Cable insulation life prediction method and system based on LSTM accelerated aging mapping

Abstract

The invention discloses a cable insulation life prediction method and system based on LSTM accelerated aging mapping, and relates to the technical field of submarine cable insulation life prediction. The existing method has the defects of insufficient multi-stress nonlinear modeling, laboratory and field data separation, difficulty in small sample modeling and the like, and the insulation life of the submarine cable is difficult to accurately predict. The method comprises the following steps: data acquisition: acquiring parameters of insulation electrical performance, physical and chemical performance and mechanical performance under an accelerated aging condition; carrying out data preprocessing: carrying out homodromous processing on the inverse indexes, improving box plot denoising, filling missing data with a space-time K nearest neighbor algorithm, and carrying out normalization; constructing an LSTM model: embedding a dielectric constant differential equation as a physical constraint, and introducing an index weight; and model training and evaluation: adopting five-fold cross validation, and quantizing prediction precision through mean square errors and decision coefficients. According to the technical scheme, the prediction precision is improved, the fault risk caused by insulation aging is reduced, the maintenance cost is reduced, and the reliability of the ocean energy transmission system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of submarine cable analysis, and particularly to a method and system for predicting the insulation life of cables based on LSTM accelerated aging mapping. Background Art

[0002] As the core infrastructure for ocean energy transmission, high-voltage submarine cables need to serve in a multi-physical field coupling environment of high voltage, large current, high humidity, high salinity and seawater pressure for a long time. The reliability of its insulation system directly determines the safe operation of the power grid, and the failures caused by insulation aging account for up to 68% (CIGRE report). Therefore, achieving accurate prediction of insulation life is crucial for submarine cable operation and maintenance decision-making.

[0003] At present, submarine cable operation and maintenance mainly rely on regular inspection and maintenance strategies (such as off-line dielectric loss test, partial discharge detection), but real-time life prediction cannot be achieved, and the maintenance cost is high. Methods for realizing real-time life prediction include the acceleration factor method and the experimental method, etc., but the above methods have the following problems: 1. Defects of traditional acceleration factor method: It relies on models based on linear assumptions such as the Arrhenius equation and inverse power law, and cannot characterize the non-linear aging under the synergistic action of multiple stresses; thermal aging models such as the Arrhenius equation only consider a single temperature stress, while in actual operation, electric field distortion (such as space charge accumulation) and mechanical vibration (ocean wave impact) will accelerate material degradation.

[0004] 2. Disconnection between laboratory and field data: The distribution of accelerated aging data and natural aging data is quite different, and there are significant differences between laboratory accelerated aging data and the real environment (IEEE Std1017 points out that the error of traditional accelerated tests is as high as 40 - 60%). Lack of an effective mapping mechanism, resulting in the evaluation results may not accurately reflect the actual aging behavior of materials.

[0005] 3. Too few actual data samples: The Weibull distribution requires a large amount of failure data to support, while the actual fault samples of high-voltage cables are not many (0.3 times per 100 kilometers per year); the acquisition period of natural aging data is long (10 - 20 years).

[0006] 4. Difficulties in multi-physical field coupling: Various stresses often exist simultaneously and are coupled with each other. For example, the presence of humidity will exacerbate the damage of the electric field to the insulation, etc. The effects of electric field distortion, thermal oxygen aging, mechanical fatigue, etc. are intertwined, and it is difficult to quantify a single factor. In actual situations, the stress effects are not linearly independent, and simple superposition may cause large errors. Summary of the Invention

[0007] The technical problem to be solved and the technical task proposed by the present invention are to improve and refine the existing technical solution, and provide a cable insulation life prediction method and system based on LSTM accelerated aging mapping, so as to achieve the purpose of real-time and accurate prediction of cable insulation life. To this end, the present invention adopts the following technical solutions.

[0008] A cable insulation life prediction method based on LSTM accelerated aging mapping, comprising the following steps: 1) Data collection: Under the set accelerated aging experiment conditions, collect key performance parameters at different aging time points, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; 2) Data preprocessing: Perform co-directional processing on inverse indicators; Use the improved box plot method for outlier detection, dynamically adjust the IQR coefficient, identify and remove outliers; Fill in missing data based on the spatio-temporal K-nearest neighbor algorithm; Compress the data to the [0, 1] interval through min-max normalization; 3) LSTM model construction: The input layer receives the preprocessed time series data; Embed the dielectric constant differential equation as a soft constraint into the loss function; The loss function introduces exponential weights to make the model pay more attention to the characteristics of sudden changes in the later stage of aging; The output layer predicts the percentage of the remaining life of the cable; 4) Model training and optimization: Divide the data into a training set and a test set, and use 5-fold cross-validation to optimize the hyperparameters; Monitor the mean squared error of the validation set to prevent overfitting; Obtain a trained LSTM model; 5) Life prediction and evaluation: Use the trained model to predict the remaining life of the cable, and evaluate the prediction accuracy through the mean squared error and the coefficient of determination.

