A method and device for early warning of short-circuit fault of rotor winding of steam turbine generator

Abstract

The present invention discloses a method and device for early warning of short-circuit fault between turns of rotor winding of steam turbine generator, which relates to the field of data processing, including: obtaining historical monitoring data and actual value of excitation current, preprocessing the historical monitoring data, obtaining preprocessed historical monitoring data and forming a monitoring sequence; constructing an excitation current prediction model based on Bi-LSTM, optimizing the training parameters to be optimized by using an improved Pelican optimization algorithm, obtaining optimized training parameters; training the excitation current prediction model based on the optimized training parameters, obtaining a trained excitation current prediction model; inputting the monitoring sequence into the trained excitation current prediction model, obtaining the predicted value of the excitation current, and judging whether the steam turbine generator has a short-circuit fault between turns of rotor winding according to the predicted value of the excitation current and the actual value of the excitation current. The present invention solves the problems of poor optimization ability, slow convergence speed and low prediction accuracy of the existing model.

Description

Technical Field

[0001] The invention relates to the field of data processing, and in particular to a method and a device for early warning of a short-circuit fault between turns of a steam turbine generator rotor winding. Background Art

[0002] Steam turbine generator is an important component of the power system architecture. Maintaining its normal and stable operation is crucial to the safety and stability of the power system. In the steam turbine generator, the rotor winding works in an extremely harsh environment. Especially when the large unit is running under load, the rotor winding needs to withstand strong electric fields, magnetic fields and mechanical forces. Therefore, in the steam turbine generator, the inter-turn short circuit of the rotor winding is a common fault phenomenon. This fault will lead to an increase in excitation current and a decrease in reactive power output. If the fault is not handled and controlled in time, it may further evolve into a more serious fault, such as damage to the rotor journal and bearing. In addition, the inter-turn short circuit of the rotor winding may also cause a series of problems such as increased vibration of the steam turbine generator, harmonics in the stator current, and changes in the temperature field of the motor. Although the inter-turn short circuit may not cause much harm to the motor itself in the early stage of the fault, if it is not handled in time, it will lead to further deterioration of the problem, thereby causing serious damage to the steam turbine generator. Therefore, the early warning and diagnosis of the inter-turn short circuit fault of the rotor winding of the steam turbine generator is particularly important, which plays a key role in maintaining the stable operation of the equipment.

[0003] It is difficult to accurately predict the excitation current of the unit using the traditional excitation current method. In recent years, data-driven power equipment fault prediction methods have made certain research progress. By collecting the distributed control system (DCS) data of the steam turbine generator and using the data-driven method to deeply mine its internal correlation information, the prediction accuracy of the excitation current can be effectively improved. In the early days, some scholars proposed an excitation current prediction method based on support vector regression (SVR), but SVR has the problem of poor generalization ability; later scholars trained the LSTM network with the data of the generator set under normal conditions to predict the excitation current. Studies have shown that LSTM can achieve better performance than other methods in time series prediction problems. However, the unidirectional LSTM network can only use the previous information, but does not consider the information of the hidden layer neurons later, and lacks the logic of the front and back. The dual attention mechanism is used to apply different weights to the parameters of LSTM and Bi-LSTM for optimization, and the optimal parameter model obtained by training is used to predict the current state of the steam turbine. It is found that Bi-LSTM has better results. In addition, there are studies that use animal population optimization to optimize the parameters of Bi-LSTM, such as using the locust optimization algorithm to optimize the motor bearing fault prediction method of Bi-LSTM. The established model has high prediction accuracy and strong robustness. However, the animal population optimization algorithm is prone to problems such as falling into local optimality and slow convergence in the later stage. Summary of the invention

[0004] The purpose of this application is to propose a method and device for early warning of short-circuit faults between turns of a steam turbine generator rotor winding in response to the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides a method for early warning of a short-circuit fault in a rotor winding of a steam turbine generator, comprising the following steps:

[0006] Acquire historical monitoring data collected by the turbine generator distributed control system and the actual value of the excitation current, the historical monitoring data including active power, reactive power, voltage, A-phase current, B-phase current and C-phase current, pre-process the historical monitoring data, obtain the pre-processed historical monitoring data and form a monitoring sequence;

[0007] Construct an excitation current prediction model based on Bi-LSTM, the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer, initialize the training parameters to be optimized in the excitation current prediction model, the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer, use the improved Pelican optimization algorithm to optimize the training parameters to obtain the optimized training parameters; train the excitation current prediction model based on the optimized training parameters to obtain the trained excitation current prediction model;

[0008] The monitoring sequence is input into the trained excitation current prediction model to obtain the predicted value of the excitation current. According to the predicted value of the excitation current and the actual value of the excitation current, it is judged whether the steam turbine generator has a rotor winding inter-turn short circuit fault.

[0009] As a preferred method, the steps of the improved Pelican optimization algorithm are as follows:

[0010] In the initialization phase, the parameters of the improved Pelican optimization algorithm are determined. The parameters of the improved Pelican optimization algorithm include the upper boundary, the lower boundary, the population size, the maximum number of iterations, and the dimension.

