Oil paper insulation polymerization degree DP value prediction method fusing frequency domain dielectric spectrum characteristics and BiLSTM

Abstract

The invention relates to an oil paper insulation polymerization degree DP value prediction method fusing frequency domain dielectric spectrum characteristics and BiLSTM, and the method comprises the steps: (1) based on an FDS test platform, measuring frequency domain dielectric spectrum curves of oil-impregnated paper samples with different polymerization degrees DP; (2) extracting segmented integral features and complex dielectric constant real and imaginary part features based on the frequency domain dielectric spectrum curve; (3) expanding the sample size through nonlinear fitting, and constructing a data set of a polymerization degree DP value prediction model; (4) constructing a bidirectional long short-term memory network BiLSTM; (5) constructing a polymerization degree DP value prediction model based on a BiLSTM network; dividing the data set into a training set and a test set, training the model by using the training set, and verifying the prediction accuracy of the model on the polymerization degree DP value through the test set; and (6) performing polymerization degree DP value prediction through the obtained polymerization degree DP value prediction model. According to the method, the prediction precision and reliability of the oil paper insulation polymerization degree DP value can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer evaluation, and in particular to a method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency-domain dielectric spectrum characteristics with BiLSTM. Background Art

[0002] Transformers are core components of power systems, and their insulation performance directly impacts the safety and stability of grid operations. The oil-paper insulation system (composed of transformer oil and insulating paper) is the transformer's primary protective insulation structure. Its degree of aging (characterized by its degree of polymerization (DP) value, with lower DP values ​​indicating more severe aging) is a key indicator for assessing transformer lifespan.

[0003] Frequency-domain Dielectric Spectroscopy (FDS) has become an important means of evaluating the status of oil-paper insulation in recent years due to its characteristics of non-destructive measurement, sensitivity to insulation degradation status, and rich insulation information. In the existing technology, the analysis of FDS data is mainly divided into two categories: one is to use a single feature index (such as the dielectric loss factor at a specific frequency) to characterize the degree of insulation degradation. Due to the single feature selection, this type of method cannot fully reflect the insulation aging information; the other is to establish a prediction model based on multiple simple feature indicators (such as the dielectric loss factor values ​​at different frequency points) combined with artificial intelligence algorithms such as support vector machines (SVM) and long short-term memory networks (LSTM). For example, the patent "A method for evaluating the aging of transformer oil-paper insulation based on LSTM" (CN117214632A) selects the dielectric loss factor values ​​at different frequencies as features and combines them with the LSTM model to predict the DP value, but this method still has the following shortcomings:

[0004] (1) Limitations of characteristic parameter selection: Relying only on the dielectric loss factor values ​​at discrete frequency points cannot fully extract the aging information of the insulation sample, and the measurement error of a single sampling point can easily interfere with the evaluation results;

[0005] (2) Algorithm limitations: LSTM only predicts future results through forward information and does not utilize the global information of the sequence, resulting in limited prediction accuracy;

[0006] (3) Insufficient sample size: The number of oil-paper insulation samples obtained in actual experiments is limited, which is difficult to meet the training requirements of complex models.

[0007] Therefore, how to more comprehensively extract the aging characteristics in FDS, optimize the prediction algorithm to utilize global information, and solve the problem of insufficient sample size have become key challenges to improve the prediction accuracy of the DP value of oil-paper insulation. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics with BiLSTM, which can improve the accuracy and reliability of the prediction of the degree of polymerization (DP) value of oil-paper insulation.

[0009] To achieve the above objectives, the present invention adopts a technical solution: a method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics with BiLSTM, comprising:

[0010] (1) Based on the FDS test platform, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees were measured;

[0011] (2) Extracting the piecewise integral characteristics and the real and imaginary part characteristics of the complex dielectric constant based on the frequency domain dielectric spectrum curve;

[0012] (3) Expand the sample size through nonlinear fitting and construct a data set for the DP value prediction model;

[0013] (4) Construct a bidirectional long short-term memory network BiLSTM;

[0014] (5) Construct a DP value prediction model based on the BiLSTM network; divide the data set into a training set and a test set, use the training set to train the model, and use the test set to verify the model's prediction accuracy of the DP value;

[0015] (6) The obtained DP value prediction model is used to predict the DP value.

