A method and device for selecting a cable insulation material, electronic equipment and a medium

Abstract

The application belongs to the technical field of data processing, and discloses a cable insulation material selection method and device, electronic equipment and medium. The method comprises the following steps: obtaining candidate material data, performing high-throughput calculation on the candidate material data to obtain corresponding target micro features, inputting the material composition and the corresponding target micro features into a target prediction model to obtain a corresponding performance prediction vector, taking the macro performance represented by the performance prediction vector as an optimization target, performing multi-objective iterative optimization solving on the candidate material data based on a non-dominated sorting genetic algorithm, introducing a dynamic coupling constraint rule in the optimization process to obtain a Pareto optimal solution set, and screening target material data from the Pareto optimal solution set. In this way, the micro features of a large amount of material data can be generated through high-throughput calculation, and the multi-objective optimization algorithm is used to realize the collaborative adaptation of the cable insulation material and the working condition, thereby solving the problems of low efficiency and multi-objective coupling optimization difficulty in traditional selection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a cable insulation material selection method and device, electronic equipment and medium. BACKGROUND

[0002] In the process of manufacturing cable insulation materials, the selection of cable insulation manufacturing materials is usually based on empirical formula and a large number of physical experiment methods, but this selection method needs to repeatedly adjust the composition and process, which is time-consuming, labor-intensive and high-cost, and also cannot intelligently select insulation materials that meet multi-dimensional performance constraints. It can be seen that the material selection efficiency of the prior art is low. SUMMARY

[0003] In view of the above defects of the prior art, the present application provides a cable insulation material selection method, device, electronic equipment and medium to solve the problem of low efficiency in selecting cable insulation materials in the prior art.

[0004] In a first aspect of the present application, a cable insulation material selection method is provided, comprising the steps of: S1, obtaining candidate material data, the candidate material data including material composition and target working conditions; S2, performing high-throughput calculation on the material composition, and extracting target micro features of the material composition, the high-throughput calculation including density functional theory calculation and molecular dynamics simulation; S3, inputting the material composition and the corresponding target micro features into a pre-trained target prediction model to obtain a corresponding performance prediction vector, the target prediction model being used to establish a mapping relationship between the material composition, the target micro features and the macro performance, the performance prediction vector including predicted values of multiple macro performances; S4, taking each macro performance indicated by the performance prediction vector as an optimization target, performing multi-objective optimization and solving on the candidate material data based on a non-dominated sorting genetic algorithm, and introducing a dynamic coupling constraint rule in the optimization process to obtain a Pareto optimal solution set, the dynamic coupling constraint rule being used to dynamically adjust the constraint conditions of the multi-objective optimization according to the target working conditions; S5, comparing the experimental performance data obtained by experiments on the candidate material data in the Pareto optimal solution set with the corresponding performance prediction vector to obtain a comparison deviation, if the comparison deviation is greater than a preset threshold, updating the target prediction model based on the experimental performance data and returning to step S1, and if the comparison deviation is not greater than the preset threshold, selecting target material data from the Pareto optimal solution set according to the preset weight of each optimization target.

[0005] In a second aspect, the application provides a cable insulation material selection device, comprising: an acquisition module configured to acquire candidate material data, the candidate material data comprising material composition and target working conditions; a calculation module configured to perform high-throughput calculation on the material composition, and extract target micro features of the material composition, the high-throughput calculation comprising density functional theory calculation and molecular dynamics simulation; a prediction module configured to input the material composition and the corresponding target micro features into a pre-trained target prediction model, to obtain a corresponding performance prediction vector, the target prediction model being configured to establish a mapping relationship between the material composition, the target micro features and macro performance, and the performance prediction vector comprising predicted values of multiple macro performances; a solving module configured to take each macro performance indicated by the performance prediction vector as an optimization target, perform multi-objective optimization solving on the candidate material data based on a non-dominated sorting genetic algorithm, and introduce a dynamic coupling constraint rule in the optimization process to obtain a Pareto optimal solution set, the dynamic coupling constraint rule being configured to dynamically adjust constraint conditions of the multi-objective optimization according to the target working conditions; and a material selection module configured to compare experimental performance data obtained by performing experiments on the candidate material data in the Pareto optimal solution set with the corresponding performance prediction vector, to obtain a comparison deviation, update the target prediction model based on the experimental performance data if the comparison deviation is greater than a preset threshold, and re-perform the step of acquiring candidate material data if the comparison deviation is not greater than the preset threshold, and select target material data from the Pareto optimal solution set according to preset weights of each optimization target.

[0006] In a third aspect, the application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0007] In a fourth aspect, the application provides a medium storing a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0008] The application has the following beneficial effects:

[0009] By acquiring candidate material data, the candidate material data including material composition and target working condition, performing high-throughput calculation on the material composition to obtain corresponding target micro features, inputting the material composition and the corresponding target micro features into a target prediction model to obtain a performance prediction vector, the target prediction model being used to establish a mapping relationship between the material composition, the target micro features and the macro performance; the performance prediction vector including predicted values of multiple macro performances, taking each macro performance represented by the performance prediction vector as an optimization target, performing multi-objective iterative optimization and solving on the candidate material data based on a non-dominated sorting genetic algorithm, and introducing a dynamic coupling constraint rule in the optimization process to obtain a Pareto optimal solution set, the dynamic coupling constraint rule being used to dynamically adjust constraint conditions of multi-objective optimization according to the target working condition; comparing experimental performance data obtained by experiments on the candidate material data in the Pareto optimal solution set with the corresponding performance prediction vector, and when the deviation is not greater than a preset threshold, screening target material data from the Pareto optimal solution set, so that micro features of a large amount of material data can be generated through high-throughput calculation, and the collaborative adaptation of material composition and working condition can be realized by using a multi-objective optimization algorithm, solving the problems of low efficiency of traditional selection and difficulty of multi-objective coupling optimization. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a cable insulation material selection method process schematic diagram provided by an embodiment of the present application;

[0011] Figure 2 is another cable insulation material selection method process schematic diagram provided by an embodiment of the present application;

[0012] Figure 3 is a cable insulation material selection device structure schematic diagram provided by an embodiment of the present application;

[0013] Figure 4 is an electronic device structure schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application will be further described below in conjunction with the drawings and embodiments, and it should be noted that in the description of the present application, the terms "upper", "lower", "left", "right", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular way, and therefore cannot be understood as limiting the present application. The terms "first", "second", "third" and the like are only for the purpose of description and cannot be understood as indicating or implying relative importance.