[0009] This technical solution collects three types of parameters, namely electrical, physical and chemical, and mechanical, to quantify the synergistic degradation effects of electric field, temperature, and mechanical stress, overcome the one-sidedness of traditional single parameters, and provide comprehensive input for nonlinear modeling. This technical solution adopts an adaptive data preprocessing mechanism, which makes the inverse indicators unidirectional, thereby unifying the direction of aging trends, improving the box plot method to dynamically adjust the IQR to adapt to non-Gaussian distribution data, using time-space correlation to repair missing values, and normalizing to eliminate dimensional differences. IQR (Interquartile Range) refers to the interquartile range. Anomaly detection and missing value filling solve the noise interference of small sample data (such as experimental errors and local data missing), which is conducive to improving data quality and making subsequent LSTM modeling more stable and reliable. Using a LSTM model enhanced by physical constraints, physical equation constraints improve the interpretability of the model, avoid pure data-driven overfitting, and solve the problem of multi-stress nonlinear coupling; dynamic weights enable the model to focus on the critical stage of aging, sensitively capture life mutations, and improve the timeliness of prediction. 5-fold cross-validation is used to optimize hyperparameters and make full use of limited data. The validation set MSE is monitored and stopped early (terminated if there is no continuous improvement) to suppress overfitting. The lack of generalization ability caused by the difference in laboratory-field data distribution is resolved to ensure that the model maintains high accuracy under small sample conditions.

[0010] As a preferred technical means: in step 1), the insulation electrical performance parameters include dielectric constant; the physical and chemical performance parameters include carbonyl index, crystallinity, thermal decomposition temperature and melting temperature; the mechanical performance parameters include elastic modulus; the dielectric constant is collected by a broadband dielectric spectrometer; the crystallinity, thermal decomposition temperature and melting temperature are collected by a synchronous thermal analyzer; the carbonyl index is collected by scanning with an attenuated total reflection infrared spectrometer.

[0011] The dielectric response at different frequencies is collected by a broadband dielectric spectrometer, which directly reflects the internal polarization behavior of the insulating material, sensitively captures the microstructural changes caused by early aging, and provides frequency domain characteristics for dielectric loss modeling. The elastic modulus quantifies the material's ability to resist elastic deformation, and combines the tensile strength to construct a "stiffness-strength" dual-dimensional evaluation system, accurately characterizing the degree of molecular chain damage under mechanical stress (such as wave impact), and directly related to the structural reliability of submarine cable laying and operation. The synchronous thermal analyzer can obtain crystallinity, melting temperature and thermal decomposition temperature through a single test, establish a physical association of "crystallization state change → thermal stability degradation", avoid sample differences in step-by-step testing and experimental errors caused by multiple transfers, and improve the accuracy and efficiency of aging mechanism analysis. The attenuated total reflection infrared spectrometer directly reflects the thermal oxidation aging process, correlates the insulation degradation rate, non-destructively detects chemical aging, avoids slicing damage, and can continuously track the oxidation degree of the same sample at multiple time points.

[0012] As a preferred technical means: the inverse index adopts the following formula to be unidirectional: ; In the formula, is the original inverse index value, is the positive index value after co-directionalization; The outlier determination condition of the improved box plot method is: if the data point satisfies X < Q1 - 1.5×IQR or X > Q3 + 1.5×IQR, then the data is considered an outlier, where ; Q1 is the first quartile and Q3 is the third quartile; The formula for filling missing data based on the spatio-temporal K-nearest neighbor algorithm is: ; In the formula, z i is the spatial position coordinate of the i-th neighbor, z t is the spatial position coordinate of the target point, σ is the spatial decay coefficient, k is the number of nearest neighbors, is the observed value of the (t - i)-th point in the time dimension; The normalization formula is:

[0013] where X min and X max are the minimum and maximum values in the dataset respectively.

[0014] Preprocessing provides high-consistency, low-noise, complete and regular input data for the subsequent prediction model, and is the cornerstone for ensuring the accuracy of life prediction. Converting the inverse index to a positive index eliminates the trend conflict of multiple parameters; improves the model convergence efficiency: avoids the training oscillation caused by the chaotic direction of the input signal of LSTM and accelerates the convergence; retains the physical meaning, and the converted is still monotonically negatively correlated with the material degradation degree and does not distort the original aging law. The traditional box plot assumes a normal distribution, while the aging data is mostly skewed (such as Weibull distribution). The dynamic IQR coefficient can flexibly adjust the threshold boundary; is beneficial to improving the rejection rate of experimental noise; at the same time, avoids misdeleting the mutated data in the later stage of aging. The spatio-temporal K-nearest neighbor missing value filling integrates the spatio-temporal two-dimensional correlation, avoids data missing caused by local sensor failure, and integrates the spatial position and time continuity. Normalization eliminates the dimensional difference and avoids the model bias towards large numerical parameters caused by the order of magnitude difference; accelerates the training of LSTM and stabilizes the gradient update.

[0015] As an optimal technical means: in step 3), the architecture of the LSTM network model adopts a bidirectional LSTM layer, and the bidirectional structure jointly models through forward and backward propagation; embedding the dielectric constant differential equation as a soft constraint into the loss function:

[0016] In the formula, γ is the dielectric constant; is the activation energy; R is the ideal gas constant; T is the absolute temperature; is the electric field strength; n is the material constant; is the correction term for LSTM output.

