networks because they cannot directly calculate losses using distribution network power data.

[0089] 2. The data required for the mechanism fusion prediction of transformer losses under unbalanced distribution network loads comes entirely from the real-time monitored power data of the distribution network and does not require other data. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 is the overall system flowchart of the present invention;

[0091] Figure 2 is the program flowchart of the power data collection and linear discriminant analysis processing module of the distribution network of the present invention;

[0092] Figure 3 is the program flowchart of the zero-sequence impedance intelligent prediction module of the present invention;

[0093] Figure 4 is the program flowchart of the mechanism fusion prediction module of transformer losses of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0094] The following combines the drawings and embodiments to further describe in detail the specific embodiments of the present invention. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0095] On the one hand, a mechanism fusion prediction system for unbalanced distribution network transformer losses, as Figure 1 shown, includes a distribution network power data collection and linear discriminant analysis module, a zero-sequence impedance intelligent prediction module, and a mechanism fusion prediction module for transformer losses;

[0096] The distribution network power data collection and linear discriminant analysis module uses the mutual information method to analyze the correlation of power data such as distribution network voltage, current, active power, and reactive power collected, and uses the linear discriminant analysis method to perform dimensionality reduction processing on the power data such as distribution network voltage, current, active power, and reactive power, and inputs it to the mechanism fusion prediction module for transformer losses;

[0097] The zero-sequence impedance intelligent prediction module establishes a new mathematical model of the transformer under unbalanced distribution network loads, and uses an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer, and inputs the predicted zero-sequence impedance of the transformer to the mechanism fusion prediction module for transformer losses;

[0098] The mechanism fusion prediction module for transformer losses calculates the total losses of the transformer under unbalanced distribution network loads using the input zero-sequence impedance of the transformer, trains the mechanism fusion prediction module for transformer losses using the reduced-dimensional input from the distribution network power data collection and linear discriminant analysis module and the corresponding calculated total losses of the transformer, and performs real-time intelligent prediction of transformer losses using the real-time input reduced by the distribution network power data collection and linear discriminant analysis module;

[0099] On the other hand, a data mechanism fusion prediction method for unbalanced transformer losses in a distribution network is implemented based on the aforementioned data mechanism fusion prediction system for unbalanced transformer losses in a distribution network, and includes the following steps:

[0100] Step 1: Collect the power grid power data to form a matrix to constitute the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}; where the power grid power data includes three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, and neutral line current. Calculate the total transformer loss T loss , and analyze the correlation between the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n} and the total transformer loss T loss .

[0101] In this embodiment, the form of the original data set is shown in Table 1;

[0102] Table 1: Original data set matrix table;

[0103]

[0104] Step 1.1: Classify and sort the collected three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, and neutral line current data of the power grid, and calculate the power factor angle from the active power and reactive power. Through classification and sorting, obtain the power grid power data {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, where U A , U B , U C are the three-phase voltages, and I A , IB , I C is the three - phase current, θ A , θ B , θ C is the three - phase power factor angle, I n is the neutral - line current.

[0105] Step 1.2: Set the total index sequence {Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11}, where Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11 are respectively U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n , T loss , select the total transformer loss Z 11 as the reference sequence, the three - phase voltages U A , U B , U C , the three - phase currents I A , I B , I C , the three - phase power factor angles θ A , θ B , θ C , the neutral - line current I n , a total of 10 items as the comparison sequences; among them, the total transformer loss is input into the zero - sequence impedance intelligent prediction module by the corresponding distribution network power data to intelligently predict the zero - sequence impedance of the transformer, and finally the total transformer loss T loss ; Mutual information can reflect the amount of information sharing between these two sequences, and the calculation formula is:

[0106]

[0107] In the formula, X and Y represent the reference sequence and the comparison sequence respectively; p(x, y) represents the joint probability distribution, that is, the probability when X = x and Y = y; p(x) represents the marginal probability distribution of the random variable X, that is, the probability that X = x; p(y) represents the marginal probability distribution of the random variable Y, that is, the probability that Y = y; log represents the logarithmic function, using the logarithm with base 2, so the unit of mutual information is bit.

[0108] Step 1.3: Calculate the original data set of each power quantity in the distribution network {U A,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}, and the correlation with the total transformer loss T loss :

[0109]

[0110] Among them, I(X; Y) is the mutual information; H(X) and H(Y) are the entropies of X and Y respectively, and the formula is:

[0111]

[0112] In the formula, H(X) represents the entropy of sequence X, that is, the uncertainty in X. The larger the entropy, the greater the randomness and uncertainty of the sequence; H(Y) represents the entropy of sequence Y; I(X; Y) represents the mutual information between X and Y, that is, the amount of information shared by the two sequences; NMI(X; Y) represents the correlation between X and Y, which is the normalized mutual information, and the value ranges from 0 to 1.