[0015] In the field of cable insulation material selection, the existing method is difficult to meet the multi-dimensional performance constraints of electrical strength, temperature adaptability, mechanical stability, etc., does not fully consider the influence of actual working conditions (such as temperature, humidity, laying conditions, etc.) on material performance, and lacks systematic modeling of the correlation between material microstructure and macrostructure, and cannot quickly predict the performance of complex component combinations. Therefore, the present application provides a cable insulation material closed-loop selection method based on machine learning and high-throughput calculation, which generates massive material data through high-throughput calculation, constructs a performance prediction model combined with machine learning, and realizes the coordinated cooperation of material composition and working conditions by using a multi-objective optimization algorithm, solving the problems of low efficiency and multi-objective coupling optimization of traditional selection.

[0016] First, the initial prediction model can be trained by a large amount of sample data to obtain a target prediction model for calculating the performance prediction vector indicating the macroscopic performance corresponding to the candidate material data.

[0017] The initial prediction model can be obtained based on XGBOOST, and a set of initial hyperparameters (hyperparameters can be default parameters or empirical values) are set for XGBOOST, for example:

[0018] The number of trees (n_estimators): 100, learning rate (learning_rate): 0.1, maximum depth (max_depth): 6;

[0019] Regularization parameters (lambda=0.1, alpha=0.01) to prevent overfitting; 5-fold cross-validation is used to optimize the hyperparameter combination through grid search.

[0020] In this way, an untrained initial prediction model can be obtained.

[0021] Further, material samples of different formulations and corresponding macroscopic performance samples are selected, and the macroscopic performance samples are obtained through standard experimental tests; high-throughput calculation is performed on the material samples to obtain micro-feature samples corresponding to the material samples; a training sample set is constructed with the material samples, micro-feature samples and macroscopic performance samples, and the training sample set includes a training set, a validation set and a test set; the initial prediction model is trained and hyperparameters are optimized according to the training sample set, and when the prediction accuracy of the macroscopic performance sample output by the initial prediction model on the test set reaches a preset requirement, a target prediction model is obtained. It should be noted that the target prediction model can also be trained by other machine learning models, which is not limited here.

[0022] Specifically, taking a 35 kV cable as an example, the cable insulation manufacturing material can include a base body, a filler, and a crosslinking agent. Before model training, material samples are obtained, for example, polyethylene (PE) is selected as the base material, the nano filler type is Al2O3 and SiO2, the filler content gradient is set to 5%, 10%, 15%, and 20%, and the crosslinking agent content is set to 1%, 3%, and 5%. Then, 20 groups of ingredient combinations are constructed. Then, through standard laboratory tests, the macroscopic performance samples of each group of material samples under the corresponding working conditions are obtained, for example, 10 macroscopic performances such as breakdown field strength (IEC 60243 standard), elongation at break (GB / T 1040.3 standard), glass transition temperature (DSC test), volume resistivity (GB / T 1410 standard), and the like. The environmental parameters (temperature 25℃±2℃, humidity 50%±5%) of the standard experiment are recorded as historical working condition inputs to the initial prediction model for training to learn the relationship between the working condition and the material sample-microscopic feature sample-macroscopic performance sample, that is, the material sample, the recorded historical working condition parameters, the macroscopic performance sample, and the microscopic feature sample are input into the prediction model for training to obtain a trained and usable target prediction model. The target prediction model not only reflects the mapping relationship between the material composition, the target microscopic feature, and the macroscopic performance, but also reflects the mapping relationship between the material composition, the corresponding working condition parameters, and the target microscopic feature and the macroscopic performance. Therefore, subsequent input of the material composition and the target working condition (the environment where the cable needs to be applied) into the model can obtain the target material data that can be applied in the target working condition.

[0023] The molecular structure parameters (such as molecular chain length and crosslinking density) of PE-based polymers are extracted from a material database (such as Materials Project) as corresponding microscopic features. Then, the microscopic features obtained through high-throughput calculation of the material samples are used as a supplement to the database, that is, the microscopic feature samples are composed of two parts, one part is the existing microscopic features obtained from the pre-constructed database, and the other part is the microscopic features obtained through high-throughput calculation.

[0024] The training sample set is constructed from the material samples, the microscopic feature samples, and the corresponding macroscopic performance samples. The training sample set can be divided into a training set, a validation set, and a test set in a ratio of, for example, 8:1:1.

[0025] The training sample set is constructed from the material samples, the microscopic feature samples, and the corresponding macroscopic performance samples. The training sample set can be divided into a training set, a validation set, and a test set in a ratio of, for example, 8:1:1.

[0026] Specifically, outliers can be removed using the Relyda criterion (3σ), missing values can be filled by linear interpolation (e.g., when the volume resistivity test of a certain group of samples fails, the mean value of the adjacent samples of the same composition is filled); the macroscopic performance (e.g., breakdown field strength 30-50kV / mm, elongation at break 150%-300%) is normalized to the [0,1] interval by Min-Max to avoid dimension influence on model training. In order to improve the efficiency of model training and the comprehensiveness of training set, virtual samples can be generated by parameter fine-tuning, for example, 20 groups of material samples obtained initially are fine-tuned to expand to 200 groups of samples, so that a more comprehensive training sample set can be constructed, and the fine-tuning can be fine-tuning of the composition ratio of the material samples or fine-tuning of the working condition parameter samples, which is not limited here.