[0017] The bidirectional LSTM layer combines forward-backward propagation modeling, learns historical aging laws along the time sequence, and captures future degradation signs in reverse time sequence, which is conducive to capturing aging inflection points and improving the sensitivity of identifying the sudden change stage of insulation life; reducing the underreporting rate of key nodes by joint modeling, and reducing the correlation modeling error between the slow change period and the drastic change period of the dielectric constant. The physical equation of dielectric constant change is embedded in the loss function as a regular term, fitting the nonlinear influence of electric field distortion (spatial charge accumulation), coupling temperature and electrochemical degradation, and learning the synergistic effect of vibration-temperature-electric field; the physical term dominates the macroscopic change trend of dielectric constant, and the correction term focuses on compensating for local nonlinear disturbances, so that the generalization error of the model is reduced under small samples. Through the fusion of physical constraints and data-driven, it not only retains the advantages of LSTM in handling complex nonlinearities, but also embeds the explainable physical mechanism of dielectric constant change, fundamentally solving the industry problem of multi-stress coupling modeling.

[0018] As a preferred technical means: the exponential weight introduced in the loss function is: ; In the formula, α is the weight adjustment coefficient, t is the current time point, and T is the total aging time.

[0019] Sudden changes in performance parameters (such as a sudden drop in crystallinity and a surge in dielectric constant) in the late stage of aging (such as when approaching the end of life) are crucial to operation and maintenance decisions. The exponential weight increases exponentially with time t, so that the model automatically gives higher weight to the later data during training, and prioritizes learning the nonlinear degradation characteristics of the late stage of aging, avoiding the early data dominating the training and ignoring the key mutation signals. Through exponential weighting, the model can optimize the prediction ability of the high-risk stage in a targeted manner and reduce the risk of sudden failures caused by prediction lags. Through the nonlinear weighting mechanism, the exponential weight forces the model to pay more attention to the steep decline section (such as exponential decay or step change) in the "time-performance" curve, so as to more accurately fit the accelerated aging law under multi-stress coupling (such as the sudden drop in performance caused by the synergistic effect of electric field and temperature). Exponential weighting can guide the model to explore such high-order features and improve the modeling ability of "non-stationary time series data" (such as performance fluctuations under time-varying stress). The exponential weight mechanism achieves automatic focus on key information through the "time-importance" exponential mapping without relying on additional labeled data. It solves the inherent defect of the traditional model of "treating all data equally" in a low-cost computing way, making the deep learning model more in line with the physical law of submarine cable insulation aging "slow in the early stage and sudden change in the later stage".

[0020] As a preferred technical measure: when monitoring the mean square error of the validation set in step 4), terminate the training if there is no improvement for multiple consecutive epochs.

[0021] When the model continuously reduces the loss on the training set but the MSE on the validation set no longer improves, it indicates that the model begins to learn the noise of the training data or the random fluctuations of specific samples (i.e., overfitting). The early stopping mechanism monitors the performance of the validation set and forces the termination of training before overfitting occurs, ensuring that the model retains the prediction ability for unseen data (such as on-site real-time monitoring data). By the early stopping method, the model terminates training after the MSE on the validation set stabilizes, which can better adapt to the differences in data distributions in multiple scenarios. Combining with the "5-fold cross-validation" strategy, the early stopping method can independently optimize the training termination point in each cross-validation fold, avoiding the bias in hyperparameter evaluation caused by a fixed number of epochs. For example, the data sets of different folds may have different convergence speeds due to differences in aging samples. The early stopping method can dynamically determine the optimal training duration for each group of data, improving the accuracy of hyperparameter optimization.

[0022] As a preferred technical means: in step 5), the mean square error formula is:

[0023] The coefficient of determination formula is:

[0024] In the formula, N is the number of test samples, is the actual life value, is the predicted life value, is the mean of the actual life values.

[0025] The mean square error formula and the coefficient of determination formula provide quantifiable risk indicators, converting "life prediction accuracy" into computable, comparable, and traceable quantitative indicators with industry-standard mathematical tools. This not only meets the technical verification requirements of the deep learning model but also fits the actual requirements of submarine cable operation and maintenance for "result reliability and decision-making scientificity", serving as a key bridge connecting algorithm research and development with engineering applications. The dual-indicator joint locates model defects, drives algorithm iteration, provides a quantifiable, diagnosable, and iterative model verification tool, and provides a closed-loop optimization basis for a high-reliability cable life prediction system.