[0113] Step 2: Scan the original dataset {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}, and analyze and process the power grid power data through the power grid power data collection and linear discriminant analysis module, that is, for the purpose of simplifying the dimension of the input variables, reduce the dimension of the original dataset, as Figure 2 shown.

[0114] Step 2.1: Collect the electrical parameters in the power grid as the original dataset, specifically including: U A ,U B ,U C (three-phase voltage); I A ,I B ,I C (three-phase current); θ A ,θ B ,θ C (three-phase power factor angle); I n (neutral line current). Before performing linear discriminant analysis, first use unsupervised clustering to generate class labels.

[0115] Step 2.2: Linear discriminant analysis is sensitive to data at different scales. Therefore, it is necessary to standardize each feature. The standardization method adopted is Z-score standardization, and its formula is:

[0116]

[0117] In the formula, \(x\) is the original data; \(\mu\) is the mean of this feature; \(\sigma\) is the standard deviation of this feature; \(x'\) is the standardized data.

[0118] Step 2.3: Calculate the within-class scatter matrix, and its formula is:

[0119]

[0120] In the formula, \(S\) W is the within-class scatter matrix, indicating the data distribution within each category; \(C\) is the number of categories; \(X\) c is the data set belonging to category \(c\); \(\mu\) c is the mean of category \(c\).

[0121] Calculate the between-class scatter matrix, and its formula is:

[0122]

[0123] In the formula, \(S\) B is the between-class scatter matrix, indicating the distribution difference between categories; \(N\) c is the number of samples in category \(c\); \(\mu\) is the mean of the global data set.

[0124] Step 2.4: Solve the generalized eigenvalue problem:

[0125]

[0126] In the formula, \(W\) is the corresponding eigenvector matrix; \(\lambda\) is the eigenvalue. Select the first \(d\) eigenvectors with the largest eigenvalues to construct the dimensionality reduction projection matrix \(W\) d .

[0127] Step 2.5: The original data set \(X\) can be projected through the eigenvector matrix \(W\) d to obtain the dimensionality-reduced data set \(X\) d :

[0128] \(X\) d = \(XW\) d

[0129] In the formula, \(X\) is the standardized original data; \(W\) d is the optimal projection matrix; \(X\) d is the dimensionality-reduced data.

[0130] Step 2.6: Apply dimensionality reduction analysis to the power quantity data of the distribution network. Input the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, where each sample data corresponds to labels of different categories. Output the data set after dimensionality reduction. Through dimensionality reduction by linear discriminant analysis, the original 10-dimensional data set can be projected into a lower-dimensional space.

[0131] Step 3: Utilize the original power quantity data set of the distribution network {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n} to establish a new mathematical model of the transformer under unbalanced load in the zero-sequence impedance intelligent prediction module, and use an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer, as Figure 3 shown.

[0132] Step 3.1: Determine the short-circuit impedance parameters of the transformer according to the model of the transformer.

[0133] Short-circuit resistance R T :

[0134]

[0135] Short-circuit reactance X T :

[0136]

[0137] Where ΔP k is the short-circuit loss of the transformer, ΔU t * is the percentage of short-circuit voltage of the transformer, U n is the rated voltage of the transformer, and S n is the rated apparent power of the transformer;

[0138] Step 3.2: Utilize the mathematical model of the transformer under unbalanced load of the distribution network to determine the neutral point voltage expression:

[0139]

[0140] Wherein, R n is the zero-sequence resistance, X n is the zero-sequence reactance, I nr is the real part of the neutral line current, and I ni is the imaginary part of the neutral line current.

[0141] Step 3.3: Expand the established complex equation system into a real part and an imaginary part equation system, and this equation system has a total of six variables.

[0142] E r = f(E, R n , X n , α a , α b , α c )

[0143] E i = f(E, R n , X n , α a , α b , α c )

[0144] Wherein, E is the equivalent voltage source on the load side of the transformer, E r is the real part of the equivalent voltage source, and E i is the imaginary part of the equivalent voltage source, R n is the zero-sequence resistance, X n is the zero-sequence reactance, and α a , α b , α c are respectively the angles between the three-phase measurement point voltages and the equivalent voltage source.

