Winding insulation residual life prediction method and system

Abstract

The invention provides a winding insulation residual life prediction method and system, and relates to the technical field of winding insulation residual life prediction. According to the method, firstly, a data set is divided through segmentation threshold values, and on the basis of an Arrhenius equation, a segmentation mathematical model is constructed to describe the degradation process of an insulating material; then, a boundary penalty function of mathematical model parameters is introduced, a neural network model is constructed, the mathematical model and a neural network are jointly optimized, and meanwhile, a weighted loss function is designed in combination with statistics of degradation data; and finally, performing progressive training on the neural network by using each sub-data set, and generating a prediction result through superposition of a theoretical aging component of the mechanism model and a residual compensation component learned by the neural network. According to the method, segmented dynamic modeling and neural network learning are combined to realize accurate prediction of the non-linear time-varying trajectory of the insulation state characteristic quantity, and the characterization capability of the model on the complex degradation process of the insulation material is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the remaining life of winding insulation, and particularly to a method and system for predicting the remaining life of winding insulation jointly driven by the degradation mechanism of insulating materials and data. Background Art

[0002] In modern industries and power systems, as the core component of many electrical equipment, windings are widely used in the manufacturing fields of transformers, motors, generators, etc., and are key components for realizing the conversion and transmission of electrical energy. A winding is usually precisely wound by a highly conductive material. Its structural design is complex and precise, requiring not only excellent electrical performance but also the ability to withstand extreme operating environments such as high temperature, high voltage, and strong electromagnetic fields to ensure the efficient and stable operation of the equipment. During actual operation, the insulating material of the winding gradually degrades due to the combined effects of thermal aging, electrical aging, mechanical stress, and environmental factors, resulting in a decline in insulation performance, which in turn threatens the reliable operation and safety stability of the entire electrical equipment. Therefore, accurately evaluating the remaining life of winding insulation and taking timely maintenance measures are of great significance for preventing sudden failures, extending the service life of equipment, and ensuring the safety of the power system.

[0003] The process of winding insulation degradation is extremely complex and affected by various internal and external factors, making it a great challenge to obtain a comprehensive and accurate mechanism model. At the same time, it is difficult and costly to collect insulation degradation data. When traditional data-driven methods deal with problems such as the highly complex and sample-scarce insulation degradation process, their prediction accuracy and generalization ability are often severely restricted. Specifically, in the absence of sufficient data support, these methods are difficult to effectively capture the non-linear relationships and potential laws in the insulation degradation process, thereby affecting the accuracy and reliability of the prediction results.

[0004] Therefore, in the field of predicting the insulation degradation process, how to improve the accuracy and generalization ability of the prediction model under the condition of scarce samples has become a key problem to be solved urgently. Given the crucial role of predicting the remaining life of winding insulation in equipment maintenance and safety management, as well as the current challenges of data scarcity and model interpretability, it is particularly urgent to propose a data-driven prediction method guided by degradation mechanism knowledge. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to propose a method for predicting the remaining life of winding insulation jointly driven by the degradation mechanism of insulating materials and data, which improves the accuracy and generalization ability of the prediction model under the condition of scarce samples. The aim is to combine the physical mechanism of winding insulation degradation with the powerful data processing ability of neural networks to achieve high-precision, high-generalization ability, and strong interpretability prediction of the remaining life of winding insulation under the condition of few-sample data, providing a more scientific and reliable basis for the preventive maintenance and safety management of electrical equipment.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] The present invention proposes a method for predicting the remaining life of winding insulation driven by the degradation mechanism of insulation materials and data, comprising the following steps:

[0008] Step 1: Collect data on the changes in winding insulation characteristics under different thermal stress levels and construct a dataset containing temperature, insulation characteristics, and measurement time points. Normalize the data to ensure the stability and efficiency of model training.

[0009] Step 2: Considering that the higher the stress, the faster the insulation degradation, the degradation data under high stress level contains more significant and rich physical information, which helps the model quickly capture key degradation characteristics. Therefore, the sample data with the highest temperature is selected for analysis to determine the number of segments of the insulation state characteristic quantity and the segmentation threshold Q of the measurement value. n , n={1,…,I-1},where i={1,…,I},I is the number of self-defined segments, I≥1. According to the segmentation threshold Q of the measured value n Divide the dataset into I groups.