[0026] Another technical solution of the present invention is to provide a cable insulation life prediction system adopting a cable insulation life prediction method based on LSTM accelerated aging mapping. The cable insulation life prediction system includes: A data acquisition module, which is used to collect key performance parameters at different aging time points under set accelerated aging experimental conditions, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; the insulation electrical performance parameters include dielectric constant; the physical and chemical performance parameters include crystallinity, melting temperature, carbonyl index, and thermal decomposition temperature; the mechanical performance parameters include elastic modulus; Data preprocessing module, including: A homogenization unit for homogenizing inverse indicators including carbonyl index and dielectric constant; An outlier detection unit for detecting outliers using the improved boxplot method, dynamically adjusting the IQR coefficient, identifying and removing outliers; A missing value filling unit for filling missing data based on the spatio-temporal K-nearest neighbor algorithm; A normalization unit for compressing data into the interval [0, 1]; LSTM prediction module, including: A bidirectional LSTM layer that receives the preprocessed time series data; A physical constraint layer that embeds the dielectric constant differential equation in the loss function; A loss function optimization unit that introduces exponential weights to make the model pay more attention to the characteristics of mutations in the later stage of aging; an output layer that predicts the percentage of the remaining life of the cable; A model training and optimization module that divides the data into a training set and a test set, optimizes the hyperparameters using 5-fold cross-validation; monitors the mean squared error of the validation set to prevent overfitting; A remaining life prediction and evaluation module; uses the trained model to predict the remaining life of the cable and evaluates the prediction accuracy through the mean squared error and the coefficient of determination.

[0027] From data acquisition, preprocessing to model construction and evaluation, it systematically addresses the deficiencies of traditional cable life prediction methods, especially having significant advantages in multi-physical field modeling, small sample processing, and engineering practicality, providing reliable technical support for the accurate assessment of the insulation status of high-voltage cables. Specifically, it simultaneously collects electrical properties, physical and chemical properties, and mechanical properties to comprehensively capture the multi-physical field coupling effect during the insulation aging process, providing richer characteristic information for life prediction. The set accelerated aging experimental conditions can be high temperature, high pressure, etc. accelerated conditions, shortening the data acquisition cycle, solving the problem of long time-consuming for obtaining natural aging data, and at the same time maintaining the physical relevance with the actual aging process. Conduct mathematical transformation on the inverse index to unify the index change direction and avoid model training deviation caused by differences in the physical meaning of the indexes. The improved box plot method can more accurately identify outliers under non-Gaussian distribution by adaptively adjusting the IQR coefficient, especially suitable for data characteristics with sudden performance changes in the later stage of aging. Fill in missing values by combining time series and spatial position information, making full use of the spatio-temporal correlation of the data, and being more in line with the physical laws of cable aging than traditional interpolation methods. Bidirectional LSTM layer: Simultaneously capture the forward and backward dependencies of time series data, effectively model the non-linear degradation characteristics (such as performance mutation points) during the aging process, and have stronger time series modeling ability than unidirectional LSTM. Incorporate the dielectric constant differential equation as physical prior knowledge into the loss function, making the model prediction results conform to the physical laws of material aging, enhancing the interpretability and engineering credibility of the model. Increase the training weight of later-stage data through the weight factor, force the model to focus on learning the mutation characteristics in the later stage of aging, and significantly improve the prediction accuracy of the life end point. Verify the model stability through multiple rounds of data partitioning, reduce the overfitting risk caused by the randomness of data partitioning, and ensure the generalization ability of the model. Real-time monitor the mean square error of the validation set, automatically terminate the training without performance improvement, avoid overtraining, and improve the model training efficiency and prediction reliability. Combine the mean square error and the coefficient of determination to comprehensively quantify the model accuracy from two dimensions of absolute error and relative fitting degree, providing a clear goal for model optimization. Through physical constraints and data preprocessing, the model can adapt to complex aging environments with multi-factor coupling such as electric field, temperature, and mechanical stress, especially suitable for harsh working conditions such as high-voltage submarine cables. The spatio-temporal K-nearest neighbor algorithm and the exponential weighting mechanism effectively alleviate the problem of scarce fault samples of high-voltage cables, reduce the dependence on large-scale failure data, and are more in line with engineering reality. Based on the time series prediction characteristics of LSTM, the system can process online monitoring data in real time, dynamically update the remaining life prediction, and support the timely adjustment of operation and maintenance decisions.

[0028] As an optimal technical means: The data acquisition module includes: The dielectric constant acquisition unit is collected by a broadband dielectric spectrometer. After multiple effective tests are completed for each group of samples, statistical distribution methods are used for data processing; The crystallinity, thermal decomposition temperature, and melting temperature acquisition unit collects data through a synchronous thermal analyzer. The sample is heated at a constant rate under a protective atmosphere. The thermal decomposition temperature is analyzed by obtaining the thermogravimetric curves at different aging times, and the crystallinity and melting temperature are analyzed by switching the atmosphere in stages. The carbonyl index acquisition unit collects data through an attenuated total reflection infrared spectrometer. By setting the reflection crystal and the incident angle, the spectral data is scanned and calculated. The elastic modulus acquisition unit collects data through a universal material testing machine. The universal material testing machine stretches the sample and records the maximum tensile force and the change in the marked distance at break.