[0145] Step 3.4: Adopt an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer. First, collect the historical operation data and sensor information of the transformer for data preprocessing to ensure data quality. Then, select an optimization algorithm, adjust the parameters of the zero-sequence impedance intelligent prediction model by optimizing the objective function (minimizing the mean square error), and continuously update the model parameters through iteration until the error reaches the set threshold or the maximum number of iterations; finally, evaluate the model through cross-validation and error analysis to ensure its generalization ability on unknown data, and apply the optimized model to real-time prediction and operation and maintenance decision support to improve the operation reliability and maintenance efficiency of the transformer;

[0146] Step 4: Use the predicted zero-sequence impedance of the transformer to calculate the additional loss of the transformer under unbalanced distribution network loads, and add other losses of the transformer to obtain the total loss of the transformer.

[0147] Step 4.1: Determine the no-load loss P0 of the transformer according to the model of the transformer.

[0148] Step 4.2: Determine the load loss P of the transformer using the original power dataset of the distribution network input. k :

[0149]

[0150] Wherein, R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively, k is the transformation ratio of the transformer, and I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively.

[0151] Step 4.3: Calculate the additional loss P of the transformer under unbalanced distribution network load using the zero-sequence impedance of the transformer predicted by intelligence, and determine the total loss P of the transformer. f0 , and determine the total loss P of the transformer.

[0152] Zero-sequence loss of the transformer under unbalanced distribution network load:

[0153] P f0 =(3I0) 2 R0

[0154] Additional copper loss of the transformer under unbalanced distribution network load:

[0155]

[0156] Total loss of the transformer:

[0157] P = P0 + P k + P′ k + P f0

[0158] Wherein, I0 is the zero-sequence current of the transformer, R0 is the zero-sequence resistance of the transformer, I A1 , I B1 and I C1 are the phase currents of the primary winding of the transformer respectively, I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively, and R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively.

[0159] Step 5: Use the calculated total loss of the transformer as the output part of the training sample of the data mechanism fusion prediction module for transformer loss.

[0160] Step 6: Use the input after dimensionality reduction by the distribution network power data collection and linear discriminant analysis module as the input part of the training sample of the data mechanism fusion prediction module for transformer loss, and train the data mechanism fusion prediction module for transformer loss.

[0161] Step 7: Use the real-time input after dimensionality reduction by the power grid power data collection and linear discriminant analysis module to perform real-time intelligent prediction on the transformer losses under unbalanced load of the distribution network, as Figure 4 shown.

[0162] Step 7.1: Obtain the label values and preliminary characteristic variables of the data model through the mechanism modeling of the unbalanced transformer of the distribution network to guide the construction of the data model. The preliminary characteristic variables, as the input part of the model, are all relevant operating parameters adopted during the model construction process, and specific data can be collected through the distribution network monitoring device. The label value of the model, as the output part, is used to guide the training of the model.

[0163] Step 7.2: Randomly divide the data set into a training set and a test set according to a certain proportion, set parameters, and establish a transformer loss prediction model under unbalanced conditions of the distribution network based on XGBoost.

[0164] The tree model used by XGBoost is the CART regression tree model, which can be expressed by the following formula:

[0165]

[0166] In the formula: x i is the sample feature, and f t (x i ) is to predict x i using the t-th tree; q and K are the regression tree structure and number; is the transformer loss value predicted for the i-th sample; F is the set of all CART regression trees; f t (x) = ω q(x) , ω q(x) represents the weight of each leaf node, and q(x) represents the serial number of the output leaf node.

[0167] The model iteration process is as follows:

[0168]

[0169] In the formula: is the regression value of the i-th sample after t iterations; f t (x i ) is the newly added tree model function; is the regression value of the i-th sample after the previous iteration.

[0170] Step 7.3: Determine the objective function of the XGBoost algorithm as follows:

[0171]

[0172] In the formula: represents the residual between the true value and the predicted value, i.e., the loss function; Ω(f t ) is the regularization term of the objective function, which determines the complexity of the tree and serves to prevent the model from overfitting; T is the number of leaf nodes; represents the square of the modulus of the leaf node weight; γ is the L1 regularization penalty term used to control the number of leaf nodes; λ is the L2 regularization penalty term to ensure that the leaf node scores are not too large; C is a constant term.

[0173] Step 7.4: To obtain the minimized objective function, perform a second-order Taylor expansion of the objective function at f t = 0 and remove the constant term, obtaining the model loss function as:

[0174]

[0175] In the formula: respectively represent the first derivative and the second derivative of the loss function with respect to the current model; I j = {i|q(x i ) = j}, representing the sample set on the leaf node with serial number j.