[0010] Step 3: According to the physical mechanism of insulation material degradation, the oxidation reaction and hydrolysis reaction involved in the degradation of insulation materials in the environment are considered, and the Arrhenius equation is used as the reaction rate model to establish a mathematical model of the degradation process. Because the material not only has oxidation reaction and hydrolysis reaction under thermal stress, but also involves multiple types of physical and chemical reactions such as creep behavior, microcrack extension, and material phase change that are difficult to model. At the same time, the intermediate products produced by the oxidation reaction and the hydrolysis reaction will form one or more continuous chain reactions, and the concentration of the reactants is time-varying. That is, the activation energy and pre-exponential factor in the Arrhenius reaction rate model vary with the reaction type and thermal stress. Therefore, a learnable reaction rate segmented equivalent factor k is defined. i To distinguish the reaction rates of different degradation stages. Then, the insulation degradation mathematical model is fitted in sections. The reaction rate section equivalent factor k is determined based on the reaction rate results of the section fitting. i The initial value of .

[0011] Step 4: Design a reaction rate scaling function based on the thermal properties of the material and the mechanism by which thermal stress affects the reaction rate To describe the response characteristics of the material under different thermal stress levels and define a learnable temperature response factor α i and the time lag coefficient σ i .

[0012] Step 5: Design the intercept compensation function χ i(T) To describe the cumulative amount of reaction rate differences over time that the reaction rate scaling function fails to describe. Define a learnable intercept factor and a thermal stress sensitivity factor v i .

[0013] Step 6: According to the reaction rate piecewise equivalent factor k i designed in Steps 3 - 5, the reaction rate scaling function and the intercept compensation function χ i (T), obtain the mathematical model of insulation degradation for the i-th segment to achieve the differentiation of the physical mechanism model. This differentiation strategy not only better conforms to the multi-stage reaction characteristics of the degradation process but also helps us understand the internal connections and conversion mechanisms between different degradation stages more deeply.

[0014] Step 7: Concatenate the time t, temperature T, t×T, and the mathematical model of insulation degradation M i (t, T) into a new feature vector, and use this feature vector as the input of the neural network model.

[0015] Step 8: To avoid overfitting of the mathematical model of insulation degradation and unreasonable values of physical parameters, add a regularization term to limit the range of variation of learnable parameters in the model. Effectively constrain the model complexity through regularization techniques to improve the generalization ability and prediction accuracy of the model. Then, considering the overall distribution and trend of the dataset, combine the differences between the predicted values and the true values and the statistical quantities of the dataset based on the statistical analysis results of the measured values of the insulation state characteristic quantities of each sample under the same stress level and the same measurement time point to construct a new loss function, providing more accurate gradient information for the neural network, thereby accelerating convergence.

[0016] Step 9: Train the i-th group of datasets in sequence, update the parameters of the neural network and the parameters of the mathematical model of insulation degradation for the i-th segment. Then, freeze the parameters of the mathematical model of insulation degradation and update the parameters of the neural network again using the complete training set. The output ΔM of the neural network model is summed with the value M i (t, T) of the mathematical model to obtain the final predicted value Y.

[0017] This method constructs a composite prediction model for predicting the remaining life of winding insulation materials by integrating the degradation mechanism of winding insulation materials and neural network technology. This method not only improves the prediction accuracy but also enhances the interpretability and robustness of the model.

[0018] Meanwhile, the present invention proposes a system for predicting the remaining life of winding insulation, including:

[0019] The data acquisition and processing unit collects the change data of the winding insulation state characteristic quantities under different thermal stress levels, constructs a data set and normalizes the data; selects the sample data with the highest temperature for analysis, determines the number of segments of the insulation state characteristic quantities and the segmentation thresholds of the measured values, and divides the data set through the segmentation thresholds.

[0020] The insulation degradation model construction unit calculates the initial theoretical value of the insulation material degradation rate using the Arrhenius equation and constructs a segmented insulation degradation mathematical model to describe the insulation material degradation process.

[0021] The neural network model construction unit splices the time, temperature and the values of the insulation degradation mathematical model into a new feature vector as the input of the neural network model; introduces the boundary penalty function of the mathematical model parameters to jointly optimize the insulation degradation mathematical model and the neural network, and constructs the loss function of the neural network model in combination with the statistics of the degradation data.

[0022] The prediction result output unit is used to perform progressive training on the neural network, and generates the prediction result by superimposing the theoretical aging component of the insulation degradation model and the residual compensation component learned by the neural network.