[0029] After each group of samples completes multiple valid tests, two-parameter Weibull distribution or normal distribution statistical analysis is used to reduce the influence of accidental errors, so that the dielectric constant data can reflect the probability distribution characteristics of the overall dielectric properties of the material and meet the requirements of reliability assessment. During the physical simulation of the protective atmosphere and staged heating, argon can be used for protection in the initial stage to simulate the anaerobic environment during submarine cable laying, accurately measure the crystallinity and melting temperature, and avoid interference from oxidation on the crystalline state; in the later stage, the atmosphere can be switched to oxygen to simulate aerobic thermal oxygen aging during service. The thermal decomposition temperature is obtained through the thermogravimetric curve to quantify the impact of oxidative degradation on thermal stability. The staged atmosphere control of the synchronous thermal analyzer and the frequency scanning of the broadband dielectric spectroscopy can simulate the multi-stage aging process during actual service, enabling the accelerated experimental data to map the natural aging law through physical mechanisms and solving the problem of "data distortion" in traditional accelerated tests. The non-destructive detection of the carbonyl index directly obtains the infrared spectrum of the oxide layer on the material surface without the need for slicing and sample preparation, and can continuously track the same sample at different aging time points, avoiding the destruction of the aging gradient structure by traditional sample preparation.

[0030] As a preferred technical means: The physical constraint layer is embedded in the dielectric constant differential equation as:

[0031] In the formula, γ is the dielectric constant; is the activation energy; R is the ideal gas constant; T is the absolute temperature; is the electric field strength; n is the material constant; is the correction term output by LSTM.

[0032] Beneficial effects: 1. Adopting a life prediction method based on the fusion of LSTM and physical constraints, enhancing the model interpretability by embedding the dielectric constant differential equation, effectively solving the problem of multi-stress non-linear coupling, and significantly improving the prediction accuracy.

[0033] 2. Introducing an exponentially weighted loss function to strengthen the data weight in the later stage of aging, significantly enhancing the prediction sensitivity of the model to the life mutation stage, and greatly reducing the prediction error compared with traditional models.

[0034] 3. Based on the mapping relationship of the traditional Arrhenius equation, introduce the correction term output by LSTM , compensate for the influence of the multi-stress coupling effect, and improve the accuracy of life assessment.

[0035] 4. The missing value filling algorithm based on spatio-temporal proximity, combined with the dynamic weight adjustment mechanism, effectively improves the robustness and reliability of small sample data modeling.

[0036] 5. By defining the reciprocal conversion method and normalization process of the inverse index, eliminate the dimensional difference of multi-physical field data, and ensure the consistency and stability of model input.

[0037] 6. The overall architecture design of model lightweight, edge deployment and real-time prediction system, improves the engineering implementation efficiency and long-term operation and maintenance adaptability. Description of the Drawings

[0038] Figure 1 is the flow chart of the present invention. Detailed Description of the Invention

[0039] The technical solutions of the present invention will be further described in detail below in conjunction with the drawings in the specification.

[0040] Example 1: As Figure 1 shown, the present invention includes the steps: S1: Data acquisition: Under the set accelerated aging experiment conditions, collect key performance parameters at different aging time points, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; S2: Data preprocessing: Perform co-directional processing on the inverse index; Use the improved box plot method for outlier detection, dynamically adjust the IQR coefficient, identify and remove outliers; Fill in the missing data based on the spatio-temporal K-nearest neighbor algorithm; Compress the data to the [0, 1] interval through min-max normalization; S3: LSTM model construction: The input layer receives the preprocessed time series data; Embed the dielectric constant differential equation as a soft constraint into the loss function; The loss function introduces exponential weights to make the model pay more attention to the characteristics of sudden changes in the later stage of aging; The output layer predicts the percentage of the remaining life of the cable; S4: Model training and optimization: Divide the data into a training set and a test set, and use 5-fold cross-validation to optimize the hyperparameters; Monitor the mean square error of the validation set to prevent overfitting; Obtain the trained LSTM model; S5: Life prediction and evaluation: Use the trained model to predict the remaining life of the cable, and evaluate the prediction accuracy through the mean square error and the coefficient of determination.

[0041] In this embodiment, cross-linked polyethylene (XLPE) cables are selected as the object for data collection, and accelerated aging test conditions (such as a high temperature environment of 110° C.) are set.

[0042] Among them, elastic modulus is one of the key properties of polymer materials, which directly affects their reasonable application. The material was tested using the Zwick / Roell Z020 universal material testing machine produced by the German Zwick company. After the dumbbell specimen was placed in the fixture of the testing machine, the chuck moved at a speed of 250±50mm / min. The maximum tensile force was measured and recorded during the test. At the same time, the distance between the two marking lines was measured on the same specimen when it broke.

[0043] The test reference standards IEC 60247-2014 and ASTM D924-15 were used. The NovocontrolConcept 80 broadband and wide-temperature dielectric spectrometer produced in Germany was used to test the dielectric loss of each XLPE. Before the test, the surface of the sample was wiped with anhydrous ethanol. After the sample was dried, its thickness was measured. Then, the two sides of the sample were ion sputtered with gold. The sputtering time was 60s and the current was 20mA. The gold-sputtered sample was placed between two copper electrodes with diameters of 20mm and 30mm respectively. The sample was fixed in the test device for measurement. The dielectric loss test can set the test temperature and the test frequency range is 1~10 6 Hz.

