Transformer winding deformation online diagnosis method and system based on machine learning

Abstract

The invention discloses a transformer winding deformation online diagnosis method and system based on machine learning, and relates to the technical field of transformer winding monitoring, and the method comprises the steps: collecting the actual normal operation data of a transformer: voltage, current, power and temperature, and building a 1: 1 simulation model to simulate the winding deformation condition, and obtaining operation data under the winding deformation working condition. And preprocessing the acquired data. And obtaining the input of model training by solving the deviation of the time series before and after the winding deformation. The winding fault position adopts one-hot coding as the output of model training. A model is trained through a multi-layer perceptron neural network, an error between output and a target is calculated by using a loss function, and model parameters are updated by using an Adam optimizer. According to the transformer winding deformation on-line diagnosis method based on machine learning, the machine learning technology is utilized, and early detection and diagnosis of winding deformation faults are realized through on-line monitoring of transformer operation data.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer winding monitoring, and in particular to an online diagnosis method and system for transformer winding deformation based on machine learning. Background Art

[0002] Transformers are important equipment in power systems, and their normal operation directly affects the safety and reliability of power systems. Transformer winding deformation is one of the important causes of transformer failure. Common deformation forms include winding extrusion, stretching, and offset. These deformations may cause local overheating, voltage imbalance, and insulation aging. In severe cases, they may even cause transformer damage, leading to large-scale power accidents. Therefore, timely and accurate diagnosis of transformer winding deformation failures is of great significance to ensure the safe operation of power systems.

[0003] Traditional transformer winding deformation monitoring methods mainly rely on offline detection technologies, such as frequency response analysis (FRA) and short-circuit impedance method. Although these methods can reflect the mechanical state of the winding to a certain extent, they require power outages for maintenance, which not only increases maintenance costs but also affects the normal operation of the power grid. In addition, the data collection and analysis process of these offline detection methods is relatively complicated, and the accuracy and reliability of the diagnostic results are also subject to certain limitations.

[0004] With the development of modern science and technology, machine learning technology has been widely used in various fields. By analyzing and learning a large amount of data, machine learning can discover the potential patterns and laws in the data, providing new ideas and methods for fault diagnosis. In particular, deep learning technology, by building a multi-layer neural network, can handle complex nonlinear relationships and achieve efficient identification and classification of faults. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: how to timely discover potential winding deformation faults and reduce the inconvenience and losses caused by power outage maintenance.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: an online diagnosis method for transformer winding deformation based on machine learning, comprising:

[0008] Collect actual operating data of the transformer under normal operating conditions;

[0009] A simulation model is established based on the actual structure and operating conditions of the transformer to simulate the winding deformation;

[0010] Simulate the winding deformation condition in the simulation model and obtain the operating data under the deformation condition;

[0011] Preprocessing the collected actual operation data and operation data under deformation conditions;

[0012] By analyzing the time series data before and after the winding deformation, the deviation value at each moment is calculated as the input feature of the model training;

[0013] The winding fault location is processed using one-hot encoding as the output label for model training;

[0014] Build a multi-layer perceptron neural network model and use the preprocessed data and features to train the model;

[0015] The error between the model output and the target is calculated through the loss function, and the model parameters are updated using the Adam optimizer;

[0016] Cross-validate and evaluate the model during the training process to verify the performance and generalization ability of the model;

[0017] Use the test data set to test the model, calculate the model's evaluation indicators, and adjust the model structure and parameters;

[0018] The trained model is used to perform online diagnosis of transformer winding deformation.

[0019] As a preferred solution of the online diagnosis method of transformer winding deformation based on machine learning described in the present invention, wherein: the transformer operation data includes voltage, current, power and temperature;

[0020] The simulation of the winding deformation includes using 3D modeling software to draw a 1:1 scale three-dimensional model of the transformer and windings of each voltage level;

[0021] The obtaining of the operating data under deformation includes performing finite element analysis on the drawn 3D model using finite element analysis software, setting winding deformations at different positions using the finite element software, and simulating the operating conditions of the transformer under winding deformation.

