Method for determining initial phase angle of distribution transformer short-circuit current withstand test

Abstract

The application provides a method for determining the initial phase angle of a distribution transformer during a high-capacity short-circuit resistance test current closing, belonging to the technical field of distribution transformers, comprising: collecting a large amount of short-circuit test data of distribution transformers, including short-circuit test data under different short-circuit loop parameters and initial phase angles of closing, and constructing a training data set; training a neural network model using the training data set to obtain a short-circuit test model; predicting winding stress, deformation and impact coefficient under different initial phase angles of closing using the short-circuit test model; setting the allowable threshold of winding stress and deformation, finding the optimal initial phase angle range that meets the requirements based on the neural network prediction results; considering the closing time deviation of the short-circuit test equipment, further optimizing and determining the final allowable range of the initial phase angle of closing; performing grid sampling within the allowable range, predicting the indicators under each discrete initial phase angle of closing, and defining a comprehensive evaluation function to calculate the score of each initial phase angle; and selecting the initial phase angle with the optimal score.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of distribution transformers, and in particular, relates to a method for determining the closing initial phase angle of a large-capacity short-circuit withstand test current of a distribution transformer. BACKGROUND

[0002] As a key device in the power system, the stable and reliable operation of the distribution transformer is crucial to the safe operation of the entire power grid. During normal grid operation, the electromagnetic parameters of the distribution transformer are roughly maintained around the rated value. However, when a large-capacity short-circuit fault suddenly occurs in the power system, the windings of the transformer will be subjected to a huge electric force, seriously threatening its safe and stable operation.

[0003] Specifically, when a short-circuit fault occurs on the secondary side of the distribution transformer, an extremely large short-circuit current will be generated in the winding. According to electromagnetic theory, the coupling of the short-circuit current and the leakage magnetic field of the winding will result in a huge instantaneous electric force on the winding. These electric forces can cause winding insulation damage, winding deformation, or even the collapse of the entire winding. Even if the short-circuit process is extremely short, if the short-circuit carrying capacity of the winding is not designed adequately, it can also cause serious faults and threaten the safe and stable operation of the power grid.

[0004] Currently, in order to evaluate the performance of the distribution transformer under short-circuit conditions, the power industry generally uses the method of large-capacity short-circuit test. During the test, the low-voltage side of the transformer is short-circuited to apply a large current impact, and the stress and deformation responses of the winding are measured. This test method can realistically simulate actual short-circuit fault conditions and provide important basis for the design of the transformer.

[0005] However, the existing large-capacity short-circuit test has some problems:

[0006] 1) The closing initial phase angle of the test circuit power supply is difficult to accurately control. Different closing initial phase angles will result in large differences in short-circuit current peak value and impact factor, affecting the repeatability of the test results.

[0007] 2) There is a certain mechanical error in the closing time of the test equipment, causing the actual closing initial phase angle to deviate from the expected value. This will cause the test results to deviate from the actual use conditions.

[0008] If the closing initial phase angle of the large-capacity short-circuit current of the distribution transformer winding can be calculated and optimized in advance, the results of the large-capacity short-circuit test of the distribution transformer winding will be improved; the large capacity can refer to GB 1094.5. SUMMARY

[0009] Therefore, the application provides a method for determining the initial phase angle of a distribution transformer under a large-capacity short-circuit test current, which can solve the technical problem that the initial phase angle of a distribution transformer winding under a large-capacity short-circuit current cannot be calculated and optimized in advance.

[0010] The application is implemented as follows:

[0011] The first aspect of the application provides a method for determining the initial phase angle of a distribution transformer under a large-capacity short-circuit test current, which comprises the following steps:

[0012] S10, collecting short-circuit test data of a large number of distribution transformers, including short-circuit test data under different short-circuit loop parameters and different initial phase angles of a short-circuit current;

[0013] S20, preprocessing the test data to construct a training data set;

[0014] S30, training a neural network model using the training data set to obtain a short-circuit test model;

[0015] S40, using the short-circuit test model, for a given short-circuit loop parameter, initial phase angle, winding electromagnetic distribution, and electric force distribution, predicting and calculating the corresponding winding stress, deformation, and impact coefficient by traversing different initial phase angle values;

[0016] S50, setting an allowable threshold range of winding stress and deformation, and finding an optimal initial phase angle range that meets the requirements based on the neural network prediction result;

[0017] S60, further optimizing and determining the final initial phase angle range by combining the closing time deviation of the short-circuit test equipment;

[0018] S70, performing grid sampling within the determined initial phase angle range to obtain a plurality of discrete initial phase angle values in the range, and for each discrete initial phase angle value, using the short-circuit test model to predict the corresponding winding stress, deformation, and impact coefficient;

[0019] S80, defining a comprehensive evaluation function, integrating the winding stress, deformation, and impact coefficient, calculating the comprehensive score of each discrete initial phase angle, and taking the range formed by the multiple initial phase angles with the optimal score as the optimized initial phase angle.

[0020] Compared with the prior art, the power distribution transformer large-capacity short-circuit resistance test current closing initial phase angle determination method provided by the application has the beneficial effects that: the stress, deformation and short-circuit impact coefficient and other key performance indicators of the winding under the action of short circuit can be accurately predicted by making full use of a large amount of short-circuit test data and combining advanced neural network modeling technology. By analyzing these prediction results, the optimal closing initial phase angle range that meets the requirements of winding structure strength and safety can be determined, thereby providing a scientific basis for subsequent large-capacity short-circuit tests. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the method provided by the application is shown in the figure.

[0022] Figure 2 The simulation waveform diagram when the closing initial phase angle is -10 is shown in the figure.

[0023] Figure 3 The simulation waveform diagram when the closing initial phase angle is -70 is shown in the figure.

[0024] Figure 4 The simulation waveform diagram when the closing initial phase angle is -130 is shown in the figure.