[0016] Furthermore, in step (1), the FDS test platform is composed of a data acquisition computer, an FDS tester, a constant temperature box, and a three-electrode test device placed in the constant temperature box;

[0017] Based on the FDS test platform, at a test temperature of 35°C, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees were measured, including the real and imaginary part test curves of the complex dielectric constant and the dielectric loss tangent test curves.

[0018] Furthermore, in step (2), the method for extracting the segmented integral feature is:

[0019] Select the frequency f∈[10 -3 Hz,10 1 The dielectric loss tangent value tanδ(f) at [Hz] is taken as the frequency sensitive interval, and tanδ(f) within this frequency sensitive interval is piecewise integrated. The piecewise integral formula is expressed as follows:

[0020]

[0021] Among them, D1, D2, D3, and D4 are characteristic parameters obtained by piecewise integration.

[0022] Furthermore, in step (2), the method for extracting the real and imaginary part characteristics of the complex dielectric constant is:

[0023] The complex dielectric constant parameter is expressed as follows:

[0024]

[0025] Where ω is the angular velocity corresponding to the frequency of the applied voltage in the FDS test; and denote the conductivity component and relaxation component of the complex dielectric constant under pressure; σ0 denotes the static conductivity; ε s and ε ∞ represent the static dielectric constant and the optical frequency dielectric constant respectively; τ represents the relaxation time constant;

[0026] The relaxation component after removing the conductivity component is:

[0027]

[0028] Among them, ε' p (ω) and ε″ p (ω) are respectively The real and imaginary parts of

[0029] The relaxation characteristics are highlighted by the first-order differential of the real part of the complex dielectric constant. The specific method is:

[0030] First, perform coordinate transformation, let ω = e x , then differentiate the real part of the complex dielectric constant once, expressed as:

[0031]

[0032] The peak point and integral amount are selected as the aging-relaxation related characteristic indicators, which are expressed as:

[0033]

[0034] Among them, D5 is the characteristic parameter characterizing the peak point, and D6 is the characteristic parameter characterizing the integral amount.

[0035] Furthermore, in step (3), the sample size is expanded by nonlinear fitting to construct a data set for the DP value prediction model, and the implementation method is as follows:

[0036] 301) Setting nonlinear function variables, and making the six characteristic parameters D1 to D6 extracted based on the frequency domain dielectric spectrum curve correspond to functions y1 to y6 respectively;

[0037] 302) respectively determining the nonlinear function type of each characteristic parameter, constructing the nonlinear function by setting the function parameters, and using the goodness of fit to characterize the fitting effect of the function, substituting the known independent variables into the solution, and identifying the function parameters;

[0038] 303) Construct an extended dataset φ based on the nonlinear function of the characteristic parameters D1~D6 A , as shown in formula (12):

[0039]

[0040] Among them, φ Aj ={y1(x 1j ),y2(x 2j ),y3(x 3j ),y4(x 4j ),y5(x 5j ),y6(x 6j )}, represents the characteristic parameter obtained by the nonlinear function corresponding to the expanded j-th group of data; x ij represents the DP value of the i-th characteristic parameter of the j-th group of data, where i represents the i-th characteristic quantity, j represents the expanded j-th group of data, y(x ij ) represents the value of the characteristic parameter corresponding to the DP value of the i-th characteristic parameter of the j-th group of data.

[0041] Furthermore, the nonlinear function of the characteristic parameter D1 is: y1 = a + bcoscx + dsincx; where a = 0.02211, b = 0.006038, c = 0.003854, d = 0.01622, and the goodness of fit is 1.0000;

[0042] The nonlinear function of characteristic parameter D2 is: y2 = a + bcoscx + dsincx; where a = 0.03898, b = 0.05704, c = 0.002259, d = 0.07015, and the goodness of fit is 1.0000;

[0043] The nonlinear function of the characteristic parameter D3 is: Among them, a = 1.001, b = 469.1, c = 381.6, and the goodness of fit is 0.9988;

[0044] The nonlinear function of the characteristic parameter D4 is: Among them, a = 7.899, b = 396.8, c = 282, and the goodness of fit is 0.9999;

[0045] The nonlinear function of the characteristic parameter D5 is: y5 = ae bx +ce dx; Among them, a = -0.1318, b = 0.001683, c = 6.13, d = -0.001353, goodness of fit 1.0000;

[0046] The nonlinear function of the characteristic parameter D6 is: Among them, a = 69.26, b = 423.8, c = 377.3, and the goodness of fit is 0.9987;

[0047] The dataset φ A 937 sets of data are expanded in the interval range [241,1178] with a DP value of 1 as a step size.