[0011] The initialization value of the training parameter to be optimized is used as the position of the pelican to initialize the pelican population, as shown in the following formula:

[0012] ;

[0013] in, Indicates the initial The first pelican Dimensional location, , represents the population size of pelicans, is a random number in the interval [0, 1], Represents the solution to the problem The lower boundary of the dimensional position, The first The upper boundary of the dimensional position, , Indicates the dimension of the problem to be solved;

[0014] The pelican population matrix is ​​represented as:

[0015] ;

[0016] in, is the initial pelican population matrix; For the initial The location of the pelican;

[0017] The objective function of the problem is used to calculate the objective function value of the pelican and form the objective function vector of the pelican population, as shown in the following formula:

[0018] ;

[0019] in, is the objective function vector of the initial pelican population; For the initial The objective function value for a pelican;

[0020] Exploration phase:

[0021] Prey is randomly generated, and the behavior of pelicans searching for food sources in a wide area is simulated to explore the search space and find the optimal area, as shown in the following formula:

[0022] ;

[0023] in, The exploration phase The first pelican Dimensional location, is a random number equal to 1 or 2, Indicates that the prey is The location in the dimension, Represents the updated first The objective function value of the position of the pelican;

[0024] Update the position of each pelican according to the pelican foraging behavior rules, as shown in the following formula:

[0025] ;

[0026] in, Represents the updated first The location of the pelican, represents the perturbation factor, tanh is the activation function, represents the standard normal distribution function;

[0027] Development phase:

[0028] By simulating a more precise search of the pelican near a known food source, the best location is found within the known optimal area, as shown in the following formula:

[0029] ;

[0030] in, For the development stage The first pelican Dimensional location, is a constant, for The neighborhood radius of is the current iteration number, is the maximum number of iterations; Represents the adaptive weight, which is calculated using the following formula:

[0031] ;

[0032] in, and are the maximum and minimum values ​​of the weight respectively;

[0033] Update the position of each pelican as follows:

[0034] ;

[0035] in, For the development stage The location of the pelican, Based on the development phase update The objective function value of the position of the pelican;

[0036] The position of the pelican during the development phase is the value of the optimized training parameters.

[0037] Preferably, during the training of the excitation current prediction model, historical monitoring data when there is no inter-turn short-circuit fault in the rotor winding of the steam turbine generator and the actual sequence of the excitation current are collected, the VMD algorithm is used to decompose the actual sequence of the excitation current into k actual values, and training data is established for each actual value in combination with the corresponding monitoring sequence obtained by preprocessing the historical monitoring data. The number of neurons in the first hidden layer and the number of neurons in the second hidden layer in the optimized training parameters of the excitation current prediction model are fixed, and the training data are trained using the number of training times and learning rate in the optimized training parameters. The predicted values ​​corresponding to each actual value are superimposed and reconstructed to obtain a prediction result, and a loss function is calculated based on each actual value and its corresponding predicted value, and the optimal parameters for each iteration are determined until the iteration ends to obtain a trained excitation current prediction model.

[0038] Preferably, the loss function comprises a mean square error loss function.

[0039] Preferably, the preprocessing includes normalization processing.

[0040] Preferably, judging whether there is a rotor winding turn-to-turn short circuit fault in the steam turbine generator according to the predicted value of the excitation current and the actual value of the excitation current specifically includes:

[0041] The relative deviation of the mean value in each window between the predicted value and the actual value of the excitation current using the sliding window method The analysis is as shown below:

[0042] ;

[0043] in, is the mean value of the actual value of the excitation current in each window, is the mean of the predicted value of the excitation current in each window;

[0044] In response to determining the relative deviation If it is greater than the threshold, it is determined that there is a short-circuit fault between the rotor winding turns of the steam turbine generator;

[0045] In response to determining the relative deviation If the value is less than or equal to the threshold, it is determined that there is no rotor winding turn-to-turn short circuit fault in the steam turbine generator.

[0046] In a second aspect, the present invention provides a turbine generator rotor winding inter-turn short-circuit fault early warning device, comprising:

[0047] The data processing module is configured to obtain historical monitoring data collected by the steam turbine generator distributed control system and the actual value of the excitation current, the historical monitoring data including active power, reactive power, voltage, A-phase current, B-phase current and C-phase current, pre-process the historical monitoring data, obtain the pre-processed historical monitoring data and form a monitoring sequence;

[0048] The model building module is configured to build an excitation current prediction model based on Bi-LSTM, the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer, initialize the training parameters to be optimized in the excitation current prediction model, the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer, and use the improved Pelican optimization algorithm to optimize the training parameters to obtain the optimized training parameters; train the excitation current prediction model based on the optimized training parameters to obtain the trained excitation current prediction model;

[0049] The prediction module is configured to input the monitoring sequence into a trained excitation current prediction model to obtain a predicted value of the excitation current, and determine whether the steam turbine generator has a rotor winding inter-turn short circuit fault based on the predicted value of the excitation current and the actual value of the excitation current.

[0050] In a third aspect, the present invention provides an electronic device comprising one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0052] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) The turbine generator rotor winding inter-turn short-circuit fault warning method proposed in the present invention introduces adaptive weights into the Pelican position update formula in the development stage of the Pelican optimization algorithm: in the early stage of the algorithm, a higher exploration weight is set to encourage the algorithm to explore a wider search space, thereby avoiding local optimality. As the algorithm iterates, the exploration weight is gradually reduced while the utilization weight is increased. This linear adjustment helps the Pelican optimization algorithm to focus more on detailed searches in the neighborhood of the current best solution in the later stages, which can greatly improve the optimization efficiency; at the same time, the introduction of disturbance factors in the exploration stage and the development stage can not only promote global search, but also effectively guide local search, avoiding the algorithm from falling into the local optimal solution too early, while not being too dispersed in the global search, affecting the search efficiency.