[0027] Then, using the assembled training set, the initial prediction model is trained by continuously inputting the data in the training set into the model, calculating the error between the predicted value and the true label (macroscopic performance sample), adjusting the parameters inside the model to minimize the error by using optimization algorithms, and achieving the model training effect. After model training, the validation set is used to adjust the hyperparameters (such as tree depth, learning rate, etc.) inside the model, and finally the test set that has never participated in the initial prediction model training is used to evaluate the model. When the prediction accuracy of the model on the test set reaches the preset requirement, the target prediction model is obtained (for example, the prediction R² of the breakdown field strength is 0.92, the mean square error (MSE) is 1.2kV / mm; the prediction R² of the elongation at break is 0.88, the MSE is 8%; and the prediction error of the breakdown field strength and the elongation at break is ≤5%, at this time the prediction accuracy of the test set reaches the preset requirement, meeting the precision requirement of engineering application).

[0028] The key influencing factors of insulation performance can also be determined by the feature importance ranking of XGBoost: filler interface bonding energy (importance ratio 35%), RMSD at high temperature (25%), and crosslinking density (20%), which provides priority reference for subsequent optimization to help determine the weight of the optimal solution after multi-objective optimization.

[0029] In this way, after obtaining the target prediction model through model training, the macroscopic performance corresponding to the candidate material data can be quickly and accurately predicted through candidate material data and computer simulation (DFT / MD), without the need for real experimental tests to obtain the macroscopic performance of the candidate material, which can greatly save research and development costs and time.

[0030] Next, a cable insulation material selection method and device according to an embodiment of the present application will be described in detail with reference to the accompanying drawings.

[0031] Figure 1 is a flowchart of a cable insulation material selection method provided by an embodiment of the present application, as shown inFigure 1 As shown, the selection method for the cable insulation material includes:

[0032] S1, Obtain candidate material data, which includes material composition and target operating conditions.

[0033] Specifically, the first step is to construct a target prediction model. After modeling, candidate material data can be obtained. This data includes the corresponding material composition and target operating conditions. Insulating materials are generally composed of a matrix, filler, and additives. Considering the application environment, the input candidate material data can be something like "polyethylene + 10% silica (SiO2) + 2% crosslinking agent, operating temperature ≤80℃, humidity ≤70%". This data represents the material composition that can be selected under operating conditions of ≤80℃ and ≤70% humidity, i.e., its formulation. Polyethylene is the matrix, silica is the filler, and the crosslinking agent is a key component of cable insulation materials used to change the molecular structure of the material, thereby giving it new properties. Of course, under special operating conditions, it can also be a pure matrix or the formulation of the candidate material can be adjusted based on the actual operating conditions. By analyzing and iteratively optimizing the candidate material data, the component ratios that perform well under the corresponding operating conditions can be obtained, and then the corresponding materials can be used to manufacture cable insulation.

[0034] It should be noted that the candidate material data can be selected from a pre-built database, that is, a database containing material components of cable insulation layers that can cope with certain complex environments can be pre-built, and a large number of material components can be selected from it for subsequent algorithm iteration optimization to select the optimal material formula. Alternatively, a large number of candidate material data can be randomly generated by the algorithm, that is, 100 sets of candidate material data can be randomly generated using a non-dominated sorting genetic algorithm. There is no restriction here. The target operating conditions can be directly input by the user. For example, if a high-voltage cable needs to be made for use in a high temperature and high humidity environment, then the target operating conditions that meet the high temperature and high humidity requirements need to be input to simulate the actual high temperature and high humidity environment. For example, the operating temperature is ≤90℃ and the operating humidity is ≥95%RH to simulate the harsh operating conditions of underground cables in tropical regions.

[0035] S2 performs high-throughput calculations on material composition, extracting the target microscopic features of the material composition. The high-throughput calculations include density functional theory calculations and molecular dynamics simulations.

[0036] The target micro-features represent the micro-features of the material composition, such as hundreds of features such as “interfacial bonding energy”, “molecular chain movement ability”, and the like, which represent the inherent and intrinsic micro-features of the material composition, and the target micro-features correspond to the macro performance (such as electrical strength, heat resistance) of the candidate material data. The calculation of the target micro-features corresponding to the material composition prepares input data for the subsequent machine learning model, and through high-throughput calculation simulation of a large number of material compositions, most of the trial and error experiments can be replaced, and the foundation for efficient material selection is laid.

[0037] In some embodiments, step S2 includes: performing density functional theory calculation on the material composition to obtain electrical performance correlation features corresponding to the material composition, the electrical performance correlation features including interfacial bonding energy, highest occupied molecular orbital energy level, lowest unoccupied molecular orbital energy level, and energy gap between the highest occupied molecular orbital energy level and the lowest unoccupied molecular orbital energy level; performing molecular dynamics simulation on the material composition to obtain molecular dynamics features corresponding to the material composition; and processing the electrical performance correlation features and the molecular dynamics features to obtain target feature data.

[0038] Specifically, the micro-features of the material composition can be obtained from the atomic nucleus and electron level through high-throughput calculation to realize efficient calculation instead of a large number of experiments.

[0039] Firstly, high-throughput calculation can be realized through density functional theory (DFT) calculation and molecular dynamics (MD) simulation.

[0040] When performing DFT calculation, VASP software can be used. The full name of the software is Vienna Ab initio Simulation Package, which is used for electronic structure calculation and quantum mechanics-molecular dynamics simulation. Based on Perdew-Burke-Ernzerhof functional, that is, PBE functional, the PBE functional considers the gradient of electron density and the limit behavior of uniform electricity based on the principles of quantum mechanics and statistical mechanics. Therefore, the VASP software can be used to calculate the bonding energy (unit: eV) of the interface between different fillers (such as Al2O3, SiO2) and the PE matrix and the charge distribution based on the PBE functional, and the interface bonding strength features are obtained to obtain the interface bonding energy.