[0048] Furthermore, in step (4), the BiLSTM network is constructed based on the LSTM network;

[0049] The LSTM network is composed of the input information x at time t t , cell state C t , temporary cell state Hidden state h t 、Forget Gate t 、Memory Gate i t and input gate o t The calculation process of the LSTM network is as follows: by forgetting the information in the cell state and memorizing new information, the information useful for subsequent moment calculations can be transmitted, while the useless information is discarded, and the hidden state h is output at each time step. t , where forgetting, memory and output are determined by the hidden state h at the previous moment t-1 and the current input x t The calculated forget gate f t 、Memory Gate i t and output gate o t To control; specifically:

[0050] First, decide what information needs to be forgotten from the cell state; by t-1 and the current input information x t The weighted sum plus the bias is processed and passed to the Sigmoid layer, so that it outputs a function f with a value between 0 and 1 t ;

[0051] f t =σ(W t ·[h t-1 ,x t ]+b f ) (13)

[0052] Secondly, determine the new information that needs to be stored in the cell state; the previous hidden state information h t-1 and current input

[0053] Input information x t Input into the Sigmoid layer and use the tanh layer to create a vector of new candidate values Combine the two outputs as the current output to update the cell state;

[0054] i t =σ(W t ·[h t-1 ,x t ]+b i ) (14)

[0055]

[0056] Finally, for the previous hidden state information h t-1 and the current input information x t Run a Sigmoid layer to determine the information that needs to be output in the cell state;

[0057] o t =σ(W o ·[h t-1 ,x t ]+b o ) (17)

[0058] The cell state is passed through the tanh layer and multiplied by the output of the sigmoid layer, thus combining the cell state and the new hidden state h t Pass to the next time step;

[0059] h t =o t tanh(C t ) (18)

[0060] The BiLSTM network introduces two LSTM networks, the forward LSTM and the backward LSTM, to simultaneously learn features from both the forward and backward directions of the sequence. The calculation process is as follows:

[0061]

[0062] Furthermore, in step (5), the optimal parameter combination of the BiLSTM network is determined to construct a DP value prediction model;

[0063] The optimal parameter combination of the BiLSTM network is: 2 hidden layers, 20 neurons in each layer, a learning rate of 0.01, and 500 iterations.

[0064] Furthermore, the degree of polymerization (DP) value prediction model achieves quantitative prediction of the DP value of unknown samples by learning the nonlinear mapping relationship between the characteristic parameters of the frequency domain dielectric spectrum and the degree of polymerization (DP) value of oil-paper insulation.

[0065] The present invention also provides an oil-paper insulation polymerization degree DP value prediction system that integrates frequency domain dielectric spectrum characteristics and BiLSTM, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above method can be implemented.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] 1. More comprehensive feature parameter selection and stronger anti-interference ability:

[0068] Existing technologies mostly use dielectric loss factor values ​​at discrete frequency points as features, which are easily affected by measurement errors at single sampling points. This application analyzes the correlation between the FDS curve morphology and the degree of aging, and selects the piecewise integral of the dielectric loss tangent value within the frequency-sensitive interval, as well as the peak point and integral of the first-order differential of the real part of the complex dielectric constant as characteristic parameters. The piecewise integral reduces the influence of single-point errors through the integration operation of continuous intervals, while the first-order differential of the real part of the complex dielectric constant highlights the correlation between relaxation characteristics and the degree of aging, significantly improving the feature's ability to characterize the aging state.