[0023] Finally, the present invention also proposes a computer-readable storage medium storing computer instructions, and the computer instructions are used to make the computer execute the steps of the method of the present invention.

[0024] The present invention adopts the above technical solutions and has the following technical effects compared with the prior art:

[0025] (1) The winding insulation remaining life prediction model provided by the present invention is based on the degradation mechanism of the insulation material, and constructs a segmented mathematical model to more accurately describe the change law of the insulation performance in different stages. Compared with the traditional segmented mathematical model with the change rate of the insulation state characteristic quantity as the judgment factor, the model proposed by the present invention does not need to carefully analyze the time-varying law of the change rate of the insulation state characteristic quantity, and only needs to directly specify the segmentation threshold of the characteristic quantity, which is simpler and more convenient.

[0026] (2) The present invention introduces a data-driven method, uses the neural network to learn the data, and adopts a joint optimization strategy of the segmented mathematical model and the neural network to adjust the parameters of the mathematical model and the neural network at the same time, so that the model can more accurately reflect the actual situation.

[0027] (3) The winding insulation remaining life prediction model provided by the present invention not only retains the interpretability of the physical mechanism model, but also improves the adaptability and prediction accuracy of the model to complex working conditions through a data-driven method. Description of the Drawings

[0028] Figure 1 is the schematic diagram of the implementation process of the present invention.

[0029] Figure 2 It is a graph of the change trajectory of partial sample insulation degradation characteristic quantities in an embodiment of the present invention.

[0030] Figure 3 It is a graph of the prediction result of the first segmented change trajectory of the insulation material degradation characteristic quantity at an aging temperature of 320 °C in an embodiment of the present invention.

[0031] Figure 4 It is a graph of the prediction result of the first segmented change trajectory of the insulation material degradation characteristic quantity at an aging temperature of 290 °C in an embodiment of the present invention.

[0032] Figure 5 It is a graph of the prediction result of the third segmented change trajectory of the insulation material degradation characteristic quantity at an aging temperature of 320 °C in an embodiment of the present invention.

[0033] Figure 6 It is a graph of the prediction result of the third segmented change trajectory of the insulation material degradation characteristic quantity at an aging temperature of 290 °C in an embodiment of the present invention. Detailed implementation manners

[0034] Next, in combination with the accompanying drawings of the present invention, the technical solutions in the embodiments of the present invention will be further described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and cannot be used to limit the protection scope of the present invention.

[0035] Embodiment 1:

[0036] This embodiment proposes a method for predicting the remaining life of winding insulation jointly driven by insulation degradation mechanism and data, as Figure 1 shown, the specific steps are as follows:

[0037] Step 1: Continuously monitor and record the changes in the winding insulation performance characteristic quantities under different thermal stress levels. At the same time, synchronously record the key parameters during the experiment, including but not limited to the ambient temperature, the actual measured values of the samples (i.e., the above insulation performance characteristic quantities), and the time points during measurement.

[0038] According to the two dimensions of temperature and time, the collected insulation performance measurement values are divided into different data groups. Trend analysis is performed on the measurement values within each data group to explore the variation laws of the measurement values with temperature and time. Calculate statistical quantities such as the average value, median, maximum value, and minimum value of each data group.

[0039] In order to ensure the stability and efficiency of model training, it is necessary to normalize the data. The extreme value normalization method is adopted to scale the data to the interval [0, 1]. The specific formula is:

[0040]

[0041] Among them, X is the original data, X min and X min are the minimum and maximum values of the data respectively.

[0042] To meet the requirements of the neural network model for the consistency of the input sequence length, it is necessary to preprocess the collected data of the winding insulation performance characteristic quantities. The specific steps are as follows:

[0043] 1. Traverse all sample sequences and determine the length of the longest sequence as the benchmark.

[0044] 2. Time feature filling: Supplement data points with a time point of 0 at the starting position of the sequence.

[0045] 3. Measurement value filling: Fill the measurement value with 0 to represent the missing or initial state.

[0046] 4. Temperature feature filling: Keep the temperature value unchanged and use the temperature value of the sample sequence itself for filling.

[0047] In step 2, considering that the higher the stress, the faster the insulation degradation, the degradation data under high stress contains more significant and rich physical information, which helps the model quickly capture the key degradation characteristics. Therefore, select the sample data with the highest temperature for analysis. First, determine the number of segments of the mathematical model and the segmentation threshold Q n ; According to the segmentation threshold Q n of the measurement value, divide the data set into group I.