[0041] DFT calculation of material composition can accurately calculate the behavior of electrons in material composition, obtain micro characteristics (such as interface binding energy, highest occupied molecular orbital energy level, lowest unoccupied molecular orbital energy level and energy gap between them) associated with electrical properties, which directly determine the insulation and electrical properties of the material composition. Among them, the interface binding energy: quantifies the degree of firmness of the combination between the filler (such as SiO2) and the matrix (such as PE), the higher the binding energy, the more stable the interface, and the less likely the material to fail due to defects under electric field and thermal stress.

[0042] It should be noted that the orbitals in the molecule can be divided into occupied orbitals, unoccupied orbitals and semi-occupied orbitals according to the electron filling condition. The highest energy level in the occupied orbital is called the highest occupied molecular orbital (Highest Occupied Molecular, HOMO), and the lowest energy level in the unoccupied orbital is called the lowest unoccupied molecular orbital (Lowest Unoccupied Molecular Orbital, LUMO). HOMO is used to represent the position of the most easily lost electron, and LUMO is used to represent the position of the most easily accepted electron. The energy level difference between the two determines the chemical stability and optical properties of the molecule. The smaller the energy level difference, the easier it is for the molecule to participate in the reaction. Therefore, obtaining the HOMO energy level, the LUMO energy level and the gap between them can directly reflect the difficulty of electron transition. The greater the energy gap, the more energy is required for the electron to transition from the valence band to the conduction band, the more difficult it is for the material to be punctured, and the higher the electrical insulation strength. Thus, predicting the core intrinsic electrical properties of the material from the electron scale.

[0043] Similarly, MD simulation calculation of material composition can use LAMMPS software (Large-scale Atomic / Molecular Massively Parallel Simulator, a large-scale atomic / molecular parallel simulator for molecular dynamics-related calculations and simulations) to construct a matrix-filler composite system model (box size 20x20x20ų) and simulate under NPT ensemble (NPT represents isothermal-isobaric ensemble, that is, the ensemble ensures that the number of atoms, pressure and temperature of the system remain unchanged):

[0044] Temperature: 353K (corresponding to actual operating temperature 80℃), pressure 1atm, time step 1fs, simulation time 100ps;

[0045] Force field: COMPASS force field is selected to describe the interaction between atoms, focusing on simulating the motion behavior of molecular chains at high temperatures.

[0046] Thus, by solving the classical motion equation between atoms, simulating the trajectory of atoms and molecules in a certain time, the molecular dynamics characteristics can be obtained, such as the calculation of the root mean square displacement, the radial distribution function, the elastic modulus, the Poisson ratio, etc.

[0047] Among them, the root mean square displacement is used to reflect the movement ability of the molecular chain at high temperature, which is directly related to the glass transition temperature and thermal stability; the radial distribution function is used to analyze the compatibility of the filler and the matrix interface, to judge whether the filler is uniformly dispersed or not; the elastic modulus and Poisson ratio reflect the macroscopic mechanical properties directly calculated by simulating the stress-strain response of the material.

[0048] In this way, the thermodynamic behavior and macroscopic performance of the material composition under certain working conditions (such as high temperature and stress) can be predicted from the atomic / molecular scale.

[0049] By processing and fusing the electrical performance-related characteristics and molecular dynamics characteristics obtained by DFT and MD calculations, the target micro-features can be obtained. By performing DFT and MD calculations on the initial multiple material compositions, the corresponding target micro-features are generated, and the performance data of the material composition in multiple dimensions can be obtained for multi-dimensional optimization. Compared with traditional experimental trial and error, the efficiency is greatly improved and the cost is greatly reduced. The fundamental reasons for the advantages and disadvantages of the corresponding material performance are explained through micro-features, and a rich and more in-depth data set is constructed, which lays a solid foundation for analyzing the matching degree of candidate material data.

[0050] Of course, it should be noted that the initial micro-features obtained after high-throughput calculation of the material composition may include hundreds or even thousands of dimensions of micro-features, and may contain dimension characteristics that have little effect on insulation performance. If all of them are used for multi-objective optimization, it may increase the algorithm pressure. Therefore, in order to reduce the pressure of the model, principal component analysis can be performed on the initial micro-features to obtain the target micro-features.

[0051] The target micro-features obtained by processing the electrical performance-related characteristics and molecular dynamics characteristics include: fusing the electrical performance-related characteristics and molecular dynamics characteristics to obtain initial micro-features; reducing the dimensionality of the initial micro-features by principal component analysis to obtain target micro-features.

[0052] Specifically, an initial comprehensive micro-feature containing hundreds or even thousands of dimensions can be formed by the initial fusion of features, thus creating a unique and extremely comprehensive "digital DNA" or "digital fingerprint" for each material component, which fully describes the internal mechanism of the formulation in terms of electronic and atomic levels, and why it exhibits certain macroscopic performance. However, if the micro-feature containing thousands of dimensions is input into the model for processing, it may increase the algorithmic pressure and cause the model to collapse, so the initial micro-feature can be reduced in dimension by principal component analysis to select the most influential features for insulation performance, obtaining the target micro-feature, such as reducing the feature to 50 dimensions or other number of dimensions of the target micro-feature. This reduces the pressure on the target prediction model without losing the main information data.

[0053] S3, inputting the material components and the corresponding target micro-features into the pre-trained target prediction model to obtain a corresponding performance prediction vector, the target prediction model being used to establish a mapping relationship between the material components, the target micro-features and the macroscopic performance, and the performance prediction vector including predicted values of multiple macroscopic performance indicators.