[0055] (2) The turbine generator rotor winding inter-turn short-circuit fault early warning method proposed in the present invention adopts the improved Pelican optimization algorithm to optimize the training times, learning rate, number of neurons in the first hidden layer and number of neurons in the second hidden layer of the Bi-LSTM model, and then trains to obtain the trained excitation current prediction model, which not only effectively improves the prediction accuracy, but also has stronger optimization ability and convergence speed, making up for the defects of SSA and POA algorithms that are easy to fall into the global optimum and have slow convergence speed.

[0056] (3) The turbine generator rotor winding inter-turn short-circuit fault early warning method proposed in the present invention uses the VMD algorithm to decompose the actual sequence of the excitation current during the training process, which can not only reduce noise, but also better fit the model and obtain higher prediction accuracy. The present invention can provide pre-fault prediction in time before the turbine generator rotor winding inter-turn short-circuit fault occurs, realize fault analysis and control in the first time, and prevent huge economic losses and safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A schematic flow chart of a method for early warning of a short-circuit fault between turns of a turbine generator rotor winding according to an embodiment of the present application;

[0059] Figure 2 A flowchart of the IPOA algorithm for optimizing the Bi-LSTM model of the turbine generator rotor winding inter-turn short-circuit fault early warning method according to an embodiment of the present application;

[0060] Figure 3 A schematic diagram of the training process of the IPOA-Bi-LSTM model of the turbine generator rotor winding inter-turn short-circuit fault early warning method according to an embodiment of the present application;

[0061] Figure 4 A flowchart of a method for early warning of a short-circuit fault between turns of a turbine generator rotor winding according to an embodiment of the present application;

[0062] Figure 5 The VMD algorithm result of the turbine generator rotor winding inter-turn short-circuit fault early warning method according to the embodiment of the present application;

[0063] Figure 6 A loss function value curve diagram of different optimization algorithms of the turbine generator rotor winding inter-turn short-circuit fault early warning method according to an embodiment of the present application;

[0064] Figure 7 It is a prediction result diagram of different types of prediction models of the turbine generator rotor winding inter-turn short-circuit fault early warning method according to the embodiment of the present application;

[0065] Figure 8 A comparison diagram of actual values ​​and predicted values ​​of a method for early warning of a short-circuit fault between turns of a steam turbine generator rotor winding in a normal state according to an embodiment of the present application;

[0066] Fig. 9 A comparison diagram of actual values ​​and predicted values ​​during an inter-turn short-circuit fault in a method for early warning of an inter-turn short-circuit fault in a turbine generator rotor winding according to an embodiment of the present application;

[0067] Fig.10 A deviation comparison diagram of a normal state and a turn-to-turn short-circuit fault of a turbine generator rotor winding early warning method according to an embodiment of the present application;

[0068] Fig.11A schematic diagram of a turbine generator rotor winding inter-turn short-circuit fault early warning device according to an embodiment of the present application;

[0069] Fig.12 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] Figure 1 A method for early warning of a short-circuit fault between turns of a steam turbine generator rotor winding provided by an embodiment of the present application is shown, comprising the following steps:

[0072] S1, obtain the historical monitoring data collected by the steam turbine generator distributed control system and the actual value of the excitation current, the historical monitoring data includes active power, reactive power, voltage, A phase current, B phase current and C phase current, pre-process the historical monitoring data, obtain the pre-processed historical monitoring data and form a monitoring sequence.

[0073] In a specific embodiment, the preprocessing includes a normalization process.

[0074] Specifically, the historical monitoring data required to predict the excitation current is determined based on the fitting relationship between the no-load electromotive force and the excitation current. When the rotor of the steam turbine generator has an inter-turn short circuit, the effective number of turns of the rotor winding decreases. In order to maintain the same air gap magnetic flux, the excitation system needs to provide a larger excitation current; in an ideal situation, that is, when the magnetic circuit of the steam turbine generator is not saturated, there is a linear relationship between the no-load electromotive force and the excitation current. This means that an increase in the excitation current will cause a proportional increase in the no-load electromotive force, and vice versa. When the excitation current increases to a certain extent, causing the magnetic circuit of the generator to enter a saturated state, this linear relationship will no longer hold. In practical applications, the magnetic circuit of the generator may exhibit nonlinear characteristics due to various factors (such as temperature, frequency, and the magnetic properties of the core material). Therefore, the actual relationship between the no-load electromotive force and the excitation current may be more complicated than the theoretical linear or nonlinear. The no-load electromotive force of a running generator is defined as:

[0075] ;

[0076] in, is the no-load electromotive force, is the power angle, is the voltage, is the current, is the DC resistance of the generator, is the synchronous reactance of the generator, and It can be directly obtained from the turbine generator distributed control system (DCS), Active power that can be read from DCS and reactive power The calculation formula is:

[0077] .