[0048] In step 3, oxygen molecules and water molecules in the ambient air will react chemically with the insulating material. The relationships between the rates of oxidation reaction and hydrolysis reaction and temperature can both be described by the Arrhenius equation. Therefore, the degradation reaction rate model:

[0049]

[0050] where k A is the reaction rate; A is the pre-exponential factor, indicating the maximum possible rate of the reaction at a sufficiently high temperature; E a is the activation energy, indicating the energy barrier for the reaction to occur; R is the gas constant (8.314 J / mol·K); T is the absolute temperature (unit: K).

[0051] On the basis of the degradation reaction rate model, further construct the initial mathematical model of the degradation process:

[0052] M A = k A t + d.

[0053] Considering that under the action of thermal stress, the material not only has oxidation reaction and hydrolysis reaction, but also involves various types of physical and chemical reactions such as creep behavior, microcrack propagation, and material phase transformation that are difficult to model. At the same time, the intermediate products generated by the oxidation reaction and hydrolysis reaction will form one or more continuous chain reactions, and the reactant concentration has time-variability. The activation energy and pre-exponential factor in the Arrhenius reaction rate model vary with different reaction types and thermal stresses. Therefore, define a learnable reaction rate piecewise equivalent factor k i , i = {1,..., I}.

[0054] Use I groups of data to fit the initial mathematical model M A , and obtain the reaction rate k A = {k A1 ,..., k Ai ,...k AI}, and the initial value of the reaction rate piecewise equivalent factor k i in the insulation degradation mathematical model M i (t, T) of the i-th segment is taken as k Ai .

[0055] In step 4, according to the thermal characteristics of the material and the influence mechanism of thermal stress on the reaction rate, define a learnable temperature response factor α i and a time lag coefficient σ i , and design a reaction rate scaling function to describe the response characteristics of the material at different temperatures. Obtain the comprehensive reaction rate f i at thermal stress T:

[0056]

[0057] From k A to f i realizes the differentiation of the physical mechanism model. This differentiation strategy not only better conforms to the multi-stage reaction characteristics of the degradation process, but also helps us to more deeply understand the internal connection and conversion mechanism between different degradation stages.

[0058] In step 5, considering that the reaction rate scaling function cannot accurately describe the difference in the reaction rate of the material under different thermal stresses, define a learnable intercept factor and a thermal stress sensitivity factor v i , and design an intercept compensation function χ i (T) to describe the cumulative amount of the reaction rate difference not described by the reaction rate scaling function over time.

[0059] In step 6, according to the reaction rate piecewise equivalent factor k i , the reaction rate scaling function and the intercept compensation function χ designed in steps 3 - 5i (T), the mathematical model of insulation degradation of the i-th segment is obtained:

[0060]

[0061] In step 7, the time t, temperature T, t×T and insulation degradation mathematical model M are i (t, T) are concatenated into a new feature vector:

[0062] x=[t,T,t×T,M i (t, T)],

[0063] This feature vector is used as the input of the neural network model.

[0064] This example uses LSTM for neural network modeling. The calculation process of the LSTM unit at each time step t is as follows:

[0065] i t =σ(W xi x t +W hi h t-1 +b i ),

[0066] f t =σ(W xf x t +W hf h t-1 +b f ),

[0067] o t =σ(W xo x t +W ho h t-1 +b o ),

[0068]

[0069] h t =o t tanh(C t ),

[0070] Among them, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, C t-1 is the cell state at the previous time step, W xi 、W xf 、W xo and W xc are the weight matrices of the input gate, forget gate, output gate and candidate value, respectively, W hi 、W hf 、Who and W hc are the weight matrices corresponding to the hidden state, b i , b f , b o and b c represent the bias terms, and σ(.) represents the activation function.

[0071] In step 8, to avoid overfitting of the model and unreasonable values of physical parameters, a regularization term is added to limit the reaction rate piecewise equivalent factor k i , the temperature response factor α i , the time lag coefficient σ i , the intercept factor and the thermal stress sensitivity factor v i . By introducing regularization technology, the generalization ability and prediction accuracy of the model are improved.

[0072] The regularization term is expressed as:

[0073]

[0074] where f(·) represents the parameter boundary penalty function to be designed, and λ k represents the weight coefficient corresponding to each parameter boundary penalty function, and k = {1, 2, 3, 4, 5}.