[0054] Specifically, the candidate material data and the corresponding target micro-features are input into the trained target prediction model, and a performance prediction vector corresponding to the material components and the target micro-features can be output, which includes predicted values of multiple performance indicators of the corresponding material components, which can be understood as a vector representation of macroscopic performance, such as breakdown field strength, elongation at break, etc. Thus, a bridge from micro to macro performance of the material components can be established. By using the target micro-features as a bridge, a quantitative mapping relationship between the components, microstructure and macroscopic performance can be established to analyze the improvement direction of the material from a micro perspective.

[0055] The target prediction model simultaneously receives the material component data and the calculated target micro-features to jointly establish the mapping relationship. The purpose is that although the target micro-features can reflect the material mechanism in depth, some macroscopic performances may be directly related to the simple linear combination or specific interaction effect of the components, and these relationships may not be fully captured by the micro-features. By simultaneously inputting both into the model, the dual advantages of'mechanism-driven' and 'data-driven' can be combined to ensure that the model can understand the deep principle and not lose the statistical rules of the component surface, thereby obtaining higher prediction accuracy and better generalization ability.

[0056] S4, taking each macroscopic performance indicated by the performance prediction vector as an optimization target, performing multi-objective optimization on the candidate material data based on a non-dominated sorting genetic algorithm, and introducing a dynamic coupling constraint rule in the optimization process to obtain a Pareto optimal solution set, the dynamic coupling constraint rule being used to dynamically adjust the constraint conditions of the multi-objective optimization according to the target working condition.

[0057] The multi-objective coupling is used to realize multi-objective optimization of each macroscopic performance, but it should be noted that other optimization objectives, such as the cost and environmental cost of the candidate material, are also considered in the actual application process.

[0058] In this way, more comprehensive multi-objective optimization can be realized by maximizing the performance indicators and minimizing the cost constraints through multi-objective coupling. The maximization of performance indicators is reflected in the electrical strength (breakdown field strength), mechanical stability (elongation at break, elastic modulus), and thermal stability (glass transition temperature). The minimization of cost constraints is reflected in the material raw material cost (such as the high price of nano filler Al2O3 compared to the PE matrix), environmental sensitivity (water tree aging resistance, weather resistance. On the other hand, the cost of the cable candidate material can also be calculated through macroscopic performance, that is, the composition and corresponding proportion of the material are calculated inversely through the macroscopic performance of the insulation, and then the corresponding cost is obtained. Of course, this is not limited here, and the cost data can also be obtained more conveniently through other algorithms.

[0059] The constraint conditions can include various types of conditional constraints, such as material proportion constraints, working condition parameterization constraints, and insulation manufacturing time constraints. From the aspect of working condition parameters, to ensure the effectiveness of the prediction data, the constraint conditions of the working conditions can be:

[0060] The operating temperature range (such as ≤90℃), the laying environment humidity (such as ≤85%RH), and the mechanical load (such as the soil pressure of buried laying), that is, the actual environment of the cable laying is avoided to introduce unnecessary data increase algorithm pressure, and the working condition parameters in all candidate material data are constrained by the above conditions. If the corresponding temperature exceeds 90 degrees or the humidity is greater than 85%, the corresponding data is directly discarded. Of course, the actual constraint conditions can be adaptively adjusted, which is not limited here.

[0061] A non-dominated sorting genetic algorithm is used for searching, and dynamic coupling constraint rules can also be introduced in the optimization process to intelligently adjust the optimization strategy according to the target working condition, for example, the system identifies that the working condition temperature exceeds 80℃, which automatically triggers the rule and requires the optimization algorithm to search only those material compositions with a high-temperature filler proportion ≥15%.

[0062] In this way, a set of Pareto optimal solutions can be obtained, which represents the best balance of solutions. This is not explained in detail here, and each solution in the set represents a different performance trade-off (such as A scheme with extreme performance but high cost, B scheme with lowest cost but performance meets the standard), and no scheme is absolutely superior to another in all aspects. Therefore, a large number of candidate material data that still perform well under a large number of complex constraints are cached. The non-dominated sorting genetic algorithm is a principle known to those skilled in the art, which is not explained in detail here.

[0063] The target working condition includes a working condition temperature and a working condition humidity, the material composition includes different proportions of a base material, a filler, and a cross-linking agent, and the dynamic coupling constraint rule is used to dynamically adjust a constraint condition of multi-objective optimization according to the target working condition, including: when the working condition temperature exceeds a preset temperature, adjusting a content of a first type of filler in the material composition to exceed a first preset proportion, and adjusting a content of the cross-linking agent according to an increase of the working condition temperature; when the working condition humidity exceeds a preset humidity, adjusting a content of a second type of filler in the material composition to exceed a second preset proportion, predicting a water tree aging rate of the adjusted material composition under the target working condition based on a target prediction model, and positively adjusting a cross-linking density according to the water tree aging rate and a change value of the working condition humidity.

[0064] Specifically, the optimization algorithm can dynamically and intelligently adjust the search strategy according to the environment (target working condition) of the cable, so as to find the material composition proportion that is really suitable for the environment, form a Pareto optimal solution set, and then select an optimal solution from the Pareto optimal solution set to obtain a material composition that can cope with the target working condition.

[0065] When the working condition temperature is greater than a preset temperature (for example, 80 DEG C), a high-temperature-resistant filler is forced to be used, and the content of the first type of filler (such as Al2O3) is adjusted to be greater than or equal to a first preset proportion (for example, Al2O3 accounts for 15% in all components), so that the excellent thermal stability and thermal conductivity of the inorganic filler such as Al2O3 can effectively inhibit the thermal motion of the polymer body at high temperature, prevent the material from softening, deforming and aging, and all formulations that are not resistant to high temperature can be screened out.

[0066] When the working condition temperature increases, the content of the cross-linking agent can be positively adjusted according to the increase of the working condition temperature, so as to intelligently increase the cross-linking degree, for example, when the temperature is 10 DEG C higher than 80 DEG C, the content of the cross-linking agent needs to be increased by 1% based on the basic content. To ensure thermal stability.