[0078] When the insulation of the generator interturn winding is normal, the air gap electromotive force and air gap flux remain basically unchanged. At this time, the generator magnetic saturation curve is also fixed, and the generator is at the inflection point of the magnetization curve, so the magnetic permeability The no-load electromotive force at this time is There is a linear correlation between the excitation current and the current, which can be fitted to the normal state of the unit. A linear function of an array is defined as:

[0079] ;

[0080] in, It is the excitation current of the unit under normal condition.

[0081] In summary, the embodiment of the present application converts the active power in the data collected by the DCS into , reactive power ,Voltage , A phase current , B phase current and C phase current These six parameters serve as input variables of the excitation current prediction model.

[0082] refer to Figure 2 , the above six historical monitoring data of the steam turbine generator are collected, and the collected historical monitoring data are preprocessed. The preprocessing process adopts normalization processing, and the preprocessed historical monitoring data constitute a monitoring sequence.

[0083] S2, construct an excitation current prediction model based on Bi-LSTM, the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer, initialize the training parameters to be optimized in the excitation current prediction model, the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer, use the improved Pelican optimization algorithm to optimize the training parameters to obtain the optimized training parameters; train the excitation current prediction model based on the optimized training parameters to obtain the trained excitation current prediction model.

[0084] In a specific embodiment, the steps of the improved Pelican optimization algorithm are as follows:

[0085] In the initialization phase, the parameters of the improved Pelican optimization algorithm are determined. The parameters of the improved Pelican optimization algorithm include the upper boundary, the lower boundary, the population size, the number of iterations, and the dimension.

[0086] The initialization value of the training parameter to be optimized is used as the position of the pelican to initialize the pelican population, as shown in the following formula:

[0087] ;

[0088] in, Indicates the initial The first pelican Dimensional location, , represents the population size of pelicans, is a random number in the interval [0, 1], Represents the solution to the problem The lower boundary of the dimensional position, The first The upper boundary of the dimensional position, , Indicates the dimension of the problem to be solved;

[0089] The pelican population matrix is ​​represented as:

[0090] ;

[0091] in, is the initial pelican population matrix; For the initial The location of the pelican;

[0092] The objective function of the problem is used to calculate the objective function value of the pelican and form the objective function vector of the pelican population, as shown in the following formula:

[0093] ;

[0094] in, is the objective function vector of the initial pelican population; For the initial The objective function value for a pelican;

[0095] Exploration phase:

[0096] Prey is randomly generated, and the behavior of pelicans searching for food sources in a wide area is simulated to explore the search space and find the optimal area, as shown in the following formula:

[0097] ;

[0098] in, The exploration phase The first pelican Dimensional location, is a random number equal to 1 or 2, Indicates that the prey is The location in the dimension, Represents the updated first The objective function value of the position of the pelican;

[0099] Update the position of each pelican according to the pelican foraging behavior rules, as shown in the following formula:

[0100] ;

[0101] in, Represents the updated first The location of the pelican, represents the perturbation factor, tanh is the activation function, represents the standard normal distribution function;

[0102] Development phase:

[0103] By simulating a more precise search of the pelican near a known food source, the best location is found within the known optimal area, as shown in the following formula:

[0104] ;

[0105] in, For the development stage The first pelican Dimensional location, is a constant, for The neighborhood radius of is the current iteration number, is the maximum number of iterations; Represents the adaptive weight, which is calculated using the following formula:

[0106] ;

[0107] in, and are the maximum and minimum values ​​of the weight respectively;

[0108] Update the position of each pelican as follows:

[0109] ;

[0110] in, For the development stage The location of the pelican, Based on the development phase update The objective function value of the position of the pelican;

[0111] The position of the pelican during the development phase is the value of the optimized training parameters.

[0112] In a specific embodiment, during the training process of the excitation current prediction model, historical monitoring data when there is no rotor winding inter-turn short-circuit fault in the steam turbine generator and the actual sequence of the excitation current are collected, and the VMD algorithm is used to decompose the actual sequence of the excitation current into k actual values. Training data is established for each actual value in combination with the monitoring sequence obtained by preprocessing the corresponding historical monitoring data, and the number of neurons in the first hidden layer and the number of neurons in the second hidden layer in the optimized training parameters of the excitation current prediction model are fixed. The training data is trained using the number of training times and learning rate in the optimized training parameters, and the predicted values ​​corresponding to each actual value are superimposed and reconstructed to obtain a prediction result. The loss function is calculated based on each actual value and its corresponding predicted value, and the optimal parameters for each iteration are determined until the iteration ends to obtain a trained excitation current prediction model.

[0113] In a specific embodiment, the loss function includes a mean square error loss function.

[0114] Specifically, in the improved pelican optimization algorithm (IPOA), the number of training times, the learning rate, the number of neurons in the first hidden layer, and the number of neurons in the second hidden layer are selected as the training parameters to be optimized. First, the IPOA parameters and the training parameters to be optimized are initialized, and prey is randomly generated. The training parameters to be optimized are respectively regarded as the pelicans in IPOA, and the values ​​of the training parameters to be optimized are the positions of the pelicans. The optimization ranges of the training parameters to be optimized are determined as follows: , , the objective function of the problem being solved can be used to calculate the objective function value of Pelican.