[0075] Considering the overall distribution and trend of the data set, the difference between the predicted value and the true value is combined with statistical quantities such as the mean and median of the data set to construct a new loss function:

[0076]

[0077] where MSE represents the mean square error term, and Penalty stat represents the penalty function designed based on statistical quantities (mean, median, maximum, minimum), and Penalty param represents the boundary penalty function of the mathematical model parameters, and η, and φ are hyperparameters.

[0078] Train I groups of data sets in sequence. The physical model parameters are jointly trained with the LSTM to update the parameters of the LSTM and the parameters of the i-th segment insulation degradation mathematical model. According to the gradient descent method, the parameters are continuously updated, and the parameter update formula is:

[0079]

[0080] Then, freeze the parameters of the insulation degradation mathematical model and update the parameters of the LSTM again using all the training sets. The output ΔM of the LSTM model is summed with the value M i (t, T) of the mathematical model to obtain the final predicted value Y:

[0081] Y = M i (t, T) + ΔM.

[0082] In this embodiment, simulation tests are carried out in Python, and the following Python libraries and modules need to be imported: Pandas, NumPy, PyTorch, Matplotlib, torch.nn, torch.optim. The dataset is the degradation data of the winding insulation material at four temperatures of 290 °C, 300 °C, 310 °C and 320 °C. There are 9 samples at each temperature, and the time length of the samples is 300 hours. Figure 2 Give the trajectories of the insulation degradation characteristic quantities of some samples to show the change trend of the degradation amount at different temperatures.

[0083] Calculate statistics such as the mean, median, maximum and minimum of each data group. Using the extreme value normalization method, the data is scaled to the interval [0, 1]. Check the sample sequence length to ensure that all sample sequence lengths are the same.

[0084] Select the number of segments of the mathematical model to be 3, and the segmentation thresholds of the measured values are selected as: Q1 = 50, Q2 = 350. Divide the dataset according to the size relationship between the measured value and Q1 and Q2 to obtain 3 sub-datasets. The 3 sub-datasets are respectively used to fit the initial mathematical model M in step 2 A , to obtain k A = {0.6, 2.7, 8.2}, and take the initial value of the reaction rate segmentation equivalent factor k in the mathematical model M i (t, T) to be k i . Ai .

[0085] Design the reaction rate scaling function

[0086]

[0087] To obtain the comprehensive reaction rate f under the thermal stress T i :

[0088]

[0089] Design the intercept compensation function x i (T):

[0090]

[0091] To obtain the insulation degradation mathematical model of the i-th segment:

[0092]

[0093] Further, the time t, temperature T, t×T, and the insulation degradation mathematical model M i (t, T) are spliced into a new feature vector:

[0094] X = [t, T, t×T, M i (t, T)],

[0095] and this feature vector is used as the input of the LSTM model.

[0096] Further, considering the overall distribution and trend of the data set, the difference between the predicted value and the true value is combined with statistical quantities such as the mean and median of the data set to form a new loss function:

[0097]

[0098] where Y pred,i is the predicted value of the i-th sample, Y true，i is the true value of the i-th sample, Y ave，i , Y med,i , Y max，i and Y min，i are the mean, median, maximum, and minimum values of the true values of all samples corresponding to the i-th sample under the thermal stress and degradation time, x min and x max are the minimum allowable value and the maximum allowable value of the parameter x,

[0099] Train the I group of data sets in sequence. The physical model parameters and the LSTM are jointly trained to update the parameters of the LSTM and the parameters of the i-th segment of the insulation degradation mathematical model. Then, freeze the parameters of the insulation degradation mathematical model and update the parameters of the LSTM again using all the training sets. The output ΔM of the LSTM model and the value M i (t, T) of the mathematical model are summed to obtain the final predicted value Y:

[0100] Y = M i (t, T) + ΔM.

[0101] Figures 3 - 6 is the result graph predicted by the trained LSTM model and the mathematical model. Comparing the results at 290°C and 320°C, it can be seen that the degradation rate shows a non-linear relationship with both temperature and time. The time to reach the threshold Q1 at 320°C is about 3 times that at 290°C, that is, the comprehensive reaction rate at 290°C must be smaller than that at 320°C, but the reaction rate cannot be processed according to the proportional relationship of time. In the present invention, the reaction rate scaling function Under the action of [mechanism], the comprehensive reaction rate at a temperature of 290 °C is consistent with its actual degradation rate. Similarly, through Figures 5 - 6 it can be seen that even if there is an intercept difference caused by different degradation rates before and after segmentation, under the action of the intercept compensation function χ i (T) in the present invention, the comprehensive reaction rate at a temperature of 290 °C is consistent with its actual degradation rate. In summary, the insulation degradation mechanism and the data-driven winding insulation remaining life prediction method proposed in the present invention are feasible and effective.