[0069] 2. Sample size expansion method to solve the limitation of small sample:

[0070] The limited number of oil-paper insulation samples obtained in actual experiments makes it difficult to train complex models. This application uses a nonlinear fitting method to fit the relationship between six characteristic parameters and DP values, expanding multiple data sets within the DP value range. This effectively solves the small sample size problem and provides sufficient data for BiLSTM model training.

[0071] 3. BiLSTM algorithm improves prediction accuracy and reliability:

[0072] Traditional LSTMs only use the forward information of the sequence for prediction. However, the BiLSTM used in this application uses two LSTM networks, the forward and backward, to simultaneously learn the global information of the sequence, which can more comprehensively capture the nonlinear relationship between feature parameters and DP values. Experimental verification results show that the prediction accuracy of the BiLSTM model in this application is significantly better than that of the LSTM model that only uses forward information.

[0073] 4. Comprehensive evaluation indicators verify the effectiveness of the model:

[0074] This application uses MAE, MSE (mean square error), RMSE (root mean square error), MAPE (mean absolute percentage error) and R2 The model was verified by multiple indicators such as the training set and the test set, and all indicators performed well, proving that the model has good generalization ability and practical application value.

[0075] In summary, this application effectively solves the problems of insufficient feature extraction, insufficient sample size and limited prediction accuracy in the existing technology through feature parameter optimization, sample size expansion and BiLSTM algorithm improvement, and provides a more reliable technical solution for the quantitative evaluation of the DP value of oil-paper insulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flowchart of a method implementation of an embodiment of the present invention;

[0077] Figure 2 is a schematic diagram of an FDS test platform in an embodiment of the present invention;

[0078] Figure 3 1 is a graph showing the complex dielectric constant test curves of DP oil-impregnated paper samples with different polymerization degrees according to an embodiment of the present invention; wherein (a) is a test curve showing the real part of the complex dielectric constant; (b) is a test curve showing the imaginary part of the complex dielectric constant;

[0079] Figure 4 1 is a test curve of dielectric loss tangent value of DP oil-impregnated paper samples with different polymerization degrees in an embodiment of the present invention;

[0080] Figure 5 This is a diagram of the overall LSTM framework in an embodiment of the present invention;

[0081] Figure 6 1 is a comparison diagram of the predicted results of the data set and the actual values ​​in an embodiment of the present invention; wherein, (a) is a comparison diagram of the predicted results of the training set; (b) is a comparison diagram of the predicted results of the test set. DETAILED DESCRIPTION

[0082] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0083] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0084] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0085] like Figure 1 As shown, this embodiment provides a method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency-domain dielectric spectrum characteristics with BiLSTM, including:

[0086] Step 1: Based on the FDS test platform, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees are measured;

[0087] Step 2: Extract the piecewise integral characteristics and the real and imaginary part characteristics of the complex dielectric constant based on the frequency domain dielectric spectrum curve;

[0088] Step 3: Expand the sample size through nonlinear fitting and construct a data set for the DP value prediction model;

[0089] Step 4: Build a bidirectional long short-term memory network BiLSTM;

[0090] Step 5: Build a DP value prediction model based on the BiLSTM network; divide the data set into a training set and a test set, use the training set to train the model, and use the test set to verify the model's prediction accuracy for the DP value;

[0091] Step 6: Predict the degree of polymerization (DP) value using the obtained DP value prediction model.

[0092] Step 1: Platform construction and sample testing

[0093] Build FDS test platform, such as Figure 2 The FDS test platform consists of a data acquisition computer, an FDS tester, a constant temperature box, and a three-electrode test device placed in the constant temperature box.

[0094] Based on the FDS test platform, at a test temperature of 35°C, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees were measured, including the real and imaginary part test curves of the complex dielectric constant and the dielectric loss tangent test curves, as shown in Figure 2. Figure 3-4 shown.

[0095] Step 2: Data analysis and feature parameter extraction

[0096] By comparing the difference in dielectric loss tangent tanδ of oil-paper insulation system under different polymerization degrees DP, it is found that when the frequency f is 10 -3 ~10 1 Hz. Aging is highly sensitive to spectral line shape, and the increased effects of insulation aging primarily affect the shape of the tanδ curve in the low-frequency region, without significantly altering the high-frequency region. Therefore, the frequency values ​​and integrals within this range can be used as degradation characteristics for analysis. As aging deepens, both the dielectric loss tangent and integral values ​​corresponding to the frequency point increase.