[0115] After initialization, the pelican enters the exploration phase, in which the pelican is simulated to search for food sources in a wide area. At this point, the improved pelican optimization algorithm focuses on exploring the search space in order to find potential optimal solutions. The improved pelican optimization algorithm conducts a wide search in the solution space to ensure that no potential good areas are missed. The first pelican The parameter I in the calculation formula for the dimensional position is randomly selected for each iteration and each pelican. When I=2, it will bring more displacement to the pelican, which may guide the pelican to a new area of ​​the search space. Therefore, the value of parameter I will affect the exploration ability of the IPOA algorithm. Update the position of the pelican according to the rules of the pelican's foraging behavior. This usually involves considering the pelican's current position, the best historical position, and information about other pelicans in the population. Make sure that the updated position is still within the defined search space. If the position of a pelican is beyond the boundary of the search space, it needs to be placed back within the boundary to prevent the algorithm from moving to a non-optimal area.

[0116] After the exploration phase is completed, the pelican enters the development phase, which simulates the pelican's more precise search near known food sources. In this phase, the IPOA algorithm focuses more on refining and improving the potential solutions that have been discovered. The IPOA algorithm conducts a more concentrated and detailed search around the good solutions that have been discovered, focusing on finding the best solution in the known good area. As the algorithm proceeds, the random component in the search is reduced to more accurately approximate the optimal solution. If the objective function value of a pelican in the new position is better than its historical best objective function value, its historical best position is updated. Introducing adaptive weights in the pelican position update formula in the exploration and development phases can be adjusted according to the current number of iterations, the individual's historical performance, or the overall performance of the group, thereby helping to guide the search process. At the same time, introducing perturbation factors in the development phase can enhance the exploration ability of the IPOA algorithm, avoid local optimal solutions, and improve robustness and generalization capabilities. Through comparative analysis, it can be found that this strategy can usually significantly improve the effect of the original POA algorithm. Introducing perturbation factors into the development phase can increase exploration capabilities, avoid premature convergence to local optimal solutions, and enhance global search capabilities.

[0117] The trained excitation current prediction model is obtained by training based on the optimized training parameters. In each iterative training process, the optimal training parameters under the current iteration number are generated until the iteration ends. In the embodiment of the present application, the mean square error loss function is used as the loss function of the excitation current prediction model, and the parameter with the smallest mean square error (MSE) is selected as the global optimal parameter. Figure 3, collect the historical monitoring data and the actual sequence of the excitation current when there is no rotor winding inter-turn short-circuit fault in the steam turbine generator, and use VMD to decompose the actual sequence of the excitation current into k actual values, recorded as . Training data is established for each actual value combined with the corresponding monitoring sequence to train and test the excitation current prediction model (IPOA-Bi-LSTM) optimized by the improved Pelican optimization algorithm, thereby obtaining a trained excitation current prediction model. Each predicted value is superimposed and reconstructed to obtain the final prediction result.

[0118] S3, inputting the monitoring sequence into the trained excitation current prediction model to obtain the predicted value of the excitation current, and judging whether the steam turbine generator has a rotor winding inter-turn short circuit fault according to the predicted value of the excitation current and the actual value of the excitation current.

[0119] In a specific embodiment, judging whether there is a rotor winding turn-to-turn short circuit fault in the steam turbine generator according to the predicted value of the excitation current and the actual value of the excitation current specifically includes:

[0120] The relative deviation of the mean value in each window between the predicted value and the actual value of the excitation current using the sliding window method The analysis is as shown below:

[0121] ;

[0122] in, is the mean value of the actual value of the excitation current in each window, is the mean of the predicted value of the excitation current in each window;

[0123] In response to determining the relative deviation If it is greater than the threshold, it is determined that there is a short-circuit fault between the rotor winding turns of the steam turbine generator;

[0124] In response to determining the relative deviation If the value is less than or equal to the threshold, it is determined that there is no rotor winding turn-to-turn short circuit fault in the steam turbine generator.

[0125] Specifically, refer to Figure 4 The trained excitation current prediction model is deployed and applied, and the monitoring sequence at each time point is input into the trained excitation current prediction model to obtain the predicted value of the excitation current. The relative deviation of the mean value in each window between the predicted value and the actual value of the generator excitation current is calculated by the sliding window method. According to the analysis, when the generator has a rotor winding interturn insulation defect, the actual value of the excitation current will be significantly higher than the predicted value, that is, the relative deviation If the value is greater than the threshold, an early warning of the short-circuit fault between turns of the turbine generator rotor winding can be issued.

[0126] The technical solutions of the embodiments of the present application are further illustrated by means of specific examples below.

[0127] The present invention uses 21 normal data from July 1, 2021 to July 21, 2021 collected in real time by the DCS of a large steam turbine generator in a power plant for training, totaling 30,240 groups, and the programming environment is Matlab2021b.

[0128] Considering that the excitation current of the steam turbine generator is a non-stationary and fluctuating time series, VMD is used to decompose the actual sequence data of the excitation current in the training set into different modes, among which The determination of the value has a great influence on the decomposition result. The embodiment of the present application uses the center frequency method to determine Value. Let the penalty factor of VMD be , Convergence criterion tolerance , obtained through multiple groups of experiments The relationship with the center frequency is shown in Table 1.