[0102] In summary, the present invention proposes a method for predicting the remaining life of winding insulation by combining insulation material degradation mechanism and data. This method realizes the accurate prediction of the non-linear time-varying trajectory of insulation state characteristic quantities by combining segmented kinetic modeling and neural network learning. At the level of physical mechanism modeling, supported by the Arrhenius equation, comprehensively considering the mechanisms of oxidation reaction and hydrolysis reaction of insulation materials and the time-variability of degradation rate, a segmented mathematical model is constructed to describe the degradation process of insulation materials. By introducing learnable segmented mathematical model parameters (including reaction rate segmented equivalent factor, temperature response factor, time lag coefficient, intercept factor and thermal stress sensitive factor), a degradation state mapping relationship for cross-stress level migration is established, so as to effectively generalize the degradation characteristics represented by the degradation data under high stress conditions of insulation materials to the actual working condition stress level. At the data-driven level, the data set is divided by segmented thresholds, and each sub-data set is used to progressively train the neural network. A regularization term with physical constraints is introduced, and a weighted loss function is designed in combination with the statistics (mean, median, maximum and minimum) of the degradation data. This design not only enhances the adaptability of the model to different samples, but also realizes the joint optimization of the mechanism model parameters and the neural network weights through collaborative iterative optimization. The final prediction result is generated by superimposing the theoretical aging component of the mechanism model and the residual compensation component learned by the neural network. While maintaining the strong interpretability of the physical model, this method significantly improves the model's characterization ability for the complex degradation process of insulation materials through segmented modeling and data driving, providing a high-precision solution with physical consistency for predicting the remaining life of winding insulation.

[0103] Embodiment 2:

[0104] This embodiment proposes a system for predicting the remaining life of winding insulation, including the following units:

[0105] A data acquisition and processing unit, configured to collect the change data of the winding insulation state characteristic quantities under different thermal stress levels, construct a data set and perform normalization processing on the data; a segmented data division unit, select the sample data with the highest temperature for analysis, determine the number of segments of the insulation state characteristic quantities and the segmented thresholds of the measured values, and divide the data set by the segmented thresholds;

[0106] An insulation degradation model construction unit calculates an initial theoretical value of the degradation rate of an insulation material using the Arrhenius equation and constructs a piecewise insulation degradation mathematical model to describe the degradation process of the insulation material;

[0107] A neural network model construction unit concatenates the values of time, temperature, and the piecewise insulation degradation mathematical model into a new feature vector as the input of the neural network model; introduces a boundary penalty function for the mathematical model parameters to jointly optimize the insulation degradation mathematical model and the neural network, and constructs a loss function for the neural network model in combination with the statistics of the degradation data;

[0108] A prediction result output unit is used to perform progressive training on the neural network and generate a prediction result by superimposing the theoretical aging component of the insulation degradation model and the residual compensation component learned by the neural network.

[0109] Embodiment 3:

[0110] This embodiment provides a computer-readable storage medium with a computer program stored thereon. When the computer program is executed by a processor, it implements the steps of the method of the present invention, which will not be elaborated here.

[0111] It should be noted that the processing flows of Embodiment 2 to Embodiment 3 correspond to the specific steps of the method provided by the embodiments of the present invention and have the corresponding functional modules and beneficial effects of the execution method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiments of the present invention.

[0112] The program code for implementing the method of the present application can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, so that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program codes can be executed entirely on the machine, partially on the machine, executed as an independent software package partially on the machine and partially on a remote machine, or executed entirely on a remote machine or server.

[0113] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0114] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0115] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the description in the specification are only illustrative of the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all such changes and improvements fall within the scope of the present invention as claimed.