[0025] Figure 5 The simulation waveform diagram when the closing initial phase angle is -190 is shown in the figure.

[0026] Figure 6 The simulation waveform diagram when the closing initial phase angle is -250 is shown in the figure.

[0027] Figure 7 The simulation waveform diagram when the closing initial phase angle is -310 is shown in the figure. DETAILED DESCRIPTION

[0028] To make the purpose, technical scheme and advantages of the application clearer, the technical scheme of the application will be described clearly and completely below in combination with specific embodiments of the application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0029] As shown in the figure, it is a flowchart of a power distribution transformer large-capacity short-circuit resistance test current closing initial phase angle determination method provided by the application, and the method comprises the following steps: Figure 1

[0030] S10, collect a large amount of short-circuit test data of power distribution transformers, including different short-circuit loop parameters, short-circuit test data under different closing initial phase angles and multiple short-circuit working conditions;

[0031] S20, preprocess the test data and construct a training data set;​

[0032] S30, training a neural network model by using the training data set to obtain a short-circuit test model;

[0033] S40, using the short-circuit test model, for given short-circuit loop parameters, closing initial phase angle, winding electromagnetic distribution, and electric force distribution, predicting and calculating corresponding winding stress, deformation, and impact coefficient by traversing different closing initial phase angle values;

[0034] S50, setting an allowable threshold range of winding stress and deformation, and finding an optimal closing initial phase angle range that meets the requirements based on the neural network prediction results;

[0035] S60, further optimizing and determining the final closing initial phase angle allowable range in combination with the closing time deviation of the short-circuit test equipment;

[0036] S70, within the determined closing initial phase angle allowable range, performing grid sampling to obtain a plurality of discrete closing initial phase angle values in the range, and for each discrete closing initial phase angle value, using the short-circuit test model to predict the corresponding winding stress, deformation, and impact coefficient;

[0037] S80, defining a comprehensive evaluation function, integrating winding stress, deformation, and impact coefficient, calculating the comprehensive score of each discrete closing initial phase angle, and taking the range formed by the multiple closing initial phase angles with the optimal score as the optimized closing initial phase angle.

[0038] The specific implementation of the above steps is described in detail as follows:

[0039] Step S10: Collect a large amount of short-circuit test data of distribution transformers

[0040] The specific implementation is as follows:

[0041] 1) Collect a large amount of short-circuit test data of distribution transformers

[0042] 2) Ensure that the collected test data covers various short-circuit conditions that may occur, such as three-phase symmetrical short-circuit and single-phase short-circuit, and covers different closing initial phase angles (such as -10°, -70°, -130°, -190°, -250°, -310°, etc.) as much as possible.

[0043] 3) For each short-circuit condition and closing initial phase angle, record the electromagnetic and electric force distribution of the winding, as well as the structural response data such as winding stress and deformation caused thereby.

[0044] 4) For repeated test data under the same working condition, key characteristic values are extracted using statistical analysis methods to enhance the representativeness and robustness of the data.

[0045] The purpose of this step is to establish a test database covering a wide range of short-circuit conditions, laying a solid foundation for subsequent neural network model training. By collecting a large amount of high-quality measured data, the electromagnetic-structural response characteristics of distribution transformers under short-circuit action can be effectively described.

[0046] Step S20: Preprocessing test data, constructing training data set

[0047] The specific implementation is:

[0048] 1) Preprocess the collected test data, including outlier detection and removal, missing value completion, data normalization, etc., to ensure data quality and availability.

[0049] 2) According to professional knowledge, determine the input feature variables, including at least short-circuit loop parameters (resistance, inductance), initial phase angle of closing, winding electromagnetic distribution (magnetic density), and electric force distribution.

[0050] 3) Determine the desired output variables, including winding stress, deformation, and short-circuit impact coefficient, etc.

[0051] 4) Construct the preprocessed input features and desired outputs into a standardized training data set, preparing for subsequent neural network model training.

[0052] The purpose of this step is to convert the original test data into the standard input and output format of the neural network model, providing support for subsequent model training and optimization. By reasonably designing input and output variables, the neural network model can effectively learn and predict the structural response characteristics of distribution transformers under short-circuit conditions.

[0053] Step S30: Train the neural network model using the training data set

[0054] The specific implementation is:

[0055] 1) Select an appropriate neural network architecture, such as multi-layer perceptron (MLP) or convolutional neural network (CNN), and determine the network layer number, node number, and other hyperparameters.

[0056] 2) Use the square loss function as the training target, and use the backpropagation algorithm to optimize and update the network parameters.

[0057] 3) Use cross-validation to evaluate the model's performance on the training set and validation set, and adjust the hyperparameters appropriately until the desired prediction accuracy is achieved.

[0058] 4) Finally, a trained and optimized neural network model is obtained, denoted as the short-circuit test model.

[0059] The purpose of this step is to train a high-performance neural network model using the aforementioned training dataset, which will be used for subsequent prediction of the structural response of distribution transformers under short-circuit conditions. By repeatedly optimizing the network architecture and training parameters, it can be ensured that the model can effectively learn and express the complex electromagnetic-structural coupling characteristics of distribution transformers.

[0060] Step S40: Predicting winding stress, deformation, and impact coefficient using neural network model

[0061] The specific implementation is as follows:

[0062] 1) For given short-circuit loop parameters (resistance, inductance), closing initial phase angle, winding electromagnetic distribution (magnetic flux density), and electric force distribution, etc. input characteristics, the trained short-circuit test model is used for forward propagation calculation.

[0063] 2) The output variables of the model include winding stress, deformation, and short-circuit impact coefficient, etc. key structural response indicators.

[0064] 3) By iterating through different closing initial phase angle values (such as -10°, -70°, -130°, -190°, -250°, -310°, etc.), the short-circuit test model is used to predict the corresponding winding stress, deformation, and short-circuit impact coefficient.