[0129] Table 1 Different The center frequency of the mode under the value:

[0130]

[0131] when When the value is constant, if the increase in the center frequency between the modes is not large, the two modes are considered similar. Figure 5 It can be obtained when When the center frequencies of mode 5 and mode 6 increase by 8.83% from 0.2179 to 0.2378, the center frequencies are similar at this time, so the embodiment of the present application selects The actual sequence data of the excitation current is decomposed, and the decomposed data is as follows: Figure 5 shown.

[0132] In order to fully demonstrate the superiority of the improved Pelican optimization algorithm, the IPOA algorithm, POA algorithm and sparrow search algorithm (SSA) are used to optimize the nodes of the first hidden layer, the nodes of the second hidden layer, the number of training times and the learning rate of the Bi-LSTM model respectively. The optimization ranges are , , , and simulate the test set. The IPOA algorithm parameters are set to the maximum number of iterations 50, the population size 30, and the other algorithm parameters are set the same as the POA algorithm. The change curves of the loss function values ​​of the three algorithms are shown in Figure 6 shown.

[0133] Depend on Figure 6 It can be seen that the IPOA algorithm converges faster than the SSA algorithm and the POA algorithm, and the number of training times used to obtain the optimal loss function value is the least, which verifies that the IPOA algorithm has stronger optimization ability and convergence speed.

[0134] In order to verify the advantages of the trained excitation current prediction model of the embodiment of the present application in the excitation current prediction, the trained excitation current prediction model of the embodiment of the present application is compared with a single LSTM prediction model, a single Bi-LSTM prediction model, a VMD-Bi-LSTM combination prediction model, a POA-Bi-LSTM combination prediction model, an SSA-Bi-LSTM combination prediction model, an IPOA-Bi-LSTM combination prediction model, a VMD-SSA-Bi-LSTM combination prediction model, and a VMD-POA-Bi-LSTM combination prediction model. All models divide the first 60% of the data into training sets and the last 40% of the data into test sets. Some prediction results of each model are shown in Figure 2. Figure 7 shown.

[0135] In order to verify the accuracy of the trained excitation current prediction model of the embodiment of the present application in predicting the excitation current, the embodiment of the present application selects the root mean square error (RMSE) and mean absolute percentage error (MAPE) indicators to evaluate the prediction accuracy of the model. These two indicators are widely used in the performance evaluation of various prediction models because they can effectively quantify the difference between the model prediction value and the actual value. The specific calculation formula is as follows:

[0136] ;

[0137] ;

[0138] in, is the number of prediction results; Indicates actual value; Represents the predicted value.

[0139] Table 2 Comparison of errors of different prediction models:

[0140]

[0141] pass Figure 7It can be found that the trained excitation current prediction model of the embodiment of the present application has the best excitation current prediction effect compared with other models. From the local magnification diagram and the relative error diagram, the prediction curve is closest to the real load curve, with the highest degree of fitting and the smallest error, which verifies the superiority of the trained excitation current prediction model of the embodiment of the present application. From Table 2, it is analyzed that the Bi-LSTM model overcomes the defect that the traditional LSTM model cannot perform bidirectional propagation training, and can better explore the characteristics of the excitation current sequence. The parameters of the Bi-LSTM model are optimized by using the improved Pelican optimization algorithm IPOA with disturbance factors and adaptive weights, which makes up for the defects of the SSA algorithm and POA algorithm that are easy to fall into the global optimum and slow convergence speed, and improves the prediction accuracy. Compared with the VMD-Bi-LSTM model that only predicts after decomposing the load sequence and the SSA-Bi-LSTM, POA-Bi-LSTM and IPOA-Bi-LSTM two-by-two combination models that only predict after optimizing the model parameters, decomposing the load sequence and then inputting it into the optimized model can further improve the prediction accuracy. The RMSE and MAPE of the trained excitation current prediction model of the embodiment of the present application are 10.0359 and 0.3438% respectively, and each evaluation index is significantly better than other models, showing a stronger learning ability. The superiority of the trained excitation current prediction model of the embodiment of the present application in excitation current prediction is verified.

[0142] The sliding window width is selected as 10 and the step length is 1. The fault data collected in real time by the DCS of the generator from January 1 to January 31, 2022 is input into the trained excitation current prediction model of the embodiment of the present application for prediction. The excitation current in each window of the measured value and the predicted value is averaged, such as Figure 8 and Fig. 9 By analyzing the variation trend of the actual value and predicted value of the excitation current in the normal state and short-circuit fault state of the rotor winding interturn, and calculating the deviation between the two, it can be found that there is a significant difference between the deviation of the actual value and predicted value of the excitation current when the interturn insulation of the generator is in different states. Some results are shown in Table 3

[0143] Table 3 Actual and predicted values ​​of partial excitation current under different conditions :

[0144]

[0145] Fig.10The deviation value of the excitation current under the fault state and the normal state of the generator can be found by comparison that when the inter-turn insulation of the generator rotor winding is in a normal state, the deviation between the predicted value and the actual value of the excitation current is in the range of 0~1%, and when the inter-turn insulation of the rotor winding has a short circuit fault, the deviation between the predicted value and the actual value of the excitation current will rise sharply, generally higher than 5%, and the measured values ​​are higher than the predicted values. Therefore, the relative deviation between the actual value and the predicted value of the generator excitation current can be calculated to determine whether the generator has an inter-turn short circuit fault, and the fault characteristics are more obvious, the diagnostic accuracy is higher, and the accurate diagnosis of the inter-turn short circuit fault of the rotor winding can be achieved.