Claims

1. A method for predicting the remaining life of winding insulation jointly driven by the degradation mechanism and data of insulating materials, characterized in that It includes the following steps: S1. Collect the change data of the winding insulation state characteristic quantities under different thermal stress levels, construct a data set and normalize the data; select the sample data with the highest temperature for analysis, determine the number of segments of the insulation state characteristic quantities and the segmentation thresholds of the measured values, and divide the data set through the segmentation thresholds; S2. Calculate the initial theoretical value of the insulation material degradation rate using the Arrhenius equation, and construct a segmented insulation degradation mathematical model to describe the insulation material degradation process; S3. Concatenate the time, temperature, and the values of the insulation degradation mathematical model into a new feature vector as the input of the neural network model; introduce the boundary penalty function of the mathematical model parameters to jointly optimize the insulation degradation mathematical model and the neural network, and construct the loss function of the neural network model in combination with the statistics of the degradation data; S4. Conduct progressive training on the neural network, and generate the prediction result by superimposing the theoretical aging component of the insulation degradation model and the residual compensation component learned by the neural network.

2. The method according to claim 1, wherein In step S2, the Arrhenius equation is fitted according to the sub-datasets after segmented division to obtain the initial theoretical value of the degradation rate of the insulating material, and the reaction rate segmented equivalent factor k is determined based on this result i for the initial value.

3. The method according to claim 1, characterized in that, In step S2, the mathematical model for establishing the insulation degradation process is as follows: Among them, M i (t, T) represents the value of the insulation degradation mathematical model for the i-th segment, and k i represents the learnable reaction rate piecewise equivalent factor used to distinguish the reaction rates in different degradation stages, where i = {1,..., I}, I is the user-defined number of segments, and I ≥; represents the reaction rate scaling function used to describe the response characteristics of the material under different thermal stress levels; represents the intercept compensation function used to describe the cumulative amount of the reaction rate difference not described by the reaction rate scaling function over time, where t represents time and T represents temperature.

4. The method according to claim 1, wherein In step S3, time t, temperature T, t×T, and the insulation degradation mathematical model M i (t, T) are concatenated into a new feature vector, and this feature vector is used as the input of the neural network model; a boundary penalty function for the mathematical model parameters is introduced, and a regularization term is added to limit the variation range of the learnable parameters in the insulation degradation mathematical model. Then, based on the statistical analysis results of the measured values of the insulation state characteristic quantities of each sample under the same stress level and the same measurement time point, the differences between the predicted values, the true values, and the data group statistics are combined to construct the loss function of the neural network model.

5. The method according to claim 1, characterized in that, In step S3, combine the differences between the predicted value, the true value, and the data set statistics, and design the loss function as: Among them, MSE represents the mean squared error term, and Penalty stat represents the penalty function designed based on statistics, and Penalty param represents the boundary penalty function of the mathematical model parameters, and η, and φ are hyperparameters.

6. The method according to claim 1, wherein Step S4 specifically is: divide the data set into group I according to the segmentation thresholds, train the i-th group of data sets in sequence, update the parameters of the neural network and the parameters of the i-th segment insulation degradation mathematical model; then, freeze the parameters of the insulation degradation mathematical model, use the complete training set to update the parameters of the neural network again, and sum the output of the neural network model and the value of the insulation degradation mathematical model to obtain the final predicted value.

7. The method according to claim 3, characterized in that, Reaction rate scaling function has the form including: Among them, σ i represents the time lag coefficient, α i represents the temperature response factor, T max represents the maximum temperature.

8. The method according to claim 3, characterized in that, Intercept compensation function χ i (T) has the following forms: Among them, represents the intercept factor, υ i represents the thermal stress sensitivity factor.

9. A winding insulation remaining life prediction system, characterized in that, It includes: A data acquisition and processing unit, which collects the change data of the winding insulation state characteristic quantities under different thermal stress levels, constructs a data set and normalizes the data; Select the sample data with the highest temperature for analysis, determine the number of segments of the insulation state characteristic quantities and the segmentation thresholds of the measured values, and divide the data set through the segmentation thresholds; An insulation degradation model construction unit, which calculates the initial theoretical value of the insulation material degradation rate using the Arrhenius equation, and constructs a segmented insulation degradation mathematical model to describe the insulation material degradation process; A neural network model construction unit, which concatenates the time, temperature, and the values of the insulation degradation mathematical model into a new feature vector as the input of the neural network model; introduces the boundary penalty function of the mathematical model parameters to jointly optimize the insulation degradation mathematical model and the neural network, and constructs the loss function of the neural network model in combination with the statistics of the degradation data; A prediction result output unit, which is used to conduct progressive training on the neural network, and generate the prediction result by superimposing the theoretical aging component of the insulation degradation model and the residual compensation component learned by the neural network.

10. A computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the method described in any one of claims 1-8.

Images