[0065] 4) The prediction results under different closing initial phase angles are stored and organized to provide support for subsequent optimization analysis.

[0066] The purpose of this step is to use the trained neural network model to predict the response of the distribution transformer winding under given short-circuit conditions. By iterating through different closing initial phase angles, the prediction results of winding stress, deformation, and short-circuit impact coefficient in the entire operating range can be obtained, providing a basis for determining the optimal initial phase angle.

[0067] Step S50: Setting the allowable threshold of winding stress and deformation, and finding the optimal closing initial phase angle range that meets the requirements

[0068] The specific implementation is as follows:

[0069] 1) According to the structural strength and insulation capacity of the distribution transformer winding, the allowable threshold of winding stress and deformation is set. For example, the allowable threshold of winding stress can be set to 100 MPa, and the allowable threshold of winding deformation can be set to 2 mm.

[0070] 2) Using the prediction results obtained in step S40, iterate through different closing initial phase angle values, and select the initial phase angle range that meets the allowable threshold of winding stress and deformation.

[0071] 3) For the initial phase angle range that meets the requirements, further analyze the corresponding short-circuit impact coefficient to ensure that the impact coefficient also meets the design requirements in this range.

[0072] 4) Considering the limitations of winding stress, deformation, and short-circuit impact coefficient, determine the final closing initial phase angle allowable range.

[0073] The purpose of this step is to determine the optimal range of closing initial phase angles that meet the conditions based on the prediction results, winding structural strength, and performance requirements. By setting reasonable thresholds, it can be ensured that the winding will not suffer excessive stress and deformation when a short circuit occurs, ensuring safe and reliable operation.

[0074] Step S60: Further optimize and determine the final closing initial phase angle allowable range considering the closing time deviation of the actual test equipment

[0075] Specific implementation is:

[0076] 1) Analyze the mechanical error of the actual test equipment (such as grid power or short-circuit generator set) during the closing process to determine its closing time deviation range, which is usually around ±1ms, equivalent to ±10° of initial phase angle deviation.

[0077] 2) Based on the closing initial phase angle allowable range determined in step S50, further consider the closing time deviation of the actual test equipment, and appropriately widen the range to ensure that the requirements of winding stress, deformation, and short-circuit impact coefficient can still be met in actual testing.

[0078] 3) For example, if the optimal initial phase angle range determined in step S50 is -190° to -250°, considering the ±10° deviation, the final allowable range can be adjusted to -180° to -260° to improve the operability of the test.

[0079] 4) After this optimization, the final determined closing initial phase angle allowable range is obtained, which provides a basis for subsequent test implementation.

[0080] The purpose of this step is to further optimize and determine the final allowable range of closing initial phase angles in combination with the characteristics of the actual test equipment. Considering that there will inevitably be some closing time deviation in actual testing, appropriately widening the initial phase angle range can improve the success rate of the test and ensure that the test results meet the actual carrying capacity of the winding.

[0081] Step S70: Grid sampling within the determined closing initial phase angle range to obtain discrete initial phase angle values

[0082] Specific implementation is:

[0083] 1) In the closing initial phase angle allowable range determined in step S60, sampling is performed according to a certain grid step (such as 5°), and a series of discrete initial phase angle values are obtained, for example, -180°, -185°, -190°, -195°, -200°, -205°, -210°, -215°, -220°, -225°, -230°, -235°, -240°, -245°, -250°, -255°, -260°.

[0084] 2) For each discrete initial phase angle value, the short-circuit test model trained in step S40 is used for prediction to obtain the corresponding winding stress, deformation, and short-circuit impact coefficient.

[0085] 3) The above prediction results are sorted and stored to provide data support for subsequent comprehensive evaluation.

[0086] The purpose of this step is to perform grid sampling in the determined closing initial phase angle allowable range to obtain a series of discrete initial phase angle values. By using the trained neural network model to predict these discrete initial phase angle values, detailed data of winding performance indicators in the entire allowable range can be obtained, which provides a basis for the final optimization decision.

[0087] Step S80: Defining a comprehensive evaluation function, calculating the comprehensive score of each discrete closing initial phase angle, and selecting the optimal range

[0088] DETAILED DESCRIPTION

[0089] 1) A comprehensive evaluation function F is defined, which integrates the winding stress σ, deformation δ, and short-circuit impact coefficient K_s into one formula, as follows:

[0090] F = w1 × (σ / σ_max) + w2 × (δ / δ_max) + w3 × (K_s / K_s,max)

[0091] Where w1, w2, and w3 are the weight coefficients of each index, and σ_max, δ_max, and K_s,max are the allowable upper limit values of each index. By default, w1, w2, and w3 = 1 / 3.

[0092] 2) Using the winding stress, deformation, and short-circuit impact coefficient data obtained in step S70 for each discrete initial phase angle, the comprehensive evaluation function F is substituted to calculate the comprehensive score of each initial phase angle.

[0093] 3) The comprehensive scores of all initial phase angles are sorted, and the range formed by the top few initial phase angle values (such as the top 3-5) with the best scores is selected as the final optimized closing initial phase angle allowable range.

[0094] The purpose of this step is to comprehensively consider three key indicators—winding stress, deformation, and short-circuit impact coefficient—to define a comprehensive evaluation function. By iterating through the prediction results under different initial phase angles, the optimal range of initial phase angles for overall performance is found. This multi-indicator fusion optimization method ensures that the final determined closing initial phase angle meets both the winding structure strength requirements and minimizes short-circuit impact, thereby improving the operational reliability of the distribution transformer.

[0095] To better implement this invention, the specific embodiments of this invention will be described in more detail below with reference to specific formulas:

[0096] Step S10: Collect a large amount of short-circuit test data of distribution transformers.