[0176] Step 7.5: Apply the XGBoost model to the prediction of the actual distribution network transformer loss value, compare with the prediction performances of multiple models, and finally calculate the model error and analyze the accuracy through the test set.

[0177] The above description is only the preferred embodiment of the present disclosure and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A data mechanism fusion prediction system for the losses of a distribution network unbalanced transformer, characterized in that It includes a power grid power data collection and linear discriminant analysis module, a zero-sequence impedance intelligent prediction module, and a data mechanism fusion prediction module for transformer losses; The power grid power data collection and linear discriminant analysis module uses the mutual information method to analyze the correlation degree of the collected power grid voltage, current, active power, reactive power, and neutral line current, and uses the linear discriminant analysis method to perform dimensionality reduction processing on the power grid voltage, current, active power, reactive power, and neutral line current, and inputs it to the data mechanism fusion prediction module for transformer losses; The zero-sequence impedance intelligent prediction module establishes a new mathematical model of the transformer under unbalanced load of the distribution network, and uses an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer, and inputs the predicted zero-sequence impedance of the transformer to the data mechanism fusion prediction module for transformer losses; The data mechanism fusion prediction module for transformer losses calculates the total losses of the transformer under unbalanced load of the distribution network by using the input zero-sequence impedance of the transformer, trains the data mechanism fusion prediction module for transformer losses by using the input after dimensionality reduction of the power grid power data collection and linear discriminant analysis module and the corresponding calculated total losses of the transformer, and performs real-time intelligent prediction of transformer losses by using the real-time input after dimensionality reduction of the power grid power data collection and linear discriminant analysis module.

2. A data mechanism fusion prediction method for the loss of a distribution network unbalanced transformer, which is implemented based on the data mechanism fusion prediction system for the loss of a distribution network unbalanced transformer described in claim 1, and is characterized in that, It includes the following steps: Step 1: Collect the power grid power data, form a matrix to constitute the original data set {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}; among them, the power grid power data includes three-phase voltage, three-phase current, three-phase active power, three-phase reactive power and neutral line current, U A ,U B ,U C are three-phase voltages, I A ,I B ,I C are three-phase currents, θ A ,θ B ,θ C are three-phase power factor angles, I n is the neutral line current, calculate the total transformer loss T loss , and use the mutual information method to analyze the original data set {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n} and the correlation degree between the total transformer loss T loss ; Step 2: Scan the original dataset {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}, and analyze and process the power grid power data through the power grid power data collection and linear discriminant analysis module, that is, for the purpose of simplifying the dimension of the input variables, reduce the dimension of the original dataset; Step 3: Using the original power dataset of the distribution network {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, a new mathematical model of the transformer under unbalanced distribution network loads is established in the zero-sequence impedance intelligent prediction module, and an optimization algorithm is used to intelligently predict the zero-sequence impedance of the transformer; the mathematical model of the transformer under unbalanced distribution network loads is a zero-sequence impedance intelligent prediction model, which is a mathematical modeling for the transformer and is used to predict the zero-sequence impedance of the transformer; Step 4: Use the predicted zero-sequence impedance of the transformer to calculate the additional losses of the transformer under unbalanced load of the distribution network, and the total losses of the transformer can be obtained by adding other losses of the transformer; Step 5: Take the calculated total losses of the transformer as the output part of the training sample of the data mechanism fusion prediction module for transformer losses; Step 6: Take the input after dimensionality reduction of the power grid power data collection and linear discriminant analysis module as the input part of the training sample of the data mechanism fusion prediction module for transformer losses, and train the data mechanism fusion prediction module for transformer losses; Step 7: Use the real-time input after dimensionality reduction of the power grid power data collection and linear discriminant analysis module to perform real-time intelligent prediction of transformer losses under unbalanced load of the distribution network.