[0067] When the working condition humidity is greater than a preset humidity (for example, 80% RH), the constraint condition is adjusted to require that the content of the second type of filler (for example, SiO2) is greater than or equal to a second preset proportion (for example, 5%), so that the SiO2 can fill the free volume inside the polymer to form a physical barrier to hinder the penetration of water molecules, while improving the hydrophobicity of the material, fundamentally inhibiting the generation and formation of "water trees", and the water tree aging rate also needs to be predicted based on a target prediction model, and the cross-linking density is positively adjusted according to the rate and humidity change (for example, when the humidity increases by 10%, the cross-linking density is increased by 5%). It should be noted that the first type of filler includes but is not limited to Al2O3, and the second type of filler includes but is not limited to SiO2.

[0068] In this way, the selected material not only contains sufficient hydrophobic filler, but also has a cross-linking structure that is specially optimized to resist the erosion of water trees for a long time, ensuring the service life of the cable.

[0069] By introducing dynamic coupling constraint rules, the embodiments of this application can move from automation to intelligence, enabling material selection to be tailored to specific application environments, greatly improving the reliability of the final product, significantly reducing the search space of the algorithm, avoiding wasting time on a large number of invalid formulas that are not applicable to the current environment, and allowing the optimization calculation to converge to the optimal solution set more quickly.

[0070] S5. Based on the experimental performance data obtained from the experiment on the material composition in the Pareto optimal solution set, compare it with the corresponding performance prediction vector. If the comparison deviation is greater than the preset threshold, update the target prediction model based on the experimental performance data and return to step S1. If the comparison deviation is not greater than the preset threshold, select the target material data from the Pareto optimal solution set according to the preset weight of each optimization objective.

[0071] Specifically, a few of the most promising formulations are selected from the Pareto optimal solution set for actual experiments. Alternatively, a few of the most promising formulations can be selected based on the preset weights of each optimization objective to obtain real performance data. In the embodiments of this application, the experimental performance data can be obtained according to the following standards: Breakdown field strength: IEC 60243-1, electrode spacing 2mm, voltage boosting rate 2kV / s, 3 tests and average value (32.1kV / mm, error of 1.2% compared with the prediction); Elongation at break: GB / T 1040.3, tensile rate 50mm / min, average value 215% (error 2.3%); Water tree resistance: IEC 62067 standard, 12kV pressure for 1000h at 85℃ / 85% RH, water tree length ≤0.5mm (compliant).

[0072] The experimental performance data is compared with the macroscopic performance indicated by the performance prediction vector obtained in S3. If the comparison deviation is greater than the preset threshold (e.g., the predicted breakdown field strength is 35kV / mm, but the actual measured value is only 28kV / mm, while the preset breakdown field strength threshold is 5kV / mm), it indicates that the model prediction is not accurate enough. In this case, new experimental performance data is added to the training set, the target prediction model is updated, and then steps S1-S5 are repeated in S1 to start optimization. This forms a closed loop, making the model more and more accurate with iteration. Of course, if the comparison deviation is less than or equal to the preset threshold, it indicates that the prediction result is accurate and reliable. Then, according to the preset weights, the final target material data is selected from the Pareto solution set. The preset weights mainly reflect the current emphasis on different optimization objectives, i.e., which performance and cost are more important. In this way, through the closed loop of prediction-verification-feedback, the system continuously learns and becomes more and more intelligent. For example, an electrical strength weight of 0.4 indicates that the user values ​​the insulation capacity of the cable very much. A cost weight of 0.3 indicates that the user is also concerned about cost, but not as important as electrical performance. Mechanical performance weight 0.3: This indicates that users also value the material's flexibility, tensile strength, and other mechanical properties. These weights can be adjusted flexibly according to the actual project requirements. For example, for projects with tight budgets, the cost weight can be adjusted to 0.5. There are no restrictions here.

[0073] The Pareto front is directly solved using a multi-objective optimization algorithm to obtain a series of equilibrium solutions. Dynamic coupling constraint rules ensure that the optimization direction always closely follows the actual application scenario (such as high temperature and high humidity), so that the selected target candidate materials perform optimally under the target operating conditions. The closed-loop iterative mechanism enables the entire system to become more and more accurate through continuous learning, from automation to intelligence, ensuring the long-term viability and accuracy of the method.

[0074] After obtaining the Pareto optimal solution set, the comprehensive evaluation value of each solution in the Pareto optimal solution set can be calculated according to the approximation ideal solution sorting method, and the solution with the highest comprehensive evaluation value can be selected as the target material data.

[0075] The Approximation to Ideal Solution Ranking Method (TOPSIS) is a multi-attribute decision-making method. It can find the best ideal solution and the worst ideal solution from the Pareto optimal solution set, and then calculate the distance between each solution in the Pareto optimal solution set and these two ideal solutions. Finally, it selects the solution that is closest to the optimal solution and furthest from the worst solution. In this way, after combining all indicators (macro performance, cost requirements) and preset weights, the most balanced comprehensive optimal solution set of target material data can be obtained.