[0097] First, it is necessary to collect relevant data on short-circuit tests of distribution transformers of different specifications and models. For each type of distribution transformer, the following key parameters need to be obtained:

[0098] 1) Short-circuit loop parameters:

[0099] Resistance R of the short-circuit loop sc and inductor L sc These are key factors affecting the magnitude of the short-circuit current. Accurate measurement of these parameters is necessary for subsequent performance prediction.

[0100] 2) Initial phase angle θ upon closing:

[0101] The initial phase angle at closing determines the peak value of the short-circuit current. Therefore, different initial phase angle conditions need to be covered in the test, such as θ = -10°, -70°, -130°, -190°, -250°, -310°, etc.

[0102] 3) Winding electromagnetic distribution:

[0103] During a short circuit, a complex electromagnetic field distribution is generated inside the winding, requiring measurement of the magnetic flux density B distribution at various parts of the winding. This parameter is closely related to the short-circuit electrodynamic force and winding stress.

[0104] 4) Winding electrodynamic distribution:

[0105] The flow of short-circuit current inside the winding generates a huge electrodynamic force, requiring measurement of the electrodynamic force F distribution at various parts of the winding. This parameter is a key basis for predicting winding deformation and stress.

[0106] 5) Winding stress and deformation:

[0107] Under short-circuit conditions, the winding will generate stress σ and deformation δ. These parameters need to be accurately measured as a basis for verifying the accuracy of the model predictions.

[0108] For repeated test data under the same working condition, statistical analysis method will be used to extract key characteristic values to enhance the representativeness and robustness of the data. Through this step, a high-quality test database covering a wide range of short-circuit working conditions can be established, laying a foundation for subsequent neural network model training.

[0109] Step S20: Preprocessing test data, constructing training data set

[0110] After collecting the original test data, it needs to be preprocessed to ensure data quality and availability. Specifically, it includes:

[0111] 1) Abnormal value detection and elimination:

[0112] For abnormal values and outliers in test data, statistical methods such as z-score are used to identify and eliminate them to ensure data representativeness.

[0113] 2) Missing value completion:

[0114] For individual missing values in test data, interpolation, regression, etc. can be used to complete them to ensure data integrity.

[0115] 3) Data normalization:

[0116] In order to eliminate the dimensional differences between different parameters, input features and output indicators need to be normalized to convert them to the [0, 1] interval. A commonly used normalization method is the minimum-maximum standardization:

[0117]

[0118] Where x norm is the normalized feature value, x min and x max are the minimum and maximum values of the feature respectively.

[0119] After the above preprocessing, the data can be organized into a standard training data set. The input features

[0120] X at least include short-circuit loop parameters R sc , L sc , initial phase angle θ, winding electromagnetic distribution B and electric force distribution F. Output indicators y include winding stress σ, deformation

[0121] δ and short-circuit impact coefficient K s . The form of the training data set is as follows:

[0122]

[0123] Through this step, the original experimental data is converted into a standard machine learning input-output format, preparing for subsequent neural network model training.

[0124] Step S30: Training the neural network model using the training data set

[0125] Based on the training data set constructed in step S20, a high-performance prediction model will be trained using neural network technology. The specific implementation steps are as follows:

[0126] 1) Select the appropriate neural network architecture:

[0127] After repeated experiments, a 4-layer fully connected network was finally selected as the basic architecture. The number of input layer nodes of this network is equal to the dimension of the input feature X, and the number of output layer nodes is equal to the dimension of the output index

[0128] y. The number of hidden layer nodes can be optimized through grid search.

[0129] 2) Define the training objective function:

[0130] The square loss function is used as the training objective, which has the mathematical form:

[0131]

[0132] where θ represents the parameters of the neural network, is the predicted output of the model for the i-th sample, and n is the total number of training samples.

[0133] 3) Use the backpropagation algorithm to optimize parameters:

[0134] In order to minimize the loss function L(θ), the backpropagation algorithm based on gradient descent is used to optimize and update the network parameters. Specifically, in each iteration, the gradient of the loss function with respect to the parameters is calculated first, and then the parameters are adjusted according to the following update rule:

[0135]

[0136] where η is the learning rate, controlling the step size of parameter update.

[0137] 4) Use cross-validation to evaluate performance:

[0138] During training, cross-validation is used to evaluate the performance of the model on the training set and validation set, and appropriate adjustments are made to hyperparameters such as hidden layer node number, learning rate, etc., until the desired prediction accuracy is achieved.

[0139] 5) Obtain the trained and optimized neural network model:

[0140] After training and optimization through the above steps, a high-performance neural network model is finally obtained, denoted as the short-circuit test model, which is used for subsequent performance prediction.

[0141] Through this step, a neural network model that can accurately predict the short-circuit performance of distribution transformers is successfully trained, providing a solid foundation for subsequent optimization analysis.

[0142] Step S40: Predicting winding stress, deformation, and impact coefficient using neural network model

[0143] With the short-circuit test model trained in step S30, it can be used to predict the winding performance indicators of distribution transformers under given short-circuit conditions. The specific implementation is as follows:

[0144] 1) Input feature preparation:

[0145] For given short-circuit loop parameters R sc ,L sc , initial phase angle θ, winding electromagnetic distribution B, and electric force distribution F, etc. input features, standardization processing is needed to make them meet the input requirements of the neural network model.

[0146] 2) Model forward calculation:

[0147] Send the standardized input features X into the trained short-circuit test model for forward propagation calculation, and the output of the model corresponding to the predicted values of winding stress, deformation, and short-circuit impact coefficient, respectively.

[0148] 3) Traverse the initial phase angle:

[0149] By changing the value of the initial phase angle θ, the short-circuit test model in step 2 can be used to predict winding performance indicators under different initial phase angles. This process can use numerical methods such as binary search to improve calculation efficiency.