[0022] As a preferred solution of the online diagnosis method of transformer winding deformation based on machine learning described in the present invention, the preprocessing is expressed as:

[0023]

[0024] Where x[t] is the input signal, y[t] is the filtered output signal, t is the position of the missing value, t1 and t2 are the positions of the adjacent known values, and x1 and x2 are the values ​​of the adjacent known values.

[0025] As a preferred solution of the online diagnosis method for transformer winding deformation based on machine learning described in the present invention, the time series data before and after the winding deformation includes the original time series data {x1, x2, ..., x n} and the deformed time series data {y1,y2,...,y n};

[0026] The deviation value at each moment is expressed as,

[0027] d t =y t -x t

[0028] Among them, d t is the deviation value at time t, x t is the value of the original data at time t, y t is the value of the deformed data at time t;

[0029] The one-hot encoding is represented as,

[0030]

[0031] Among them, i,j Represents the jth position of the fault label at the i-th position in the one-hot encoding.

[0032] As a preferred solution of the online diagnosis method of transformer winding deformation based on machine learning described in the present invention, wherein: the multi-layer perceptron neural network model includes an input layer, a hidden layer and an output layer;

[0033] The hidden layer is represented as,

[0034] h i =σ(∑ j w ij x j +b i )

[0035] Where hi is the output of the i-th node in the hidden layer, wij is the weight, xj is the input, bi is the bias, and σ(·) is the activation function;

[0036] The output layer is represented as,

[0037] y k =σ(∑ i w ki h i +b k )

[0038] Among them, yk is the output of the kth node in the output layer, hi is the output of the hidden layer, wki is the weight, and bk is the bias;

[0039] The loss function is expressed as,

[0040]

[0041] Where n is the number of samples; y i is the true value of the i-th sample; is the predicted value of the ith sample.

[0042] As a preferred solution of the online diagnosis method for transformer winding deformation based on machine learning described in the present invention, wherein: the gradient of the loss function relative to each parameter is calculated using the chain rule, starting from the output layer, the gradient is propagated backward layer by layer, and the weights and biases of the neural network are updated;

[0043] Compute the gradient of the output layer’s loss with respect to its input, expressed as,

[0044]

[0045] Among them, δ (L) is the error term of the output layer; L is the loss function; z (L) is the linear combination of the output layer; σ' is the derivative of the activation function; a (l) is the activation value of the lth layer; W( l) is the weight matrix of the lth layer; b (l) is the bias vector of the lth layer; σ is the activation function;

[0046] δ (L) =(W (l+1) ) T δ (l+1) ⊙σ′(z (l) )

[0047]

[0048] Where η is the learning rate.

[0049] As a preferred solution of the online diagnosis method of transformer winding deformation based on machine learning described in the present invention, wherein: the input of the training model is 18 inputs of three-phase voltage and current, 6 inputs of three-phase power and 1 input of top oil temperature;

[0050] The output of model training is 9 outputs of three-phase winding deformation positions;

[0051] The input data set for model training uses time series as input, where the length of each time series of voltage and current is one cycle of the waveform, containing 50 points. Data processing is performed on the waveform of this cycle to capture the changes before and after the winding deformation.

[0052] The input dataset contains 70 600-dimensional time series; the output dataset contains 70 corresponding fault locations, using one-hot encoding;

[0053] 20% of the data set was selected as a test set and 12 samples; after the model training was completed, the 12 samples were used for diagnostic verification;

[0054] The three-phase voltage and current 18 inputs include the voltage and current of the high, medium and low voltage side of the A phase of the transformer; the voltage and current of the high, medium and low voltage side of the B phase of the transformer; the voltage and current of the high, medium and low voltage side of the C phase of the transformer;

[0055] The three-phase power six inputs include active and reactive power on the high, medium and low voltage sides of the transformer;

[0056] The nine outputs of the three-phase winding deformation positions include the high, medium and low voltage sides of the transformer A phase; the high, medium and low voltage sides of the transformer B phase; and the high, medium and low voltage sides of the transformer C phase.

[0057] Another object of the present invention is to provide an online diagnosis system for transformer winding deformation based on machine learning, which can solve the problem of ensuring security and confidentiality of data transmission in existing distribution network operations by constructing a remote secure communication system for distribution network operations, and the problem that using a third-party platform is not conducive to data confidentiality.