3. A data mechanism fusion prediction method for the losses of a distribution network unbalanced transformer according to claim 2, characterized in that The said Step 1 includes the following steps: Step 1.1: Classify and organize the collected data of three-phase voltage, three-phase current, three-phase active power, three-phase reactive power, and neutral line current in the distribution network, calculate the power factor angle from the active power and reactive power, and obtain the original data set {U A ,U B ,U C ,I A ,I B ,I C ,θ A ,θ B ,θ C ,I n}; Step 1.2: Let the total index sequence be {Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11}, where Z1, Z2, Z3, Z4, Z5, Z6, Z7, Z8, Z9, Z 10 , Z 11 are respectively U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n , T loss . Select the total transformer loss Z 11 as the reference sequence, the three-phase voltages U A , U B , U C , the three-phase currents I A , I B , I C , the three-phase power factor angles θ A , θ B , θ C , the neutral line current I n , a total of 10 items as the comparison sequences; among them, the total transformer loss is input into the zero-sequence impedance intelligent prediction module by the corresponding distribution network power data to intelligently predict the zero-sequence impedance of the transformer, and finally the total transformer loss T loss is obtained; Mutual information can reflect the amount of information sharing between these two sequences, and the calculation formula is: In the formula, X and Y represent the reference sequence and the comparison sequence respectively; p(x,y) represents the joint probability distribution, that is, the probability when X = x and Y = y; p(x) represents the marginal probability distribution of the random variable X, that is, the probability that X = x; p(y) represents the marginal probability distribution of the random variable Y, that is, the probability that Y = y; log represents the logarithmic function, using the logarithm with base 2, so the unit of mutual information is bit; Step 1.3: Calculate the correlation degree between the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n} of each power quantity in the distribution network and the total transformer loss T loss : Among them, I(X;Y) is the mutual information; H(X) and H(Y) are the entropies of X and Y respectively, and the formula is: In the formula, H(X) represents the entropy of the sequence X, that is, the uncertainty in X; the larger the entropy, the greater the randomness and uncertainty of the sequence; H(Y) represents the entropy of the sequence Y; I(X;Y) represents the mutual information between X and Y, that is, the amount of information shared by the two sequences; NMI(X;Y) represents the correlation degree between X and Y, which is the normalized mutual information, and the value is between 0 and 1.

4. A data mechanism fusion prediction method for unbalanced transformer losses in a distribution network according to claim 2, characterized in that The said Step 2 includes the following steps: Step 2.1: Collect electrical parameters in the distribution network as the original data set, specifically including: three-phase voltages U A , U B , U C , three-phase currents I A , I B , I C , three-phase power factor angles θ A , θ B , θ C , neutral line current I n . Before performing linear discriminant analysis, first use unsupervised clustering to generate class labels; Step 2.2: Perform standardization processing on each feature; The standardization adopts Z-score standardization, and its formula is: In the formula, x is the original data; μ is the mean of this feature; σ is the standard deviation of this feature; x′ is the standardized data; Step 2.3: Calculate the within-class scatter matrix, and its formula is: where S W is the within-class scatter matrix, representing the data distribution within each class; C is the number of classes; X c is the data set belonging to class c; μ c is the mean of class c; Calculate the between-class scatter matrix, and its formula is: where S B is the between-class scatter matrix, representing the distribution difference between classes; N c is the number of samples in class c; μ is the mean of the original data set; Step 2.4: Solve the generalized eigenvalue problem: Where W is the corresponding eigenvector matrix; λ is the eigenvalue; the eigenvectors with the largest d eigenvalues are selected to construct the dimensionality reduction projection matrix W d ; Step 2.5: The original dataset is projected through the feature vector matrix W d to obtain the dimension-reduced dataset X d : X d = XW d where X is the original data after standardization; W d is the optimal projection matrix; X d is the data after dimensionality reduction; Step 2.6: Apply dimensionality reduction analysis to the power consumption data of the distribution network. Input the original data set {U A , U B , U C , I A , I B , I C , θ A , θ B , θ C , I n}, where each sample data corresponds to labels of different categories. Output the data set after dimensionality reduction. Through linear discriminant analysis for dimensionality reduction, project the original 10-dimensional data set into a lower-dimensional space.