[0150] 4) Store the prediction results:

[0151] Store and organize the prediction results under different initial phase angles for subsequent optimization analysis.

[0152] Through this step, the trained neural network model can be used to effectively predict the winding stress, deformation, and short-circuit impact coefficient of distribution transformers under various short-circuit conditions, providing a basis for determining the optimal initial phase angle.

[0153] Step S50: Set the allowable threshold of winding stress and deformation, and find the optimal initial phase angle range that meets the requirements

[0154] Based on the performance prediction results obtained in step S40, the optimal closing initial phase angle range needs to be further determined. The specific implementation is as follows:

[0155] 1) Set the threshold of winding performance index:

[0156] According to the structural strength and insulation capacity of the distribution transformer winding, set the allowable upper limit σ max = 100 MPa of winding stress and the allowable upper limit δ max = 2 mm of winding deformation. The setting of these thresholds needs to consider sufficient safety margin.

[0157] 2) Screen the initial phase angle range that meets the requirements:

[0158] Using the prediction results obtained in step S40, traverse different closing initial phase angles θ, and screen the initial phase angle range that meets the conditions σ ≤ σ max and δ ≤ δ max . This process can use efficient algorithms such as binary search.

[0159] 3) Analyze the short-circuit impact coefficient:

[0160] For the initial phase angle range that meets the winding stress and deformation requirements, the corresponding short-circuit impact coefficient K s also needs to be analyzed to ensure that this index also meets the design requirements. If necessary, optimization methods such as linear programming can be used to minimize K s .

[0161] 4) Determine the final closing initial phase angle range:

[0162] Considering the limiting conditions of winding stress, deformation, and short-circuit impact coefficient, the final closing initial phase angle allowable range that meets the requirements is determined.

[0163] Through this step, based on the performance prediction, the optimal closing initial phase angle range is determined according to the structural strength and performance requirements of the distribution transformer, providing a scientific basis for subsequent test implementation.

[0164] Step S60: Further optimize and determine the final closing initial phase angle allowable range considering the closing time deviation of the actual test equipment

[0165] Based on the determination of the initial phase angle allowable range in step S50, the characteristics of the actual test equipment also need to be considered to further optimize and determine the final closing initial phase angle range. The specific implementation is as follows:

[0166] 1) Analyze the closing time deviation of the test equipment:

[0167] Through actual tests, the mechanical errors of different types of test equipment (such as grid power supply or short-circuit generator set) in the closing process are statistically analyzed, and it is determined that the closing time deviation range is usually about ±1ms, which is equivalent to a phase angle deviation of ±10°.

[0168]

[0169] 2) Optimize the allowable range of initial phase angle:

[0170] On the basis of the closing initial phase angle range determined in step S50, the range is appropriately widened to compensate for the initial phase angle deviation that may occur in actual tests. For example, if the range obtained in step S50 is -190° to -250°, considering a deviation of ±10°, the final allowable range is adjusted to -180° to -260°.

[0171] 3) Determine the final closing initial phase angle range:

[0172] After this step of optimization, the final closing initial phase angle allowable range is obtained, which provides a basis for subsequent test implementation. This range not only considers the requirements of winding performance indicators, but also takes into account the characteristics of actual test equipment, improving the operability of the test.

[0173] The relevant variables are explained as follows: R sc : equivalent resistance of short-circuit loop; L sc : equivalent inductance of short-circuit loop; θ: closing initial phase angle; B: magnetic flux density distribution inside the winding;

[0174] F: electric force distribution inside the winding; σ: stress borne by the winding; δ: deformation of the winding; K s : short-circuit impact coefficient; X: input feature matrix of neural network;

[0175] y: output indicator matrix of neural network; predicted output of neural network model;

[0176] θ: parameter of neural network; L: loss function of neural network; η: learning rate of neural network; σ max : allowable upper limit of winding stress; δ max : allowable upper limit of winding deformation; (i): represents the i-th sample; t: represents the t-th iteration; n: total number of training samples; x min ,x max : minimum and maximum values of feature x.

[0177] Specifically, the principle of the present application is:

[0178] ​1. Establish a high-quality short-circuit test database: First, extensive collection of test data of different types of distribution transformers under various short-circuit conditions, including short-circuit loop parameters, initial phase angle, winding electromagnetic distribution, electric force distribution, and winding stress, deformation and other indicators. These data cover a wide range of short-circuit fault conditions, providing a solid foundation for subsequent neural network modeling. For multiple repeated test data under the same conditions, statistical analysis method is used to extract key characteristic values to enhance the representativeness and robustness of the data. Through this step, a high-quality test database covering a wide range of short-circuit conditions is established.

[0179] 2. Performance prediction using neural networks: Based on the test database established above, a high-performance prediction model short-circuit test model is trained using neural network technology. The input of the model includes short-circuit loop parameters, initial phase angle, winding electromagnetic distribution and electric force distribution, and the output is winding stress, deformation and short-circuit impact coefficient and other key performance indicators. In the training process, a 4-layer fully connected network architecture with strong prediction ability is selected, and the square loss function is used as the optimization objective. Through the back propagation algorithm, the network parameters are updated, and the hyperparameters are adjusted continuously using cross-validation, and finally a high-precision neural network model is obtained. Using this trained short-circuit test model, the winding performance of distribution transformers under various short-circuit conditions can be effectively predicted, providing a solid data foundation for subsequent optimization analysis.

[0180] 3. Determine the optimal initial phase angle range: After the prediction results of the neural network model are obtained, the allowable threshold of winding stress and deformation is set, and by traversing different initial phase angle values, the initial phase angle range that meets these threshold requirements is selected. This process can use an efficient binary search algorithm. For the initial phase angle range that meets the winding stress and deformation requirements, the corresponding short-circuit impact coefficient is also analyzed to ensure that this indicator also meets the design requirements. If necessary, linear programming and other optimization methods can be used to minimize the impact coefficient. Finally, considering the restriction conditions of winding stress, deformation and short-circuit impact coefficient, the final initial phase angle allowable range is determined.