[0076] Figure 2This is a schematic flowchart of another method for selecting insulating materials provided in an embodiment of this application, as shown below. Figure 2 As shown:

[0077] After model training is complete, a non-dominated sorting genetic algorithm is used for iterative optimization to select target material data. First, high-throughput computation and feature extraction are performed on the acquired material composition to obtain the target microscopic features. Then, after the performance prediction model is built, machine learning is used to predict performance based on the material composition and corresponding target microscopic features. The macroscopic performance indicated by the performance prediction vector is used as the optimization objective to achieve multi-objective collaborative optimization and operating condition coupling. Afterwards, experiments are conducted based on the Pareto optimal solution set obtained from the multi-objective optimization to determine whether the experimental performance data meets the corresponding indicators, i.e., whether the comparison deviation with the corresponding performance prediction vector is less than a preset threshold. This determines whether to re-iterate or output the target material data. If the deviation between the experimental value and the predicted value is >5% (e.g., the breakdown field strength of a batch of samples is only 28kV / mm), then:

[0078] Trace the data chain: Check whether the DFT calculation missed the interface defect model, or whether the MD simulation force field needs to be updated;

[0079] Supplementary experimental data: Collect additional microstructure data of this component (such as TEM characterization of interface distribution) to expand the training set;

[0080] Retrain the model: Adjust the XGBoost parameters or introduce a deep learning model (such as LSTM) to improve prediction accuracy under complex conditions.

[0081] Once the target material data is obtained, it can be used for practical industrial applications.

[0082] It should be noted that the optimization method of this application embodiment can be applied to different laying scenarios (such as underground cable trenches and submarine cables). By adjusting the operating parameters (such as submarine high voltage and high salt spray), adaptable material solutions can be quickly generated to verify the transferability of the method.

[0083] Through iterative optimization of the closed-loop framework, the selection of 35kV cable insulation materials has been upgraded from "experience-based trial and error" to "data-driven" methods, demonstrating significant advantages, especially in adaptability to complex working conditions and balance of multiple objectives.

[0084] Figure 3 This application provides a cable insulation material selection device, such as... Figure 3 As shown, the cable insulation material selection device includes:

[0085] The acquisition module 301 is used to acquire candidate material data, which includes material composition and target working conditions.

[0086] Calculation module 302 is used to perform high-throughput calculations on material composition and extract target microscopic features of material composition. High-throughput calculations include density functional theory calculations and molecular dynamics simulations.

[0087] The prediction module 303 is used to input the material composition and the corresponding target micro-features into the pre-trained target prediction model to obtain the corresponding performance prediction vector. The target prediction model is used to establish the mapping relationship between the material composition, target micro-features and macro-performance. The performance prediction vector includes the predicted values ​​of multiple macro-performances.

[0088] The solution module 304 is used to perform multi-objective optimization on candidate material data based on a non-dominated sorting genetic algorithm with each macroscopic performance indicated by the performance prediction vector as the optimization objective. Dynamic coupling constraint rules are introduced during the optimization process to obtain the Pareto optimal solution set. The dynamic coupling constraint rules are used to dynamically adjust the constraints of multi-objective optimization according to the target working conditions.

[0089] The material selection module 305 compares the experimental performance data obtained from experiments on candidate material data in the Pareto optimal solution set with the corresponding performance prediction vector to obtain the comparison deviation. If the comparison deviation is greater than a preset threshold, the target prediction model is updated based on the experimental performance data, and the step of obtaining candidate material data is re-executed. If the comparison deviation is not greater than the preset threshold, the target material data is selected from the Pareto optimal solution set according to the preset weights of each optimization objective.

[0090] In some embodiments, the calculation module 302 is specifically used to perform density functional theory calculations on the material composition to obtain the electrical performance correlation characteristics corresponding to the material composition. The electrical performance correlation characteristics include the interface binding energy, the highest occupied molecular orbital energy level, the lowest unoccupied molecular orbital energy level, and the band gap between the highest occupied molecular orbital energy level and the lowest unoccupied molecular orbital energy level; to perform molecular dynamics simulations on the material composition to obtain the molecular dynamics characteristics corresponding to the material composition; and to process the electrical performance correlation characteristics and molecular dynamics characteristics to obtain the target feature data.

[0091] In some embodiments, the calculation module 302 is specifically used to fuse electrical performance correlation features and molecular dynamics features to obtain initial microscopic features; and to reduce the dimensionality of the initial microscopic features through principal component analysis to obtain target microscopic features.

[0092] In some embodiments, the prediction module 303 is specifically used to select material samples with different compositions and corresponding macroscopic performance samples. The macroscopic performance samples are obtained by performing standard experimental tests on the material samples. High-throughput calculations are performed on the material samples to obtain microscopic feature samples corresponding to the material samples. A training sample set is constructed using the material samples, microscopic feature samples, and macroscopic performance samples. The training sample set includes a training set, a validation set, and a test set. The initial prediction model is trained and its hyperparameters are optimized based on the training set. When the prediction accuracy of the initial prediction model on the test set reaches the preset requirements, the target prediction model is obtained.

[0093] In some embodiments, the prediction module 303 is specifically used to clean and normalize the macroscopic performance samples; and associate the cleaned and normalized macroscopic performance samples with the corresponding material samples and microscopic feature samples to form a training sample set.

[0094] In some embodiments, the solving module 304 is specifically used to adjust the content of the first type of filler in the material composition to exceed the first preset ratio when the operating temperature exceeds the preset temperature, and adjust the crosslinking agent content according to the increase in operating temperature; when the operating humidity exceeds the preset humidity, adjust the content of the second type of filler in the material composition to exceed the second preset ratio, and predict the water tree aging rate of the adjusted material composition under the target operating conditions based on the target prediction model, and positively adjust the crosslinking density according to the change in water tree aging rate and operating humidity.

[0095] In some embodiments, the material selection module 305 is specifically used to calculate the comprehensive evaluation value of each solution in the Pareto optimal solution set according to the approximation ideal solution sorting method, and select the solution with the highest comprehensive evaluation value as the target material data.

[0096] Figure 4 This is a schematic diagram of the electronic device 4 provided in an embodiment of this application. Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the various method embodiments described above. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the various device embodiments described above.

[0097] Electronic device 4 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 4 may include, but is not limited to, processor 401 and memory 402. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or different components.