[0058] To solve the above technical problems, the present invention provides the following technical solutions: a transformer winding deformation online diagnosis system based on machine learning, comprising: a data acquisition module, a simulation model construction module, a deformation condition simulation module, a data preprocessing module, a feature calculation module, a label encoding module, a neural network training module, a model optimization module, a cross-validation module, a model testing module, a model adjustment module and an online diagnosis module; the data acquisition module is used to collect the actual operating data of the transformer under normal operating conditions; the simulation model construction module is used to establish a simulation model according to the actual structure and operating conditions of the transformer, and simulate the winding deformation; the deformation condition simulation module is used to simulate the winding deformation condition in the simulation model and obtain the operating data under the deformation condition; the data preprocessing module is used to preprocess the collected actual operating data and the operating data under the deformation condition; the feature The calculation module is used to calculate the deviation value at each moment by analyzing the time series data before and after the winding deformation, which is used as the input feature of the model training; the label encoding module is used to process the winding fault position using one-hot encoding as the output label of the model training; the neural network training module is used to construct a multi-layer perceptron neural network model, and use the preprocessed data and features for model training; the model optimization module is used to calculate the error between the model output and the target through the loss function, and use the Adam optimizer to update the model parameters; the cross-validation module is used to cross-validate and evaluate the model during the training process to verify the performance and generalization ability of the model; the model testing module is used to test the model using a test data set, calculate the evaluation index of the model, and adjust the model structure and parameters; the online diagnosis module is used to perform online diagnosis of transformer winding deformation through the trained model.

[0059] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the online diagnosis method for transformer winding deformation based on machine learning as described above are implemented.

[0060] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned online diagnosis method for transformer winding deformation based on machine learning.

[0061] Beneficial effects of the present invention: The online diagnosis method for transformer winding deformation based on machine learning provided by the present invention utilizes the actual operation data of the transformer, combined with the simulation model, to simulate the winding deformation, obtain the operation data under different working conditions, and realize the online diagnosis of transformer winding deformation through data preprocessing, feature extraction and neural network model training. This method not only improves the accuracy and efficiency of fault diagnosis, but also can perform real-time monitoring during the normal operation of the transformer, timely discover potential winding deformation faults, and reduce the inconvenience and losses caused by power outages and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0063] Figure 1 An MLP neural network structure diagram of an online diagnosis method for transformer winding deformation based on machine learning provided in one embodiment of the present invention.

[0064] Figure 2 A neural network training process diagram of an online diagnosis method for transformer winding deformation based on machine learning provided by one embodiment of the present invention.

[0065] Figure 3 A diagram of a technical solution for transformer fault diagnosis based on a neural network, which provides an online diagnosis method for transformer winding deformation based on machine learning, according to an embodiment of the present invention.

[0066] Figure 4 A three-dimensional simulation model diagram of a transformer for an online diagnosis method of transformer winding deformation based on machine learning provided by an embodiment of the present invention.

[0067] Figure 5 A primary and secondary voltage waveform diagram collected by an online diagnosis method for transformer winding deformation based on machine learning provided by an embodiment of the present invention.

[0068] Figure 6 A primary and secondary current waveform diagram collected by an online diagnosis method for transformer winding deformation based on machine learning provided by an embodiment of the present invention.

[0069] Figure 7 A timing input waveform diagram before data processing of an online diagnosis method for transformer winding deformation based on machine learning provided by an embodiment of the present invention.

[0070] Figure 8A timing input waveform diagram after data processing of an online diagnosis method for transformer winding deformation based on machine learning provided by an embodiment of the present invention.

[0071] Fig. 9 A loss function iteration process diagram of an online transformer winding deformation diagnosis method based on machine learning provided by an embodiment of the present invention.