[0181] 4. Optimization for test equipment characteristics: Based on the determination of the optimal initial phase angle range, the time deviation that may exist in the actual test equipment during the closing process is also considered, which is usually around ±1ms, equivalent to ±10° of initial phase angle deviation. In order to improve the operability of the test, the initial phase angle allowable range determined in the previous step is appropriately relaxed to ensure that the winding performance requirements can still be met in actual tests. For example, if the original range is -190° to -250°, considering the ±10° deviation, it is adjusted to -180° to -260°.

[0182] By this step optimization, the final determination of the closing initial phase angle to meet the test equipment characteristics of the allowable range, for the subsequent test implementation to provide a scientific basis.

[0183] Two specific embodiments of the present application are provided below, which are detailed simulation analysis and test verification for a 315 kVA distribution transformer. The simulation calculation of the impact coefficient is carried out by using PSCAD software, it can be seen that the impact coefficient depends on the loop inductance and resistance, and is directly related to the initial phase angle of the power closing.

[0184] Table 1 Closing initial phase angle-impact coefficient simulation data table

[0185] switching initial phase angle impulse coefficient -10 1.57 -70 1.29 -130 1.15 -190 1.48 -250 1.51 -310 1.07

[0186] The relevant simulation waveforms are shown in Figures 2-7 The following is the implementation process and results of determining the optimal closing initial phase angle according to the present application.

[0187] 1. Short-circuit test data collection

[0188] First, a wide range of relevant data of different types of distribution transformers in large-capacity short-circuit tests are collected. Taking a 315 kVA transformer as an example, it mainly includes:

[0189] Short-circuit loop parameters:

[0190] According to the test measurement, the short-circuit loop resistance R sc of the transformer is 0.02Ω, and the short-circuit loop inductance L sc is 0.12H.

[0191] Closing initial phase angle:

[0192] Six different closing initial phase angle conditions are designed for the transformer, which are θ=-10°, -70°, -130°, -190°, -250°, and -310°, respectively.

[0193] Winding electromagnetic distribution:

[0194] Through finite element analysis, the magnetic flux density distribution B inside the winding of the transformer under short-circuit state is obtained. Taking t=0.01s as an example, the magnetic flux density distribution of the high-voltage and low-voltage windings is shown in Figure 1 .

[0195] Winding electromagnetic distribution:

[0196] Similarly, the winding electromagnetic distribution F of the transformer under the action of short-circuit is calculated by using finite element analysis. Figure 2 The winding electromagnetic distribution at t=0.01s is shown.

[0197] Winding stress and deformation:

[0198] The stress and deformation data of the transformer winding under different initial phase angles of closing are obtained through short-circuit test measurement

[0199] For example, when θ =-190°, the maximum stress of the winding is 85 MPa, and the maximum deformation is 1.2 mm.

[0200] The above test data covers a wide range of short-circuit conditions, providing a solid foundation for subsequent neural network modeling and performance analysis.

[0201] 2. Neural network model training

[0202] Based on the aforementioned collected test data, a neural network model short-circuit test model for predicting the short-circuit performance of distribution transformers is constructed. The specific training process is as follows:

[0203] (1) Data preprocessing

[0204] First, the collected test data is preprocessed, including outlier detection, missing value completion, and data normalization operations. The normalization method used is the min-max normalization method:

[0205]

[0206] where x norm is the normalized feature value, x min and x max are the minimum and maximum values of the feature, respectively.

[0207] (2) Construction of training data set

[0208] The preprocessed input features (short-circuit loop parameters, initial phase angle of closing, winding electromagnetic distribution, and electric force distribution) and output indicators (winding stress, deformation, and short-circuit impact coefficient) are organized into a standard training data set for the neural network model to learn.

[0209] (3) Network architecture selection

[0210] After repeated experiments, a 4-layer fully connected neural network is finally selected as the basic architecture. The input layer node number of the network is 10, corresponding to the dimension of the input features; the hidden layer node numbers are 50, 50, and 30; and the output layer node number is 3, corresponding to the winding stress, deformation, and short-circuit impact coefficient, respectively.

[0211] (4) Model training

[0212] The square loss function is used as the training target, and the backpropagation algorithm is used to optimize and update the network parameters. The specific loss function expression is:

[0213]

[0214] where θ represents the parameters of the neural network, is the predicted output of the model for the i-th sample, and n is the total number of training samples.

[0215] During the training process, techniques such as momentum term and L2 regularization are used to improve the convergence speed and generalization performance of the model. At the same time, through grid search, the optimal learning rate η = 0.001 and batch size batch_size = 32 and other hyperparameters are determined.

[0216] (5) Model evaluation

[0217] The performance of the model on the training set and the validation set is evaluated using cross-validation. The results show that on the validation set, the average prediction error of the winding stress is 5.2%, the average prediction error of the deformation is 7.1%, and the average prediction error of the short-circuit impact coefficient is 3.9%.

[0218] In summary, a high-performance neural network model short-circuit test model is successfully trained, which can accurately predict the key performance indicators of distribution transformers under large-capacity short-circuit conditions.

[0219] 3. Switching initial phase angle optimization

[0220] With the aforementioned trained short-circuit test model, it can be used to predict the winding performance of a 315 kVA distribution transformer under different switching initial phase angles and determine the optimal initial phase angle range. The specific process is as follows:

[0221] (1) Performance prediction

[0222] First, the short-circuit loop parameters R sc = 0.02 Ω, L sc = 0.12 H, 6 kinds of switching initial phase angles

[0223] θ and the aforementioned obtained winding electromagnetic distribution B and electric force distribution F and other input features are sent to the short-circuit test model for forward calculation and prediction.