[0098] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0099] The memory 402 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM of the electronic device 4. The memory 402 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 4. The memory 402 can also include both internal and external storage units of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0100] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0102] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for selecting cable insulation materials, characterized in that, Includes the following steps: S1, Obtain candidate material data, the candidate material data including material composition and target working conditions; S2, perform high-throughput calculations on the material composition to extract the target microscopic features of the material composition. The high-throughput calculations include density functional theory calculations and molecular dynamics simulations. S3, input the material composition and the corresponding target micro-features into the pre-trained target prediction model to obtain the corresponding performance prediction vector. The target prediction model is used to establish the mapping relationship between the material composition, the target micro-features and macro-performance. The performance prediction vector includes the predicted values ​​of multiple macro-performances. S4, taking the macroscopic performances indicated by the performance prediction vector as optimization objectives, multi-objective optimization is performed on the candidate material data based on the non-dominated sorting genetic algorithm, and dynamic coupling constraint rules are introduced in the optimization process to obtain the Pareto optimal solution set. The dynamic coupling constraint rules are used to dynamically adjust the constraint conditions of multi-objective optimization according to the target working conditions. S5, based on the experimental performance data obtained by experimenting on the candidate material data in the Pareto optimal solution set, compare it with the corresponding performance prediction vector to obtain the comparison deviation. If the comparison deviation is greater than a preset threshold, update the target prediction model based on the experimental performance data and return to step S1. If the comparison deviation is not greater than the preset threshold, select the target material data from the Pareto optimal solution set according to the preset weights of each optimization objective.

2. The method according to claim 1, characterized in that, Step S2 includes: Density functional theory calculations were performed on the material composition to obtain the electrical performance correlation characteristics corresponding to the material composition. The electrical performance correlation characteristics include the interface binding energy, the highest occupied molecular orbital energy level, the lowest unoccupied molecular orbital energy level, and the band gap between the highest occupied molecular orbital energy level and the lowest unoccupied molecular orbital energy level. Molecular dynamics simulations were performed on the material composition to obtain the molecular dynamics characteristics corresponding to the material composition. The electrical performance correlation features and the molecular dynamics features are processed to obtain target feature data.

3. The method according to claim 2, characterized in that, The process of processing the electrical performance correlation features and the molecular dynamics features to obtain target feature data includes: The electrical performance correlation characteristics and the molecular dynamics characteristics are fused to obtain the initial microscopic characteristics; Principal component analysis is used to reduce the dimensionality of the initial micro-features to obtain the target micro-features.

4. The method according to claim 1, characterized in that, Before step S3, the following is also included: Material samples with different compositions and corresponding macroscopic performance samples were selected, and the macroscopic performance samples were obtained by performing standard experimental tests on the material samples; The high-throughput calculation is performed on the material sample to obtain the microscopic feature sample corresponding to the material sample; A training sample set is constructed using the material sample, the microscopic feature sample, and the macroscopic performance sample, and the training sample set includes a training set, a validation set, and a test set; The initial prediction model is trained and its hyperparameters are optimized based on the training set. When the prediction accuracy of the initial prediction model on the test set reaches the preset requirement, the target prediction model is obtained.

5. The method according to claim 4, characterized in that, The construction of the training sample set using the material sample, the microscopic feature sample, and the macroscopic performance sample includes: The macroscopic performance samples were cleaned and normalized. The cleaned and normalized macroscopic performance samples are associated with the corresponding material samples and microscopic feature samples to form the training sample set.

6. The method according to claim 1, characterized in that, The target operating conditions include operating temperature and operating humidity; the material composition includes matrix materials, fillers, and crosslinking agents in different proportions; and the dynamic coupling constraint rules are used to dynamically adjust the constraints of multi-objective optimization according to the target operating conditions, including: When the operating temperature exceeds the preset temperature, the content of the first type of filler in the material composition is adjusted to exceed the first preset ratio, and the content of the crosslinking agent is adjusted according to the increase in the operating temperature. When the operating humidity exceeds the preset humidity, the content of the second type of filler in the material composition is adjusted to exceed the second preset ratio, and the water tree aging rate of the adjusted material composition under the target operating conditions is predicted based on the target prediction model. The crosslinking density is positively adjusted according to the water tree aging rate and the change value of the operating humidity.

7. The method according to claim 1, characterized in that, The step of selecting target material data from the Pareto optimal solution set according to the preset weights of each optimization objective includes: The comprehensive evaluation value of each solution in the Pareto optimal solution set is calculated according to the approximation ideal solution sorting method, and the solution with the highest comprehensive evaluation value is selected as the target material data. The comprehensive evaluation value is calculated based on a preset weight.

8. A device for selecting cable insulation materials, characterized in that, include: The acquisition module is used to acquire candidate material data, which includes material composition and target working conditions; The computation module is used to perform high-throughput calculations on the material composition and extract the target microscopic features of the material composition. The high-throughput calculations include density functional theory calculations and molecular dynamics simulations. The prediction module is used to input the material composition and the corresponding target micro-features into a pre-trained target prediction model to obtain the corresponding performance prediction vector. The target prediction model is used to establish the mapping relationship between the material composition, the target micro-features and macro-performance. The performance prediction vector includes the predicted values ​​of multiple macro-performances. The solution module is used to perform multi-objective optimization on the candidate material data based on the non-dominated sorting genetic algorithm, with each macroscopic performance indicated by the performance prediction vector as the optimization objective. Dynamic coupling constraint rules are introduced during the optimization process to obtain the Pareto optimal solution set. The dynamic coupling constraint rules are used to dynamically adjust the constraints of the multi-objective optimization according to the target working conditions. The material selection module compares the experimental performance data obtained from experiments on candidate material data in the Pareto optimal solution set with the corresponding performance prediction vector to obtain the comparison deviation. If the comparison deviation is greater than a preset threshold, the target prediction model is updated based on the experimental performance data, and the step of obtaining candidate material data is executed again. If the comparison deviation is not greater than the preset threshold, the target material data is selected from the Pareto optimal solution set according to the preset weights of each optimization objective.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images