[0146] Further references Fig.11 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a turbine generator rotor winding inter-turn short-circuit fault warning device. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0147] The embodiment of the present application provides a turbine generator rotor winding inter-turn short circuit fault early warning device, comprising:

[0148] The data processing module 1 is configured to obtain historical monitoring data collected by the steam turbine generator distributed control system and the actual value of the excitation current, the historical monitoring data including active power, reactive power, voltage, A-phase current, B-phase current and C-phase current, pre-process the historical monitoring data, obtain the pre-processed historical monitoring data and form a monitoring sequence;

[0149] The model building module 2 is configured to build an excitation current prediction model based on Bi-LSTM, the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer, initialize the training parameters to be optimized in the excitation current prediction model, the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer, and use the improved Pelican optimization algorithm to optimize the training parameters to obtain the optimized training parameters; train the excitation current prediction model based on the optimized training parameters to obtain the trained excitation current prediction model;

[0150] The prediction module 3 is configured to input the monitoring sequence into the trained excitation current prediction model to obtain the predicted value of the excitation current, and determine whether the steam turbine generator has a rotor winding inter-turn short circuit fault based on the predicted value of the excitation current and the actual value of the excitation current.

[0151] Fig.12 Schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present invention. Fig.12As shown, the electronic device of this embodiment includes: a processor 1201 and a memory 1202; wherein the memory 1202 is used to store computer-executable instructions; the processor 1201 is used to execute the computer-executable instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant description in the above method embodiment.

[0152] Optionally, the memory 1202 may be independent or integrated with the processor 1201 .

[0153] When the memory 1202 is independently provided, the electronic device further includes a bus 1203 for connecting the memory 1202 and the processor 1201 .

[0154] The embodiment of the present invention further provides a computer storage medium, in which computer execution instructions are stored. When the processor 1201 executes the computer execution instructions, the above method is implemented.

[0155] The embodiment of the present invention further provides a computer program product, including a computer program. When the computer program is executed by the processor 1201, the above method is implemented.

[0156] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0157] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0158] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The unit formed by the above modules may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0159] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor 1201 to perform some steps of the methods of various embodiments of the present application.

[0160] It should be understood that the processor 1201 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or the processor 1201 may be any conventional processor 1201, etc. The steps of the method disclosed in the invention may be directly embodied in the hardware processor 1201 for execution, or may be executed by a combination of hardware and software modules in the processor 1201.

[0161] The memory 1202 may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

[0162] The bus 1203 may be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus 1203 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 1203 in the drawings of the present application is not limited to only one bus 1203 or one type of bus 1203.

[0163] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0164] An exemplary storage medium is coupled to the processor 1201, so that the processor 1201 can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor 1201. The processor 1201 and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor 1201 and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0165] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0166] Finally, 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of short-circuit faults between turns of a turbine generator rotor winding, characterized in that: The following steps are involved: Acquire historical monitoring data collected by the steam turbine generator distributed control system and the actual value of the excitation current, wherein the historical monitoring data includes active power, reactive power, voltage, A-phase current, B-phase current and C-phase current, pre-process the historical monitoring data, obtain the pre-processed historical monitoring data and form a monitoring sequence; An excitation current prediction model based on Bi-LSTM is constructed, wherein the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer. The training parameters to be optimized in the excitation current prediction model are initialized, wherein the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer. The training parameters to be optimized are optimized using an improved Pelican optimization algorithm to obtain optimized training parameters. The excitation current prediction model is trained based on the optimized training parameters to obtain a trained excitation current prediction model. During the training process of the excitation current prediction model, the information that there is no inter-turn short circuit in the rotor winding of the steam turbine generator is collected. The historical monitoring data at the time of the fault and the actual sequence of the excitation current are used to decompose the actual sequence of the excitation current into k actual values ​​by using the VMD algorithm, and the monitoring sequence obtained by preprocessing the corresponding historical monitoring data is used to establish training data, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer in the optimized training parameters of the excitation current prediction model are fixed, and the training data is trained using the number of training times and the learning rate in the optimized training parameters, and the prediction values ​​corresponding to each actual value are superimposed and reconstructed to obtain the prediction result, and the loss function is calculated based on each actual value and its corresponding prediction value, and the optimal parameters of each iteration are determined until the iteration ends, so as to obtain the trained excitation current prediction model; The monitoring sequence is input into the trained excitation current prediction model to obtain a predicted value of the excitation current, and it is determined whether the steam turbine generator has a rotor winding turn-to-turn short circuit fault based on the predicted value of the excitation current and the actual value of the excitation current.