[0224] The output results include: winding stress winding deformation short-circuit impact coefficient

[0225]

[0226] Table 2 Predicted results under different switching initial phase angles

[0227] switching initial phase angle θ (°) maximum winding stress σ (MPa) maximum winding deformation δ (mm) short-circuit impulse coefficient Ks -10 105 2.3 1.57 -70 92 1.8 1.51 -130 88 1.5 1.49 -190 85 1.2 1.48 -250 90 1.7 1.46 -310 98 2.1 1.07

[0228] The predicted results under different initial phase angles are listed in Table 2. It can be seen that the winding performance indicators show obvious differences with the change of initial phase angle. For example, when θ = -10°, the short-circuit impact factor reaches the maximum value of 1.57, while when θ = -310°, the impact factor is only 1.07. This shows that the selection of initial phase angle has an important influence on the short-circuit performance of the transformer.

[0229] (2) Performance indicator threshold setting

[0230] According to the structural strength and insulation capacity of the distribution transformer winding, the following allowable upper limits of performance indicators are set:

[0231] Winding stress allowable upper limit σ max = 100 MPa

[0232] Winding deformation allowable upper limit δ max = 2 mm

[0233] These threshold settings take into account sufficient safety margins to ensure reliable operation of the transformer under short-circuit action.

[0234] (3) Optimal initial phase angle range determination

[0235] Using the prediction results of the short-circuit test model, different initial phase angles θ are traversed to screen the initial phase angle range that meets the condition After calculation, the range is -190° ≤ θ ≤ -250°.

[0236] Further analysis of the short-circuit impact factor corresponding to this initial phase angle range shows that its value is less than 1.5, meeting the design requirements. Therefore, -190° ≤ θ ≤ -250° is finally determined as the optimal initial phase angle range for the 315 kVA transformer.

[0237] (4) Test equipment characteristic optimization

[0238] In actual large-capacity short-circuit tests, due to mechanical errors in the closing process of test equipment (such as grid power or short-circuit generator set), the closing time deviation is usually around ±1 ms, which is equivalent to an initial phase angle deviation of ±10°.

[0239] In order to improve the operability of the test, the aforementioned optimal initial phase angle range -190° ≤ θ ≤ -250° is appropriately relaxed to -180° ≤ θ ≤ -260°. In this way, even if a ±10° initial phase angle deviation occurs in actual tests, it can be ensured that the winding performance indicators still meet the requirements.

[0240] Through the above steps, the optimal closing initial phase angle range of the 315 kVA distribution transformer in the large-capacity short-circuit test is finally determined as -180°≤θ≤-260°, which provides a reliable basis for subsequent test implementation.

[0241] 4. Technical effect verification

[0242] In order to verify the effectiveness of the method of the present application, a series of comparative tests and analysis are carried out, as follows:

[0243] (1) Short-circuit test under different initial phase angles

[0244] Under the conditions of θ=-10°, -70°, -130°, -190°, -250° and -310°, the 315 kVA distribution transformer is subjected to a large-capacity short-circuit test, and the winding stress, deformation and short-circuit impact coefficient are measured. The test results are shown in Table 1.

[0245] As can be seen from Table 1, there are obvious differences in the short-circuit performance of the transformer under different initial phase angles. For example, when θ=-10°, the maximum winding stress reaches 95 MPa, which exceeds the allowable upper limit of 100 MPa; while when θ=-190°, the winding stress is only 85 MPa, which meets the requirements. Similarly, the short-circuit impact coefficient also shows a large difference under different initial phase angles, reaching a maximum value of 1.57 when θ=-10°, and only 1.07 when θ=-310°.

[0246] This fully demonstrates that the selection of the closing initial phase angle has an important influence on the performance of the distribution transformer under short-circuit conditions, and if not properly selected, it may cause serious damage to the winding. Therefore, it is crucial to determine the optimal closing initial phase angle range for the safe and stable operation of the transformer.

[0247] (2) Prediction performance of the method of the present application

[0248] The optimal initial phase angle range -180°≤θ≤-260° determined in step 3 is applied to the actual short-circuit test, and the winding performance indicators of the transformer in this range are measured. The test results are shown in Table 3, and are compared with the prediction results of the neural network model.

[0249] Table 3 Comparison of neural network model prediction and actual measurement data

[0250]

[0251] As can be seen from Table 3, the prediction results of the method based on the neural network of the application are in good agreement with the actual test data. For example, when θ = -190°, the maximum stress of the winding predicted by the neural network model is 85 MPa, which is very close to the 84 MPa measured by the test, with an error of only 5.2%; the maximum deformation predicted is 1.15 mm, which is also very close to the 1.20 mm measured by the test, with an error of 7.1%. Similarly, the prediction error of the short-circuit impact coefficient is only 3.9%.

[0252] This shows that the neural network model proposed by the application can accurately predict the key performance indicators of the distribution transformer under the condition of large-capacity short circuit, providing a reliable basis for determining the optimal closing initial phase angle range. Compared with traditional theoretical analysis and empirical formula, this method is more scientific, systematic and accurate, greatly improving the accuracy of prediction.

[0253] (3) Limitations of existing methods

[0254] In order to further verify the superiority of the method of the application, the short-circuit performance of the 315 kVA transformer was predicted and analyzed by using the traditional theoretical analysis and empirical formula.

[0255] According to electromagnetic theory, the electromagnetic force distribution of the transformer winding under short-circuit condition can be calculated. Assuming that the axial and radial forces of the winding reach the critical value, the critical buckling load can be estimated. Based on the linear buckling analysis theory, the critical buckling load of the low-voltage winding is calculated to be -168.86 kN / m.

[0256] According to the empirical formula, the impact coefficient K s of the short-circuit current can be expressed as:

[0257]

[0258] Where I peak is the peak value of the short-circuit current, and I rms is the effective value of the short-circuit current. Using this formula, when θ = -190°, the impact coefficient K s = 1.52.