[0072] Fig.10 A fault diagnosis model test result diagram of a method for online diagnosis of transformer winding deformation based on machine learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0074] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0075] Example 1

[0076] Reference Figure 1-Figure 3 , as an embodiment of the present invention, provides an online diagnosis method for transformer winding deformation based on machine learning, comprising:

[0077] S1: Use sensors and data acquisition equipment to acquire data from transformers in normal operation, including voltage, current, power and temperature, and use 3D modeling software SolidWorks to draw 1:1 scale 3D models of transformers and windings of various voltage levels. Combining 1:1 simulation models with actual data can better simulate the actual operating status of different transformer winding deformation conditions, making the model training data more comprehensive and realistic, thereby improving the model's fault detection accuracy.

[0078] S2: Use the finite element analysis software COMSOL to perform finite element analysis on the drawn 3D model. Use the finite element software to set the winding deformation at different positions, simulate the operating conditions of the transformer under winding deformation, and obtain relevant data.

[0079] S3: Preprocess the actual operation data and simulation data. Since the main frequency of the signal is 50 Hz, a low-pass filter is set to remove high-frequency noise in the data while retaining the low-frequency signal components.

[0080]

[0081] Where x[t] is the input signal and y[t] is the output signal after filtering. Since it is continuous data, the missing values ​​are filled by interpolation. Interpolation estimates missing values ​​by using the relationship between known data points, which is suitable for processing periodic data.

[0082]

[0083] Where t is the location of the missing value, t1 and t2 are the locations of the adjacent known values, and x1 and x2 are the values ​​of the adjacent known values.

[0084] S4: The time series deviation value is obtained to measure the difference between each moment before and after the winding deformation. The following is the formula and steps for calculating the deviation value:

[0085] Original time series data:

[0086] {x1,x2,...,x n}

[0087] Time series data after deformation:

[0088] {y1,y2,...,y n}

[0089] The calculation formula of the deviation value is:

[0090] d t =y t -x t

[0091] Among them, d t is the deviation value at time t, x t is the value of the original data at time t, y t is the value of the deformed data at time t. In the process of processing and calculation, the time series data deviation is used to capture the changes before and after the winding deformation. Such feature extraction helps the model to better identify the fault type.

[0092] One-hot encoding is used to convert categorical data into binary vectors, suitable for output labels in model training. Fault location {1, 2, …, 7}

[0093]

[0094] Among them, i,jRepresents the jth position of the fault label at the i-th position in the one-hot encoding.

[0095] S5: Build an MLP multi-layer perceptron neural network model. Design an appropriate network structure, including input layer, hidden layer and output layer. MLP consists of an input layer, one or more hidden layers and an output layer.

[0096] h i =σ(∑ j w ij x j +b i )

[0097] where hi is the output of the i-th node in the hidden layer, wij is the weight, xj is the input, bi is the bias, and σ(·) is the activation function.

[0098] The output layer nodes calculate the weighted sum of the hidden layer outputs and obtain the final output through the activation function:

[0099] y k =σ(∑ i w ki h i +b k )

[0100] Where yk is the output of the kth node in the output layer, hi is the output of the hidden layer, wki is the weight, and bk is the bias. This enables the model to learn from the data and find a suitable set of parameters to accurately predict or classify new data. The standard multi-layer perceptron model is used, but combined with the unique processing method of actual scenarios and data, as well as the use of specific optimization algorithms such as the Adam optimizer, the model has improved convergence speed and diagnostic accuracy.

[0101] S6: The loss function of a neural network is a function used to measure the difference between the predicted value and the true value. The lower the loss function value, the more accurate the predicted value of the neural network. The mean square error (MSE) is often used in regression problems, and its expression is:

[0102]

[0103] Where n is the number of samples; y i is the true value of the i-th sample; is the predicted value of the ith sample.

[0104] MSE calculates the average of the squared differences between the predicted value and the true value, reflecting the distance between the predicted value and the true value. The smaller the MSE value, the more accurate the model prediction.

[0105] S7: The process is to use the chain rule to calculate the gradient of the loss function relative to each parameter, starting from the output layer, and propagating the gradient backward layer by layer to update the weights and biases of the neural network. First, calculate the gradient of the output layer's loss relative to its input, expressed as:

[0106]

[0107] where δ (L) is the error term of the output layer; L is the loss function; z (L) is the linear combination of the output layer; σ' is the derivative of the activation function; a (l) is the activation value (output) of layer l; W (l) is the weight matrix of the lth layer; b (l) is the bias vector of the lth layer; σ is the activation function.