2. The method for early warning of short-circuit fault of a turbine generator rotor winding according to claim 1, characterized in that: The steps of the improved Pelican optimization algorithm are as follows: In the initialization stage, the parameters of the improved Pelican optimization algorithm are determined, and the parameters of the improved Pelican optimization algorithm include an upper boundary, a lower boundary, a population size, a maximum number of iterations, and a dimension; The initialization value of the training parameter to be optimized is used as the position of the pelican to initialize the pelican population, as shown in the following formula: x i,j =l j +rand×(and j -l j ); Among them, x i,j represents the j-th dimensional position of the initial i-th pelican, i = 1, 2, ..., N, N represents the population of pelicans, rand is a random number in the interval [0, 1], l j Indicates the lower boundary of the j-th dimension of the problem, u j Represents the upper boundary of the j-th dimension position of the problem to be solved, j = 1, 2, ..., m, m represents the dimension of the problem to be solved; The pelican population matrix is ​​represented as: Among them, X is the initial pelican population matrix; X i is the initial position of the i-th pelican; The objective function of the problem is used to calculate the objective function value of the pelican and form the objective function vector of the pelican population, as shown in the following formula: Where F is the objective function vector of the initial pelican population; F i is the initial objective function value of the i-th pelican; Exploration phase: Prey is randomly generated, and the behavior of pelicans searching for food sources in a wide area is simulated to explore the search space and find the optimal area, as shown in the following formula: in, represents the j-th position of the ith pelican in the exploration phase, I is a random number equal to 1 or 2, and p j represents the position of the prey in the jth dimension, represents the objective function value based on the updated position of the i-th pelican in the exploration phase; Update the position of each pelican according to the pelican foraging behavior rules, as shown in the following formula: in, represents the updated position of the i-th pelican based on the exploration phase, (1+tanh(randn)) represents the disturbance factor, tanh is the activation function, and randn represents the standard normal distribution function; Development phase: By simulating a more precise search of the pelican near a known food source, the best location is found within the known optimal area, as shown in the following formula: in, is the j-th dimensional position of the ith pelican in the development phase, R is a constant, and R·(1-t / T) is x i,j The neighborhood radius, t is the current iteration number, T is the maximum iteration number; ω represents the adaptive weight, which is calculated using the following formula: Among them, ω max and ω min are the maximum and minimum values ​​of the weight respectively; Update the position of each pelican as follows: in, is the position of the i-th pelican in the development phase, is the objective function value based on the position of the i-th pelican after the development phase update; The position of the pelican in the development stage is the value of the optimized training parameter.

3. The method for early warning of short-circuit fault of a turbine generator rotor winding according to claim 1, characterized in that: The loss function includes a mean square error loss function.

4. The method for early warning of short-circuit fault of rotor winding of a steam turbine generator according to claim 1, characterized in that: The preprocessing includes normalization processing.

5. The method for early warning of inter-turn short-circuit fault of a turbine generator rotor winding according to claim 1, characterized in that: Judging whether the steam turbine generator has a rotor winding turn-to-turn short circuit fault according to the predicted value of the excitation current and the actual value of the excitation current, specifically includes: The sliding window method is used to analyze the relative deviation a of the mean value in each window between the predicted value and the actual value of the excitation current, as shown in the following formula: Among them, I f is the mean value of the actual value of the excitation current in each window, I f′ is the mean of the predicted value of the excitation current in each window; In response to determining that the relative deviation a is greater than a threshold, determining that the steam turbine generator has a rotor winding turn-to-turn short circuit fault; In response to determining that the relative deviation a is less than or equal to a threshold, it is determined that no rotor winding inter-turn short circuit fault exists in the steam turbine generator.

6. A turbine generator rotor winding inter-turn short-circuit fault early warning device, characterized in that: include: A data processing module is configured to obtain historical monitoring data collected by the steam turbine generator distributed control system and an actual value of the excitation current, wherein the historical monitoring data includes active power, reactive power, voltage, A-phase current, B-phase current and C-phase current, and pre-process the historical monitoring data to obtain the pre-processed historical monitoring data and form a monitoring sequence; The model building module is configured to build an excitation current prediction model based on Bi-LSTM, wherein the excitation current prediction model includes an input layer, a first hidden layer, a second hidden layer and an output layer, and initialize the training parameters to be optimized in the excitation current prediction model, wherein the training parameters to be optimized include the number of training times, the learning rate, the number of neurons in the first hidden layer and the number of neurons in the second hidden layer, and use an improved Pelican optimization algorithm to optimize the training parameters to obtain optimized training parameters; based on the optimized training parameters, the excitation current prediction model is trained to obtain a trained excitation current prediction model, and during the training process of the excitation current prediction model, the information that the steam turbine generator does not have a rotor winding is collected. The historical monitoring data of the inter-turn short-circuit fault and the actual sequence of the excitation current are used, and the actual sequence of the excitation current is decomposed into k actual values ​​by using the VMD algorithm, and the monitoring sequence obtained by preprocessing each actual value in combination with the corresponding historical monitoring data is used to establish training data, and the number of neurons in the first hidden layer and the number of neurons in the second hidden layer in the optimized training parameters of the excitation current prediction model are fixed, and the training data is trained using the number of training times and the learning rate in the optimized training parameters, and the prediction values ​​corresponding to each actual value are superimposed and reconstructed to obtain the prediction result, and the loss function is calculated based on each actual value and its corresponding prediction value, and the optimal parameters of each iteration are determined until the iteration ends, and the trained excitation current prediction model is obtained; The prediction module is configured to input the monitoring sequence into the trained excitation current prediction model to obtain a predicted value of the excitation current, and determine whether the steam turbine generator has a rotor winding inter-turn short circuit fault based on the predicted value of the excitation current and the actual value of the excitation current.

7. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images