[0259] The prediction results of the above theoretical analysis and empirical formula are compared with the prediction of the neural network model of the application and the actual test data, as shown in Table 4.

[0260] Table 4 Comparison of results of different prediction methods

[0261] prediction method critical buckling load of low-voltage winding (kN / m) maximum radial force (kN / m) short-circuit impulse coefficient Ks (θ = -190°) theoretical analysis -168.86 -43.4019 - empirical formula - - 1.52 neural network - -46.2893 1.48 measured value - -43.4019 1.43

[0262] As can be seen from Table 4, there is a certain deviation between the prediction results of theoretical analysis and empirical formula and the measured data. For example, the critical buckling load of the low-voltage winding obtained by theoretical analysis is-168.86 kN / m, while the maximum radial force measured is only-43.4019 kN / m, with a safety margin of 3.89. This shows that the critical value given by the theoretical analysis is too conservative and has a large gap with the actual situation.

[0263] Similarly, the short-circuit impact coefficient predicted by the empirical formula is 1.52, while the predicted value of the neural network model is 1.48, which is closer to the measured value of 1.43. This shows that the neural network model trained based on a large amount of experimental data can more accurately reflect the actual performance of the transformer under short-circuit conditions.

[0264] Compared with the prior art, the method of the present application has the following significant advantages:

[0265] 1. A high-quality test database covering a wide range of short-circuit conditions is established: The present application widely collects test data of different types of distribution transformers under various short-circuit conditions and initial phase angles when closing, including winding electromagnetic distribution, electric force distribution, and key response indicators such as stress and deformation. This lays a solid foundation for subsequent neural network modeling.

[0266] 2. Advanced neural network technology is used for performance prediction: The present application trains a high-performance neural network model short-circuit test model that can accurately predict the stress, deformation and short-circuit impact coefficient of the winding of the distribution transformer under various short-circuit conditions. This greatly improves the accuracy and reliability of performance prediction and provides strong support for the determination of the optimal initial phase angle.

[0267] 3. Optimization of test equipment characteristics: When determining the optimal initial phase angle range, the present application takes into account the influence of the closing time deviation of actual test equipment (such as power grid power supply, short-circuit generator set), appropriately widens the allowable range, and improves the operability of the test. This is conducive to shortening the test preparation work and improving the test success rate.

[0268] 4. Comprehensive optimization of winding performance indicators: When determining the optimal initial phase angle range, the present application not only considers the limiting conditions of winding stress and deformation, but also takes into account the short-circuit impact coefficient, striving to further optimize the performance under the premise of meeting the requirements of various indicators. This ensures the safe and stable operation of the transformer under short-circuit conditions.

[0269] In summary, the method for determining the initial phase angle of the closing of the large-capacity short-circuit current of the winding of the distribution transformer proposed by the present application fully utilizes the advantages of big data analysis and machine learning technology, can effectively predict the key performance of the transformer under short-circuit conditions, and provides a scientific basis for optimizing the test conditions.

Claims

1. A method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer, characterized in that, Includes the following steps: S10. Collect a large amount of short-circuit test data of distribution transformers, including: short-circuit circuit parameters with different historical values ​​and short-circuit test data under various short-circuit conditions with different initial closing phase angles in history. S20. Preprocess the short-circuit test data to construct a training dataset; S30. Train a neural network model using the training dataset to obtain a short-circuit test model; S40. Using the short-circuit test model, for given short-circuit loop parameters, initial closing phase angle, winding electromagnetic distribution, and electrodynamic distribution, by traversing different historical initial closing phase angle values ​​in S10, the corresponding first winding stress, deformation, and impact coefficient are predicted and calculated. S50. Set the allowable threshold ranges for the initial second winding stress, deformation, and impact coefficient. Based on the neural network prediction results, find the optimal initial phase angle range for closing that meets the requirements. S60. Based on the closing time deviation of the test equipment in the short-circuit test, further optimize and determine the final allowable range of the initial phase angle for closing; S70. Within the determined allowable range of the initial phase angle of closing, perform grid sampling to obtain multiple discrete initial phase angle values ​​of closing within the range. For each discrete initial phase angle value of closing, use the short-circuit test model to predict the corresponding stress, deformation, and impact coefficient of the third winding. S80. Define a comprehensive evaluation function that integrates the stress, deformation, and impact coefficient of the third winding, calculates the comprehensive score of each discrete initial phase angle of closing, and takes the range formed by multiple initial phase angles of closing with the best scores as the optimized initial phase angle of closing.

2. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The short-circuit test data includes at least the winding electromagnetic distribution, electrodynamic distribution, structural stress, and deformation data.

3. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, When training the neural network model using the training dataset, the input to the training is a first vector formed by short-circuit loop parameters, initial phase angle of closing, winding electromagnetic distribution, and electrodynamic distribution. The output to the training is a second vector formed by winding stress, deformation, and impact coefficient.

4. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The various short-circuit conditions include three-phase symmetrical short circuit and single-phase short circuit; among them, the short-circuit test data for each short-circuit condition are short-circuit test data covering different initial phase angles of closing.

5. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The neural network model uses a multilayer perceptron or a convolutional neural network.

6. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, In step S50, the allowable threshold range for winding stress is 0~100MPa, and the allowable threshold range for deformation is 0~2mm.

7. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The comprehensive evaluation function is the normalized average value of the winding stress, deformation, and impact coefficient.

8. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The method of traversing different initial phase angle values ​​for closing is the binary search algorithm.

9. The method for determining the initial phase angle of closing the large-capacity short-circuit withstand test current of a distribution transformer according to claim 1, characterized in that, The methods for preprocessing the experimental data include outlier detection and removal, missing value completion, and data normalization.

Images