[0044] The chemical structure of the exposed surface of the sample was characterized by attenuated total reflection infrared spectrometer, with ZnSe as the reflective crystal, an incident angle of 45°, 32 scans, and a scanning range of 700-4000 cm -1 , resolution 4cm -1 , to calculate the carbonyl index.

[0045] The crystallinity, melting temperature and thermal decomposition temperature of the sample were analyzed using a TGA / DSC3+ synchronous thermal analyzer. Under argon protection, the sample was heated from room temperature 30°C to 200°C at a rate of 10°C / min, then replaced with an oxygen atmosphere and continued to be heated to 250°C at the same rate. The synchronous thermal analyzer combines DSC and TGA to obtain the heat flow change and weight change curves of the sample in one test.

[0046] In step 2 of the method provided by the present invention, the elastic modulus, dielectric constant (50 Hz), carbonyl index, crystallinity, thermal decomposition temperature and melting temperature in the above step 1 are set as variables X1, X2, X3, X4, X5, X6 that change with aging time.

[0047] Since the dielectric constant (X2) and carbonyl index (X3) are inverse indicators in the multi-physics field aging process of submarine cables, they need to be converted into the same direction using the following formula: .

[0048] Outlier detection is performed using the improved box plot method, dynamically adjusting the IQR coefficient to adapt to non-Gaussian distributed data. First, calculate the first quartile (Q1) and the third quartile (Q3) of the data:

[0049] If the data point X < Q1 - 1.5 × IQR or X > Q3 + 1.5 × IQR, then the data is considered an outlier.

[0050] Meanwhile, based on the spatio-temporal K-Nearest Neighbors (KNN) algorithm, fill the missing data with the mean of the nearest K observations:

[0051] z i is the position coordinate, and σ = 0.1 is the spatial decay coefficient (axial position of the cable).

[0052] Finally, perform normalization. To eliminate the influence of different dimensions and make all eigenvalue within the same range, adopt the min-max normalization, as shown in the following formula:

[0053] where X min and X max are the minimum and maximum values in the dataset respectively; after normalization, all data is compressed into the interval [0, 1], improving the stability of model training.

[0054] In step 3 of the method provided by the present invention, the architecture of the LSTM network model adopts a bidirectional LSTM layer. The bidirectional structure jointly models through forward and backward propagation, enhancing the sensitivity to the aging inflection point. Embed the dielectric constant differential equation as a soft constraint into the loss function:

[0055] where, γ: dielectric constant; Ea: activation energy; R: ideal gas constant; T: absolute temperature; : electric field strength; : the correction term output by LSTM, used to compensate for the deficiencies of the traditional model.

[0056] The data closer to the end of life is more critical for operation and maintenance decisions. Therefore, an exponential weight is introduced into the loss function to make the model pay more attention to the characteristics of sudden changes in the later stage of aging:

[0057] In step 4 of the method provided by the present invention, the preprocessed data is divided into an 80% training set and a 20% test set, and 5-fold cross-validation is used to optimize the hyperparameters. Monitor the mean squared error (MSE) of the validation set. If no lower value appears within 10 consecutive epochs, terminate the training to prevent overfitting.

[0058] In step 5 of the method provided by the present invention, evaluate the training accuracy of the model. The metrics include calculating the mean squared error (MSE) and the coefficient of determination (R 2 ), and the formulas are as follows:

[0059] where is the actual aging data, is the model prediction result.

[0060] The following uses specific test data and test curve results to prove the effect of the present invention. According to the cable insulation life prediction method based on LSTM accelerated aging mapping provided by the present invention, a multi-physical field accelerated aging experiment is carried out on the high-voltage submarine cable insulation material, and the changes in aging time and key performance parameters are recorded. The original data is shown in Table 1.

[0061] Table 1 Original data of key performance parameters

[0062] In step 2 of the method provided by the present invention, in order to eliminate the influence of dimensions in the original data, inverse index co-directionalization is carried out. Since X2 (dielectric constant) and X3 (carbonyl index) are inverse indices, reciprocal transformation is used. At the same time, in order to avoid the risk of the denominator being zero, a very small amount is introduced to solve numerical calculation anomalies:

[0063] Then, standardization processing is carried out: use min-max normalization to compress the data into the [0, 1] interval, and the results are shown in Table 2.

[0064] Table 2 Data after standardization processing

[0065] Select the parameters with the highest contribution to life prediction according to their standardized change rates. The main parameters are X3 (carbonyl index), X4 (crystallinity), and X5 (thermal decomposition temperature); the backup parameter is X1 (elastic modulus). Then construct an LSTM network structure: the input layer has 3 time steps (24h window) and 3 feature numbers (X3, X4, X5); the bidirectional LSTM layer has 64 neurons, and the activation function is tanh; the physical constraint layer is embedded with the dielectric constant differential equation constraint; the output layer predicts the remaining life percentage in the next 12h.