[0108] δ (L) =(w (L+1) ) T δ (L+1) ⊙σ′(z (l) )

[0109]

[0110] Where η is the learning rate. The Adam optimization algorithm provides an efficient and adaptive gradient descent method by combining the momentum method and the RMSProp algorithm, which significantly improves the convergence speed and stability of neural network training.

[0111] S8: In machine learning, dividing the dataset into training set and test set is a common method for model training and evaluation. The training set is used to train the model, while the test set is used to evaluate the performance of the model. Use 80% of the dataset as the training set and 20% as the test set.

[0112] K-fold cross validation (K=5) was performed on the training set to optimize the model parameters and verify the performance of the model. The steps of cross validation include: dividing the training set into K subsets. Using each subset as the validation set in turn, and the remaining subsets as the training set for model training and validation. The performance index of each validation was calculated, and the average was taken as the performance evaluation of the model. 12 samples in the test set were randomly selected, and the fault types of these samples were predicted. The predicted results were then compared with the actual fault types of these samples. The accuracy of the model was further verified.

[0113] The input of the training model is: 18 three-phase voltage and current inputs (voltage and current on the high, medium and low voltage side of the transformer phase A; voltage and current on the high, medium and low voltage side of the transformer phase B; voltage and current on the high, medium and low voltage side of the transformer phase C), 6 three-phase power inputs (active and reactive power on the high, medium and low voltage side of the transformer), and 1 top oil temperature input. The output of the model training is: 9 outputs of the three-phase winding deformation position (high, medium and low voltage side of the transformer phase A; high, medium and low voltage side of the transformer phase B; high, medium and low voltage side of the transformer phase C). The input data set for model training uses time series as input, where the length of each time series of voltage and current is one cycle of the waveform, containing 50 points, and then the waveform of this cycle is processed to capture the changes before and after the winding deformation. The input data set contains 70 600-dimensional time series. The output data set contains 70 corresponding fault locations, using one-hot encoding, which is equivalent to 70 7-dimensional data. 20% of the data set is selected as a test set and 12 samples. After the model training was completed, 12 samples were diagnosed and verified. The diagnosis results showed that the model trained by the proposed method can correctly diagnose the location of winding deformation faults.

[0114] Compared with traditional offline detection, the key data (such as voltage, current, power and temperature) required by the present invention can be obtained in real time during actual operation. Traditional diagnostic methods can usually only give a judgment on whether a fault exists, while the output of the present invention can provide more refined fault location information, helping maintenance personnel to find and repair problems more quickly.

[0115] Example 2

[0116] According to an embodiment of the present invention, a transformer winding deformation online diagnosis system based on machine learning is provided, comprising:

[0117] Data acquisition module, simulation model building module, deformation condition simulation module, data preprocessing module, feature calculation module, label encoding module, neural network training module, model optimization module, cross-validation module, model testing module, model adjustment module and online diagnosis module;

[0118] The data acquisition module is used to collect the actual operating data of the transformer under normal operating conditions;

[0119] The simulation model building module is used to build a simulation model according to the actual structure and operating conditions of the transformer and simulate the deformation of the winding;

[0120] The deformation condition simulation module is used to simulate the winding deformation condition in the simulation model and obtain the operation data under the deformation condition;

[0121] The data preprocessing module is used to preprocess the collected actual operation data and operation data under deformation conditions;

[0122] The feature calculation module is used to calculate the deviation value at each moment by analyzing the time series data before and after the winding deformation, which is used as the input feature for model training;

[0123] The label encoding module is used to process the winding fault location using one-hot encoding as the output label for model training;

[0124] The neural network training module is used to build a multi-layer perceptron neural network model and use preprocessed data and features for model training;

[0125] The model optimization module is used to calculate the error between the model output and the target through the loss function, and the Adam optimizer is used to update the model parameters;

[0126] The cross-validation module is used to cross-validate and evaluate the model during the training process to verify the performance and generalization ability of the model;

[0127] The model testing module is used to test the model using the test data set, calculate the model's evaluation indicators, and adjust the model structure and parameters;

[0128] The online diagnosis module is used to perform online diagnosis of transformer winding deformation through the trained model.