[0097] Therefore, the frequency f∈[10 -3 Hz,10 1 The dielectric loss tangent value tanδ(f) at [Hz] is taken as the frequency sensitive interval. The tanδ(f) within this frequency sensitive interval is segmentedly integrated, and D1 to D4 are named as characteristic parameters. The segmented integral formula is expressed as follows:

[0098]

[0099] Among them, D1, D2, D3, and D4 are characteristic parameters obtained by piecewise integration.

[0100] Furthermore, the real and imaginary parts of the complex dielectric constant also contain a wealth of aging factors. The complex dielectric constant parameter is shown below, which can be used to distinguish between the conductivity component and the relaxation component.

[0101]

[0102] Where ω is the angular velocity corresponding to the frequency of the applied voltage in the FDS test; and denote the conductivity component and relaxation component of the complex dielectric constant under pressure; σ0 denotes the static conductivity; ε s and ε ∞ represent the static dielectric constant and the optical frequency dielectric constant respectively; τ represents the relaxation time constant.

[0103] The conductivity component can be eliminated according to the Kramers-Kronig complex transform equation. Finally, the relaxed complex dielectric constant after eliminating the conductivity component can be decomposed into:

[0104]

[0105] Among them, ε' p (ω) and ε″ p (ω) are respectively However, due to the existence of multiple relaxation processes in the composite oil-paper insulation system and their mutually coupled nature, the real and imaginary parts of the complex dielectric constant are difficult to characterize the degree of aging. Therefore, the first-order differential of the real part of the complex dielectric constant is provided to highlight the relaxation characteristics.

[0106] First, perform coordinate transformation, let ω = e x , then differentiate the real part of the complex dielectric constant once, expressed as:

[0107]

[0108] It can be seen that the first-order differential of the real part of the complex dielectric constant is proportional to the square of the imaginary part of the relaxation complex dielectric constant after removing the conductivity component. Using this method, a dielectric spectrum with prominent relaxation characteristics can be obtained. Its peak point and integral are closely related to the degree of aging. Therefore, the peak point and integral are selected as the aging-relaxation related characteristic indicators, which are expressed as:

[0109]

[0110] Among them, D5 is the characteristic parameter characterizing the peak point, and D6 is the characteristic parameter characterizing the integral amount.

[0111] Step 3: Construction of the DP value prediction model dataset

[0112] Although the characteristic parameter D that characterizes the insulating state of the sample can be obtained by experimental means i (i = 1, 2, 3, 4, 5, 6), but its small sample data cannot be used to train the DP value prediction model. Therefore, in order to train this model, a large number of samples need to be prepared to obtain sufficient training and test data sets. However, due to the long sample preparation time, the amount of data required for model construction is insufficient. Therefore, the present invention uses nonlinear fitting to expand data density to balance the limited sample size. The specific method is as follows:

[0113] 301) Set the nonlinear function variables and let the six characteristic parameters D1 to D6 extracted based on the frequency domain dielectric spectrum curve correspond to functions y1 to y6 respectively.

[0114] 302) Determine the nonlinear function type for each characteristic parameter. Construct the nonlinear function by setting the function parameters a, b, c, and d. Use the goodness of fit to characterize the function's fit. Substitute the known independent variables into the solution to identify the function parameters. The nonlinear functions for each characteristic parameter are shown in Table 1.

[0115] Table 1 Nonlinear functions

[0116]

[0117]

[0118] 303) Based on Table 1, we construct an extended dataset φ A , as shown in formula (12):

[0119]

[0120] Among them, φ Aj ={y1(x 1j ),y2(x 2j ),y3(x 3j ),y4(x 4j ),y5(x5j ),y6(x 6j )}, represents the characteristic parameter obtained by the nonlinear function corresponding to the expanded j-th group of data; x ij represents the DP value of the i-th characteristic parameter of the j-th group of data, where i represents the i-th characteristic quantity, j represents the expanded j-th group of data, y(x ij ) represents the value of the characteristic parameter corresponding to the DP value of the i-th characteristic parameter of the j-th group of data.