[0066] During training, the data is divided into the first 6 groups for training and the last 2 groups for testing. Terminate if the validation loss does not decrease for 10 consecutive epochs. The prediction results of the LSTM model are shown in Table 3: Table 3 Calculated life prediction results

[0067] In step 5 of the method provided by the present invention, calculate the mean square error and the coefficient of determination to evaluate the model accuracy, and compare it with the results obtained by traditional linear regression, and compare the RMSE and R of the two 2 , and the specific evaluation results are shown in Table 4: Table 4 Evaluation results

[0068] The smaller the RMSE root mean square error, the better, and the closer R 2 is to 1, the better. It can be seen that the prediction results are relatively accurate and better than the traditional linear regression method.

[0069] Example 2: A cable insulation life prediction system based on LSTM accelerated aging mapping, comprising: I. A data acquisition module, which is used to collect key performance parameters at different aging time points under the set accelerated aging experiment conditions, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; the insulation electrical performance parameters include dielectric constant; the physical and chemical performance parameters include crystallinity, melting temperature, carbonyl index, and thermal decomposition temperature; the mechanical performance parameters include elastic modulus; The data acquisition module includes: A dielectric constant acquisition unit, which places the sputtered sample between two copper electrodes with diameters of 20mm and 30mm respectively, fixes the sample in the test device for measurement; A crystallinity and melting temperature acquisition unit, where the sample is heated at a constant rate under a protective atmosphere and analyzed by switching the atmosphere in stages; A carbonyl index acquisition unit, which is measured by an attenuated total reflection infrared spectrometer, and calculates the spectral data by scanning using a set reflection crystal and incident angle; The thermal decomposition temperature acquisition unit heats the sample at a constant rate under a protective atmosphere to obtain thermogravimetric curves at different aging times; The elastic modulus acquisition unit uses a universal material testing machine to stretch the specimen and records the maximum tensile force and the change in the marked distance at fracture.

[0070] II. The data preprocessing module includes: The same-direction transformation unit is used to perform the same-direction transformation on the inverse indicators including the dielectric constant and the carbonyl index; The outlier detection unit is used to detect outliers using the improved box plot method, dynamically adjust the IQR coefficient, and identify and remove outliers; The missing value filling unit is used to fill in the missing data based on the spatio-temporal K-nearest neighbor algorithm; The normalization unit is used to compress the data into the interval [0, 1]; III. The LSTM prediction module includes: The bidirectional LSTM layer receives the preprocessed time series data; The physical constraint layer embeds the dielectric constant differential equation in the loss function; The loss function optimization unit introduces exponential weights to make the model pay more attention to the characteristics of sudden changes in the later stage of aging; the output layer predicts the percentage of the remaining life of the cable; IV. The model training and optimization module divides the data into a training set and a test set, and uses 5-fold cross-validation to optimize the hyperparameters; monitors the mean squared error of the validation set to prevent overfitting; V. The life prediction and evaluation module; uses the trained model to predict the remaining life of the cable, and evaluates the prediction accuracy through the mean squared error and the coefficient of determination.

[0071] This embodiment systematically solves the deficiencies of traditional cable life prediction methods from data acquisition, preprocessing to model construction and evaluation, and has significant advantages especially in multi-physical field modeling, small sample processing and engineering practicability, providing reliable technical support for the accurate evaluation of the insulation state of high-voltage cables.

[0072] It can be understood that the detailed functional implementation of the above modules can be referred to the introduction in the foregoing method embodiment, and no other elaboration will be made here.

[0073] The above-described cable insulation life prediction method and system based on LSTM accelerated aging mapping are specific embodiments of the present invention, which have already reflected the substantial characteristics and progress of the present invention. According to actual usage needs, under the inspiration of the present invention, equivalent modifications can be made to it, and all are within the protection scope of this solution.

Claims

1. A method for predicting the insulation life of a cable based on LSTM accelerated aging mapping, characterized in that It includes the following steps: 1) Data acquisition: Under the set accelerated aging test conditions, key performance parameters at different aging time points are collected, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; 2) Data preprocessing: Perform co-directional processing on inverse indicators; Use the improved box plot method for outlier detection, dynamically adjust the IQR coefficient, identify and remove outliers; Fill in missing data based on the spatio-temporal K-nearest neighbor algorithm; Compress the data to the [0, 1] interval through min-max normalization; 3) LSTM model construction: The input layer receives the preprocessed time series data; Embed the dielectric constant differential equation as a soft constraint into the loss function; The loss function introduces exponential weights to make the model pay more attention to the characteristics of sudden changes in the later stage of aging; The output layer predicts the percentage of the remaining life of the cable; 4) Model training and optimization: Divide the data into a training set and a test set, and use 5-fold cross-validation to optimize hyperparameters; Monitor the mean squared error of the validation set to prevent overfitting; Obtain the trained LSTM model; 5) Life prediction and evaluation: Use the trained model to predict the remaining life of the cable, and evaluate the prediction accuracy through the mean squared error and the coefficient of determination.

2. A method for predicting the insulation life of a cable based on LSTM accelerated aging mapping according to claim 1, characterized in that: In step 1), the insulation electrical performance parameters include the dielectric constant; The physical and chemical performance parameters include the carbonyl index, crystallinity, thermal decomposition temperature, and melting temperature; The mechanical performance parameters include the elastic modulus; The dielectric constant is collected by a broadband dielectric spectrometer; The crystallinity, thermal decomposition temperature, and melting temperature are collected by a synchronous thermal analyzer; The carbonyl index is obtained by scanning with an attenuated total reflection infrared spectrometer.