5. A data mechanism fusion prediction method for the losses of a distribution network unbalanced transformer according to claim 2, characterized in that The said Step 3 includes the following steps: Step 3.1: Determine the short-circuit impedance parameters of the transformer according to the model of the transformer; Short-circuit resistance R T : Short-circuit reactance X T : where ΔP k is the short-circuit loss of the transformer, and ΔU t * is the percentage of short-circuit voltage of the transformer, U n is the rated voltage of the transformer, and S n is the rated apparent power of the transformer; Step 3.2: Use the mathematical model of the transformer under unbalanced distribution network load to determine the expression of the neutral point voltage: Wherein, R n is the zero-sequence resistance, X n is the zero-sequence reactance, I nr is the real part of the neutral line current, I ni is the imaginary part of the neutral line current; Step 3.3: Expand the established complex equation system into a real part and an imaginary part equation system, and there are a total of six variables in this equation system; E r = f(E, R n , X n , α a , α b , α c ) E i = f(E, R n , X n , α a , α b , α c ) Among them, E is the equivalent voltage source on the load side of the transformer, E r is the real part of the equivalent voltage source, E i is the imaginary part of the equivalent voltage source, R n is the zero-sequence resistance, X n is the zero-sequence reactance, α a , α b , α c are respectively the angles between the three-phase measurement point voltages and the equivalent voltage source; Step 3.4: Adopt an optimization algorithm to intelligently predict the zero-sequence impedance of the transformer; First, collect the historical operation data and sensor information of the transformer for data preprocessing. Then, select an optimization algorithm, adjust the parameters of the zero-sequence impedance intelligent prediction model through the optimization objective function, and continuously update the model parameters through iteration until the error reaches the set threshold or the maximum number of iterations; finally, evaluate the model through cross-validation and error analysis.

6. The data mechanism fusion prediction method for the loss of a distribution network unbalanced transformer according to claim 2, characterized in that, The said Step 4 includes the following steps: Step 4.1: Determine the no-load loss P0 of the transformer according to the model of the transformer; Step 4.2: Determine the load loss P of the transformer using the input original power consumption dataset of the distribution network k : Wherein, R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively, k is the turns ratio of the transformer, I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively; Step 4.3: Calculate the additional loss \(P\) of the transformer under unbalanced distribution network loads by using the zero-sequence impedance of the intelligent prediction transformer f0 , and determine the total loss \(P\) of the transformer; Zero-sequence loss of the transformer under unbalanced distribution network load: P f0 = (3I0) 2 R0 Additional copper loss of the transformer under unbalanced distribution network load: Total loss of the transformer: P = P0 + P k + P' k + P f0 Wherein, I0 is the zero-sequence current of the transformer, R0 is the zero-sequence resistance of the transformer, I A1 , I B1 and I C1 are the phase currents of the primary winding of the transformer respectively, I A2 , I B2 and I C2 are the phase currents of the secondary winding of the transformer respectively, and R1 and R2 are the DC resistances of the primary and secondary windings of the transformer respectively.

7. A data mechanism fusion prediction method for the losses of a distribution network unbalanced transformer according to claim 2, characterized in that, The said Step 7 includes the following steps: Step 7.1: Obtain the label value and preliminary feature variables of the data model through mechanism modeling of the unbalanced distribution network transformer to guide the construction of the data model; the preliminary feature variables are used as the input part of the model, all of which are relevant operation parameters adopted in the model construction process, and specific data can be collected through the distribution network monitoring device; the label value of the model is used as the output part to guide the training of the model; Step 7.2: Randomly divide the data set into a training set and a test set according to a certain proportion, set parameters, and establish a transformer loss prediction model under unbalanced distribution network conditions based on XGBoost; The tree model used by XGBoost is the CART regression tree model, which is represented by the following formula: Where: x i is the sample feature, f t (x i ) is the prediction of x i using the t-th tree; q and K are the regression tree structure and number; is the transformer loss value obtained from the prediction of the i-th sample; F is the set of all CART regression trees; f t (x) = ω q(x) , ω q(x) represents the weight of each leaf node, and q(x) represents the serial number of the output leaf node; The iterative process of the transformer loss prediction model is as follows: In the formula: is the regression value after the t-th iteration of the i-th sample; f t (x i ) is the newly added tree model function; is the regression value after the previous iteration of the i-th sample; Step 7.3: Determine the objective function of the transformer loss prediction model, as follows: In the formula: represents the residual between the true value and the predicted value, that is, the loss function; Ω(f t ) is the regularization term of the objective function, which determines the complexity of the tree and serves to prevent the model from overfitting; T is the number of leaf nodes; represents the square of the modulus of the leaf node weight; γ is the L1 regularization penalty term used to control the number of leaf nodes; λ is the L2 regularization penalty term to ensure that the leaf node scores are not too large; C is a constant term; Step 7.4: To obtain the minimized objective function, perform a second-order Taylor expansion on the objective function at f t = 0, and remove the constant term, to obtain the loss function of the transformer loss prediction model as follows: In the formula: respectively represent the first-order derivative and the second-order derivative of the loss function with respect to the current model; I j = {i|q(x i ) = j}, represents the sample set on the leaf node with serial number j; Step 7.5: Apply the transformer loss prediction model to predict the loss value of the actual distribution network transformer, and calculate the model error and analyze the accuracy through the test set.

Images