[0121] Considering the impact of accuracy, 937 sets of data were expanded in the interval range [241,1178] with a DP value of 1 as the step size.

[0122] In summary, this step uses nonlinear function fitting to expand the aging characteristic database, and the goodness of each fitting model reaches an ideal value, which can provide a data basis for the subsequent polymerization degree DP value prediction model.

[0123] Step 4: Construction of BiLSTM network

[0124] Because RNNs (recurrent neural networks) often face problems like long-term dependencies, vanishing gradients, or exploding gradients when processing time series, researchers have proposed the LSTM (Long-Term Memory) to address these issues. However, LSTMs can only process forward information input into the neural network to obtain predictions. BiLSTMs, on the other hand, obtain predictions through both forward and backward information input into the neural network. BiLSTMs offer superior prediction results. A BiLSTM network is constructed based on the LSTM network.

[0125] like Figure 5 As shown, the LSTM network is composed of the input information x at time t t , cell state C t , temporary cell state Hidden state h t 、Forget Gate t 、Memory Gate i t and input gate o t The calculation process of the LSTM network can be summarized as follows: by forgetting the information in the cell state and memorizing new information, the information useful for subsequent calculations can be transmitted, while useless information is discarded, and the hidden state h is output at each time step. t , where forgetting, memory and output are determined by the hidden state h at the previous moment t-1 and the current input x t The calculated forget gate f t 、Memory Gate i t and output gate o t to control.

[0126] First, decide what information needs to be forgotten from the cell state. t-1 and the current input information x t The weighted sum plus the bias is processed and passed to the Sigmoid layer, so that it outputs a function f with a value between 0 and 1 t The closer the function value is to 0, the more information is forgotten, and the closer it is to 1, the more information is retained.

[0127] f t =σ(W t ·[h t-1 ,x t ]+b f ) (34)

[0128] Next, determine the new information that needs to be stored in the cell state. t-1 and the current input information x t Input into the Sigmoid layer and use the tanh layer to create a vector of new candidate values The two outputs are combined as the current output to update the cell state.

[0129] i t =σ(W t ·[h t-1 ,x t ]+b i ) (35)

[0130]

[0131] Finally, for the previous hidden state information h t-1 and the current input information x t Run a Sigmoid layer to determine the information that needs to be output in the cell state.

[0132] o t =σ(W o ·[h t-1 ,x t ]+b o ) (38)

[0133] The cell state is passed through the tanh layer (so that the output value is between -1 and 1) and multiplied by the output of the Sigmoid layer to combine the cell state and the new hidden state h t Pass to the next time step.

[0134] h t =o t tanh(C t ) (39)

[0135] The BiLSTM network introduces two LSTM networks (forward LSTM and backward LSTM) to simultaneously learn features from both the forward and backward directions of the sequence. The calculation process is as follows:

[0136]

[0137] Step 5: Construction of BiLSTM-based DP value prediction model, model training and feasibility verification

[0138] Through repeated experiments, the optimal parameter combination of the BiLSTM network, including the number of hidden layers, number of neurons, learning rate, and number of iterations, was determined, and a DP value prediction model was constructed.

[0139] In this embodiment, the optimal parameter combination is determined as shown in Table 2 below.

[0140] Table 2 Optimal parameter combination of BiLSTM network

[0141]

[0142] The data set obtained in step 3 is divided into a training set and a test set. The training set is used to train the constructed DP value prediction model. Finally, the test set is used to verify the accuracy of the DP value prediction model in predicting the DP value. The prediction results are shown in Table 3 and Figure 6 shown.

[0143] Table 3 Prediction results of the model for polymerization degree BP

[0144]

[0145] The present invention selects the integral of the FDS dielectric loss factor tanδ and the real and imaginary parts of the complex dielectric constant in different frequency bands as characteristic parameters; trains a BiLSTM model using a training set from expanded sample data; and verifies the degree of polymerization (DP) predicted by the trained BiLSTM model using a test set from the expanded sample data, ultimately establishing a reasonable BiLSTM-based DP value prediction model. Finally, the DP value is predicted using the obtained DP value prediction model. This DP value prediction model quantitatively predicts the DP value of unknown samples by learning the nonlinear mapping relationship between the characteristic parameters of the frequency-domain dielectric spectrum and the DP value of oil-paper insulation.