3. A method for predicting the insulation life of a cable based on LSTM accelerated aging mapping according to claim 1, characterized in that: The inverse indicator is converted into the same direction using the following formula: , where is the original inverse index value, is the positive indicator value after homogenization; The outlier determination condition of the improved box plot method is: if a data point satisfies X < Q1 - 1.5 × IQR or X > Q3 + 1.5 × IQR, then the data is considered an outlier, where ; Q1 is the first quartile and Q3 is the third quartile; The formula for filling in missing data based on the spatio-temporal K-nearest neighbor algorithm is: ； where z i is the spatial position coordinate of the i-th neighbor, and z t is the spatial position coordinate of the target point, σ is the spatial attenuation coefficient, k is the number of neighbors, is the observation value of the (t - i)-th point in the time dimension; The normalization formula is: where X min and X max are the minimum and maximum values in the dataset, respectively.

4. A method for predicting the insulation life of a cable based on LSTM accelerated aging mapping according to claim 1, characterized in that: In step 3), the architecture of the LSTM network model uses a bidirectional LSTM layer, and the bidirectional structure jointly models through forward and backward propagation; Embed the dielectric constant differential equation as a soft constraint into the loss function: Where γ is the dielectric constant; is the activation energy; R is the ideal gas constant; T is the absolute temperature; is the electric field strength; n is the material constant; is the correction term output by LSTM.

5. The cable insulation life prediction method based on LSTM accelerated aging mapping according to claim 4, wherein: The exponential weight introduced in the loss function is as follows: ; In the formula, α is the weight adjustment coefficient, t is the current time point, and T is the total aging time.

6. The cable insulation life prediction method based on LSTM accelerated aging mapping according to claim 1, characterized in that: When monitoring the mean squared error of the validation set in step 4), if there is no improvement for multiple consecutive epochs, the training is terminated.

7. A method for predicting the insulation life of a cable based on LSTM accelerated aging mapping according to claim 1, characterized in that: In step 5), the mean squared error formula is: The coefficient of determination formula is: where N is the number of test samples, is the actual life value, is the predicted life value, is the mean value of the actual life values.

8. A cable insulation life prediction system adopting the cable insulation life prediction method based on LSTM accelerated aging mapping described in claim 1, characterized in that It includes: A data acquisition module, which is used to collect key performance parameters at different aging time points under the set accelerated aging test conditions, including insulation electrical performance parameters, physical and chemical performance parameters, and mechanical performance parameters; The insulation electrical performance parameters include the dielectric constant; The physical and chemical performance parameters include the carbonyl index, crystallinity, thermal decomposition temperature, and melting temperature; The mechanical performance parameters include the elastic modulus; A data preprocessing module, including: A co-directional unit, which is used to perform co-directional processing on inverse indicators including the dielectric constant and the carbonyl index; An outlier detection unit, which is used to detect outliers by using the improved box plot method, dynamically adjust the IQR coefficient, identify and remove outliers; A missing value filling unit, which is used to fill in missing data based on the spatio-temporal K-nearest neighbor algorithm; A normalization unit, which is used to compress the data to the [0, 1] interval; An LSTM prediction module, including: A bidirectional LSTM layer, which receives the preprocessed time series data; A physical constraint layer, which embeds the dielectric constant differential equation in the loss function; The loss function optimization unit introduces exponential weights to make the model pay more attention to the characteristics of mutations in the later stage of aging; the output layer predicts the percentage of the remaining life of the cable. The model training and optimization module divides the data into a training set and a test set, and uses 5-fold cross-validation to optimize the hyperparameters; monitors the mean square error of the validation set to prevent overfitting. The life prediction and evaluation module uses the trained model to predict the remaining life of the cable, and evaluates the prediction accuracy through the mean square error and the coefficient of determination.

9. The cable insulation life prediction system according to claim 8, characterized in that: The data acquisition module includes: The dielectric constant acquisition unit is acquired by a broadband dielectric spectrometer. After multiple effective tests are completed for each group of samples, statistical distribution methods are used for data processing. The crystallinity, thermal decomposition temperature, and melting temperature acquisition unit is acquired by a synchronous thermal analyzer. The sample is heated at a constant rate in a protective atmosphere. The thermal decomposition temperature is analyzed by obtaining the thermogravimetric curve at different aging times, and the crystallinity and melting temperature are analyzed by switching the atmosphere in stages. The carbonyl index acquisition unit is acquired by an attenuated total reflection infrared spectrometer. The set reflection crystal and incident angle are used, and the spectral data is calculated by scanning. The elastic modulus acquisition unit is acquired by a universal material testing machine. The universal material testing machine stretches the sample and records the maximum tensile force and the change in the marked distance at break.

10. The cable insulation life prediction system according to claim 8, characterized in that: The physical constraint layer embeds the dielectric constant differential equation as: where γ is the dielectric constant; is the activation energy; R is the ideal gas constant; T is the absolute temperature; is the electric field strength; n is the material constant; is the correction term output by LSTM.

Images