[0129] Example 3

[0130] An embodiment of the present invention is different from the first two embodiments in that:

[0131] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0132] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0133] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0134] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0135] Example 4

[0136] Reference Figure 4-Figure 10 , which is an embodiment of the present invention, provides an online diagnosis method for transformer winding deformation based on machine learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0137] Collect the basic parameters of the actual transformer and establish a 1:1 finite element simulation model as shown in the attached Figure 4As shown in the figure, according to the determined training model input, the data of normal operation is collected through the actual transformer, and the operation data of the transformer after deformation is simulated through the model, as shown in the attached figure. Figure 5 and attached Figure 6 The following are the voltage and current waveforms of the high-voltage and low-voltage sides of phase A. The abnormal data points are supplemented by S3, and the data are preprocessed by S4. Figure 7 The time series input of the current on the high voltage side of phase A before data processing includes the original time series data corresponding to seven winding states: deformation of the high voltage side and low voltage side windings of phase A, deformation of the high voltage side and low voltage side windings of phase B, deformation of the high voltage side and low voltage side windings of phase C, and no deformation. Figure 8 The time series data after data processing. Compared with before processing, the difference between the waveforms under different winding states is significantly amplified, which is more conducive to model training. Through forward propagation and loss function calculation (the iterative process of the loss function is shown in the attached figure) Fig. 9 As shown, it can converge quickly and complete the model training), back propagate gradients, and update the neural network model parameters. The trained neural network model is tested using the test data set. The final test determines that the accuracy of the winding deformation coil is 100%. The test results are shown in the attached Fig.10 As shown, 12 sets of predicted values ​​completely match the true values.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for online diagnosis of transformer winding deformation based on machine learning, characterized in that: include: Collect actual operating data of the transformer under normal operating conditions; A simulation model is established based on the actual structure and operating conditions of the transformer to simulate the winding deformation; Simulate the winding deformation condition in the simulation model and obtain the operating data under the deformation condition; Preprocessing the collected actual operation data and operation data under deformation conditions; By analyzing the time series data before and after the winding deformation, the deviation value at each moment is calculated as the input feature of the model training; The winding fault location is processed using one-hot encoding as the output label for model training; Build a multi-layer perceptron neural network model and use the preprocessed data and features to train the model; The error between the model output and the target is calculated through the loss function, and the model parameters are updated using the Adam optimizer; Cross-validate and evaluate the model during the training process to verify the performance and generalization ability of the model; Use the test data set to test the model, calculate the model's evaluation indicators, and adjust the model structure and parameters; The trained model is used to perform online diagnosis of transformer winding deformation.

2. The online diagnosis method for transformer winding deformation based on machine learning according to claim 1, characterized in that: The transformer operation data includes voltage, current, power and temperature; The simulation of the winding deformation includes using 3D modeling software to draw a 1:1 scale three-dimensional model of the transformer and windings of each voltage level; The obtaining of the operating data under deformation includes performing finite element analysis on the drawn 3D model using finite element analysis software, setting winding deformations at different positions using the finite element software, and simulating the operating conditions of the transformer under winding deformation.

3. The online diagnosis method for transformer winding deformation based on machine learning according to claim 2, characterized in that: The preprocessing is expressed as, Where x[t] is the input signal, y[t] is the filtered output signal, t is the position of the missing value, t1 and t2 are the positions of the adjacent known values, and x1 and x2 are the values ​​of the adjacent known values.

4. The online diagnosis method for transformer winding deformation based on machine learning according to claim 3 is characterized in that: The time series data before and after the winding deformation includes the original time series data {x1, x2, ..., x n } and the deformed time series data {y1,y2,...,y n }; The deviation value at each moment is expressed as, d t =y t -x t Among them, d t is the deviation value at time t, x t is the value of the original data at time t, y t is the value of the deformed data at time t; The one-hot encoding is represented as, Among them, i,j Represents the jth position of the fault label at the i-th position in the one-hot encoding.