[0146] This embodiment also provides an oil-paper insulation polymerization degree DP value prediction system that integrates frequency domain dielectric spectrum characteristics and BiLSTM, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above method can be implemented.

[0147] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0148] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0149] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

Claims

1. A method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM, characterized in that: include: (1) Based on the FDS test platform, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees were measured; (2) Extracting the piecewise integral characteristics and the real and imaginary part characteristics of the complex dielectric constant based on the frequency domain dielectric spectrum curve; (3) Expand the sample size through nonlinear fitting and construct a data set for the DP value prediction model; (4) Construct a bidirectional long short-term memory network BiLSTM; (5) Construct a DP value prediction model based on the BiLSTM network; divide the data set into a training set and a test set, use the training set to train the model, and use the test set to verify the model's prediction accuracy of the DP value; (6) The obtained DP value prediction model is used to predict the DP value.

2. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1 is characterized in that: In step (1), the FDS test platform consists of a data acquisition computer, an FDS tester, a constant temperature box, and a three-electrode test device placed in the constant temperature box; Based on the FDS test platform, at a test temperature of 35°C, the frequency domain dielectric spectrum curves of DP oil-impregnated paper samples with different polymerization degrees were measured, including the real and imaginary part test curves of the complex dielectric constant and the dielectric loss tangent test curves.

3. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1 is characterized in that: In step (2), the method for extracting the segmented integral feature is: Select the frequency f∈[10 -3 Hz,10 1 The dielectric loss tangent value tanδ(f) at [Hz] is taken as the frequency sensitive interval, and tanδ(f) within this frequency sensitive interval is piecewise integrated. The piecewise integral formula is expressed as follows: D4=∫1 10 tanδ(f)df (4) Among them, D1, D2, D3, and D4 are characteristic parameters obtained by piecewise integration.

4. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1 is characterized in that: In step (2), the method for extracting the real and imaginary part characteristics of the complex dielectric constant is: The complex dielectric constant parameter is expressed as follows: Where ω is the angular velocity corresponding to the frequency of the applied voltage in the FDS test; and denote the conductivity component and relaxation component of the complex dielectric constant under pressure; σ0 denotes the static conductivity; ε s and ε ∞ represent the static dielectric constant and the optical frequency dielectric constant respectively; τ represents the relaxation time constant; The relaxation component after removing the conductivity component is: Among them, ε' p (ω) and ε″ p (ω) are respectively The real and imaginary parts of The relaxation characteristics are highlighted by the first-order differential of the real part of the complex dielectric constant. The specific method is: First, perform coordinate transformation, let ω = e x , then differentiate the real part of the complex dielectric constant once, expressed as: The peak point and integral amount are selected as the aging-relaxation related characteristic indicators, which are expressed as follows: Among them, D5 is the characteristic parameter characterizing the peak point, and D6 is the characteristic parameter characterizing the integral amount.

5. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1 is characterized in that: In step (3), the sample size is expanded by nonlinear fitting to construct a data set for the DP value prediction model. The implementation method is as follows: 301) Setting nonlinear function variables, and making the six characteristic parameters D1 to D6 extracted based on the frequency domain dielectric spectrum curve correspond to functions y1 to y6 respectively; 302) respectively determining the nonlinear function type of each characteristic parameter, constructing the nonlinear function by setting the function parameters, and using the goodness of fit to characterize the fitting effect of the function, substituting the known independent variables into the solution, and identifying the function parameters; 303) Construct an extended dataset φ based on the nonlinear function of the characteristic parameters D1~D6 A , as shown in formula (12): Among them, φ Aj ={y1(x 1j ),y2(x 2j ),y3(x 3j ),y4(x 4j ),y5(x 5j ),y6(x 6j )}, represents the characteristic parameter obtained by the nonlinear function corresponding to the expanded j-th group of data; x ij represents the DP value of the i-th characteristic parameter of the j-th group of data, where i represents the i-th characteristic quantity, j represents the expanded j-th group of data, y(x ij ) represents the value of the characteristic parameter corresponding to the DP value of the i-th characteristic parameter of the j-th group of data.

6. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 5, characterized in that: The nonlinear function of characteristic parameter D1 is: y1 = a + bcoscx + dsincx; where a = 0.02211, b = 0.006038, c = 0.003854, d = 0.01622, and the goodness of fit is 1.0000; The nonlinear function of characteristic parameter D2 is: y2 = a + bcoscx + dsincx; where a = 0.03898, b = 0.05704, c = 0.002259, d = 0.07015, and the goodness of fit is 1.0000; The nonlinear function of the characteristic parameter D3 is: Among them, a = 1.001, b = 469.1, c = 381.6, and the goodness of fit is 0.9988; The nonlinear function of the characteristic parameter D4 is: Among them, a = 7.899, b = 396.8, c = 282, and the goodness of fit is 0.9999; The nonlinear function of the characteristic parameter D5 is: y5 = ae bx +ce dx ; Among them, a = -0.1318, b = 0.001683, c = 6.13, d = -0.001353, goodness of fit 1.0000; The nonlinear function of the characteristic parameter D6 is: Among them, a = 69.26, b = 423.8, c = 377.3, and the goodness of fit is 0.9987; The dataset φ A 937 sets of data are expanded in the interval range [241,1178] with a DP value of 1 as a step size.

7. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1, characterized in that: In step (4), the BiLSTM network is constructed based on the LSTM network; The LSTM network is composed of the input information x at time t t , cell state C t , temporary cell state Hidden state h t 、Forget Gate t 、Memory Gate i t and input gate o t The calculation process of the LSTM network is as follows: by forgetting the information in the cell state and memorizing new information, the information useful for subsequent moment calculations can be transmitted, while the useless information is discarded, and the hidden state h is output at each time step. t , where forgetting, memory, and output are determined by the hidden state h at the previous moment t-1 and the current input x t The calculated forget gate f t 、Memory Gate i t and output gate o t To control; specifically: First, decide what information needs to be forgotten from the cell state; by t-1 and the current input information x t The weighted sum plus the bias is processed and passed to the Sigmoid layer, so that it outputs a function f with a value between 0 and 1 t ; f t =σ(W t ·[h t-1 ,x t ]+b f ) (13) Secondly, determine the new information that needs to be stored in the cell state; the previous hidden state information h t-1 and the current input information x t Input into the Sigmoid layer and use the tanh layer to create a vector of new candidate values Combine the two outputs as the current output to update the cell state; i t =σ(W t ·[h t-1 ,x t ]+b i ) (14) Finally, for the previous hidden state information h t-1 and the current input information x t Run a Sigmoid layer to determine the information that needs to be output in the cell state; the t =σ(W o ·[h t-1 ,x t ]+b o ) (17) The cell state is passed through the tanh layer and multiplied by the output of the sigmoid layer, thus combining the cell state and the new hidden state h t Pass to the next time step; h t =o t ·tanh(C t ) (18) The BiLSTM network introduces two LSTM networks, the forward LSTM and the backward LSTM, to simultaneously learn features from both the forward and backward directions of the sequence. The calculation process is as follows:

8. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1, characterized in that: In step (5), the optimal parameter combination of the BiLSTM network is determined and a DP value prediction model is constructed; The optimal parameter combination of the BiLSTM network is: 2 hidden layers, 20 neurons in each layer, a learning rate of 0.01, and 500 iterations.

9. The method for predicting the degree of polymerization (DP) value of oil-paper insulation by integrating frequency domain dielectric spectrum characteristics and BiLSTM according to claim 1, characterized in that: The polymerization degree DP value prediction model realizes quantitative prediction of the DP value of unknown samples by learning the nonlinear mapping relationship between the characteristic parameters of the frequency domain dielectric spectrum and the polymerization degree DP value of oil-paper insulation.

10. A system for predicting the degree of polymerization (DP) value of oil-paper insulation that integrates frequency domain dielectric spectrum characteristics and BiLSTM, characterized in that: The method comprises a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method according to any one of claims 1 to 9 can be implemented.

Images