5. The online diagnosis method for transformer winding deformation based on machine learning according to claim 4, characterized in that: The multi-layer perceptron neural network model includes an input layer, a hidden layer and an output layer; The hidden layer is represented as, h i =σ(∑ j w ij x j +b i ) Where hi is the output of the i-th node in the hidden layer, wij is the weight, xj is the input, bi is the bias, and σ(·) is the activation function; The output layer is represented as, y k =σ(∑ i w ki h i +b k ) Among them, yk is the output of the kth node in the output layer, hi is the output of the hidden layer, wki is the weight, and bk is the bias; The loss function is expressed as, Where n is the number of samples; y i is the true value of the i-th sample; is the predicted value of the ith sample.

6. The online diagnosis method for transformer winding deformation based on machine learning according to claim 5, characterized in that: Use the chain rule to calculate the gradient of the loss function with respect to each parameter, starting from the output layer, and propagating the gradient backward layer by layer to update the weights and biases of the neural network; Compute the gradient of the output layer’s loss with respect to its input, expressed as, Among them, δ (L) is the error term of the output layer; L is the loss function; z (L) is the linear combination of the output layer; σ' is the derivative of the activation function; a (l) is the activation value of layer l; W (l) is the weight matrix of the lth layer; b (l) is the bias vector of the lth layer; σ is the activation function; Where η is the learning rate.

7. The online diagnosis method for transformer winding deformation based on machine learning according to claim 6, characterized in that: The inputs of the training model are 18 inputs of three-phase voltage and current, 6 inputs of three-phase power, and 1 input of top oil temperature; The output of model training is 9 outputs of three-phase winding deformation positions; The input data set for model training uses time series as input, where the length of each time series of voltage and current is one cycle of the waveform, containing 50 points. Data processing is performed on the waveform of this cycle to capture the changes before and after the winding deformation. The input dataset contains 70 600-dimensional time series; the output dataset contains 70 corresponding fault locations, using one-hot encoding; 20% of the data set was selected as a test set and 12 samples; after the model training was completed, the 12 samples were used for diagnostic verification; The three-phase voltage and current 18 inputs include the voltage and current of the high, medium and low voltage side of the A phase of the transformer; the voltage and current of the high, medium and low voltage side of the B phase of the transformer; the voltage and current of the high, medium and low voltage side of the C phase of the transformer; The three-phase power six inputs include active and reactive power on the high, medium and low voltage sides of the transformer; The nine outputs of the three-phase winding deformation positions include the high, medium and low voltage sides of the transformer A phase; the high, medium and low voltage sides of the transformer B phase; and the high, medium and low voltage sides of the transformer C phase.

8. A system using the online transformer winding deformation diagnosis method based on machine learning as claimed in any one of claims 1 to 7, characterized in that: include: Data acquisition module, simulation model building module, deformation condition simulation module, data preprocessing module, feature calculation module, label encoding module, neural network training module, model optimization module, cross-validation module, model testing module, model adjustment module and online diagnosis module; The data acquisition module is used to collect actual operating data of the transformer under normal operating conditions; The simulation model building module is used to establish a simulation model according to the actual structure and operating conditions of the transformer to simulate the deformation of the winding; The deformation condition simulation module is used to simulate the winding deformation condition in the simulation model and obtain the operation data under the deformation condition; The data preprocessing module is used to preprocess the collected actual operation data and operation data under deformation; The feature calculation module is used to calculate the deviation value at each moment by analyzing the time series data before and after the winding deformation, which is used as the input feature of the model training; The label encoding module is used to process the winding fault position using one-hot encoding as the output label of the model training; The neural network training module is used to construct a multi-layer perceptron neural network model, and use the preprocessed data and features to perform model training; The model optimization module is used to calculate the error between the model output and the target through the loss function, and update the model parameters using the Adam optimizer; The cross-validation module is used to cross-validate and evaluate the model during the training process to verify the performance and generalization ability of the model; The model testing module is used to test the model using a test data set, calculate the evaluation index of the model, and adjust the model structure and parameters; The online diagnosis module is used to perform online diagnosis on transformer winding deformation through a trained model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the online diagnosis method for transformer winding deformation based on machine learning described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the online diagnosis method for transformer winding deformation based on machine learning described in any one of claims 1 to 7 are implemented.

Images