A large model-based high-voltage cable main insulation state evaluation method and system

Abstract

This invention discloses a method and system for assessing the main insulation condition of high-voltage cables based on a large model, belonging to the field of power equipment condition assessment technology. The method includes: collecting mechanical, physicochemical, and electrical multi-dimensional characteristic parameters of the cable to be assessed and its batch of comparative samples to form an original data matrix and recording the service life; obtaining a normalized matrix through extreme value normalization; calculating the initial weights of each characteristic parameter using the entropy weight method; determining the corrected bathtub curve stage of the cable based on the service life and normalized values; generating temporary weights through reinforcement learning iterative optimization starting from the initial weights, and obtaining expert weight adjustment suggestions from the large model at preset intervals, obtaining optimized weights after revising the suggestions; calculating the Euclidean distance and cosine similarity between the sample to be assessed and the positive and negative ideal solutions based on the optimized weights, and fusing them to obtain the relative closeness as the condition score. This invention significantly improves the accuracy and reliability of high-voltage cable insulation condition perception.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition assessment technology, and in particular to a method and system for assessing the condition of the main insulation of high-voltage cables based on a large model. Background Technology

[0002] As the core artery of the power transmission network, the reliability of the main insulation condition of high-voltage cables directly affects the safe and stable operation of the entire power grid. Traditional high-voltage cable condition assessment methods mainly rely on periodic power outage preventive tests, such as dielectric loss testing and partial discharge detection, or judgments based on a single statistical model. These methods have significant limitations: First, the assessment process is mostly static and intermittent, making it impossible to achieve continuous and dynamic perception and tracking of the insulation condition; second, the weights of various characteristic parameters (such as mechanical, electrical, and chemical indicators) in the assessment model largely depend on subjective settings based on expert experience, lacking an objective and adaptive adjustment mechanism, and making it difficult to accurately reflect the actual changes in the contribution of each characteristic parameter at different aging stages.

[0003] In recent years, large language models have demonstrated outstanding capabilities in natural language understanding and complex reasoning, providing a novel approach to encoding and integrating insulation aging expertise into assessment systems. Meanwhile, reinforcement learning, as a machine learning paradigm capable of continuous self-optimization through interaction with the environment, offers insights into the dynamic adjustment of assessment model parameters. However, how to deeply integrate the deep knowledge reasoning capabilities of large models and the online adaptive optimization capabilities of reinforcement learning with the aging patterns of cable insulation to construct an integrated framework that combines physical interpretability, dynamic adaptability, and high-precision condition assessment remains a critical technical challenge to be overcome in this field.

[0004] Therefore, developing a method and system that can fully utilize multi-source data and integrate domain knowledge to achieve accurate status assessment is of great theoretical and engineering significance for improving the intelligence level of high-voltage cable operation and maintenance and ensuring the safe and economical operation of the power grid. Summary of the Invention

[0005] The purpose of this invention is to overcome at least one technical problem existing in the prior art and to provide a method and system for evaluating the main insulation status of high-voltage cables based on a large model.

[0006] On one hand, this invention provides a method for evaluating the main insulation status of high-voltage cables based on a large model. The method includes: Step S1, collecting mechanical performance characteristic parameters, physicochemical performance characteristic parameters, and electrical performance characteristic parameters of the high-voltage cable sample to be evaluated and its batch of comparative samples to form an original data matrix, and recording the service life of the cable to be evaluated; Step S2, receiving the original data matrix, performing dimensionless processing using an extreme value normalization formula to obtain a normalized matrix; Step S3, calculating the objective weight values ​​of each characteristic parameter of the high-voltage cable sample to be evaluated using the entropy weight method based on the normalized matrix as the initial weights corresponding to each characteristic parameter; Step S4, according to the service life and the characteristic parameter values ​​of the high-voltage cable sample to be evaluated in the normalized matrix, determining the current stage of the modified bathtub curve of the cable as the current stage label of the high-voltage cable sample to be evaluated according to a preset stage division rule, wherein the modified bathtub curve stage includes early failure period, random failure period, and loss failure period; Step S5, using the initial weights as the initial state, and using a preset reward... The function is the optimization objective. The initial weights are iteratively updated using a reinforcement learning algorithm to generate a temporary optimized weight vector. The reward function is used to evaluate the degree of conformity between the evaluation result under the current weight and the current stage label. Step S6: According to the preset calling cycle, the current stage label, service life, normalized value of feature parameters, and temporary optimized weight vector are encapsulated into prompt words. The large model is called to obtain and parse the expert weight adjustment suggestions returned by the large model. Step S7: A weight correction amount is generated according to the expert weight adjustment suggestions. The weight correction amount is superimposed on the temporary optimized weight vector and re-normalized to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated. Step S8: The optimized weight vector is used to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive ideal solution and the negative ideal solution. Step S9: Based on the Euclidean distance and cosine similarity, the relative closeness is calculated using a preset fusion formula as a quantitative score of the main insulation state of the high-voltage cable to be evaluated.

[0007] Furthermore, the mechanical performance characteristics include elongation at break and / or tensile strength; the physicochemical performance characteristics include melting temperature, crystallinity, and / or carbonyl index; the electrical performance characteristics include power frequency dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric constant, 0.1Hz dielectric loss factor, the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor, and / or power frequency breakdown field strength; the characteristics are divided into positive and negative characteristics; the positive characteristics are indicators where a larger value represents a better cable insulation condition, including elongation at break, tensile strength, melting temperature, crystallinity, and power frequency breakdown field strength; the negative characteristics are indicators where a smaller value represents a better cable insulation condition, including carbonyl index, power frequency dielectric constant, 0.1Hz dielectric constant, power frequency dielectric loss factor, and 0.1Hz dielectric loss factor. Dielectric loss factor, the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor.

[0008] Furthermore, step S2 includes: ; ; ; In the formula, This represents the dimensionless value of the j-th feature parameter of the i-th sample in the positive feature parameters. Let j be the dimensionless value of the j-th feature parameter of the i-th sample in the negative feature parameters. Depend on and constitute, This indicates that the i-th sample corresponds to the original data value of the j-th feature parameter; m is the number of samples. and These are the maximum and minimum values ​​of the j-th feature parameter across all samples, respectively. It is the dimensionless value of the j-th feature parameter of the i-th sample after normalization.

[0009] Furthermore, step S3 includes: ; In the formula, This represents the information entropy corresponding to the j-th feature parameter; The objective weighting values ​​for each parameter are calculated based on the entropy value, including: ; In the formula, The entropy weight of the j-th feature parameter is used as the initial weight for each feature parameter, and n is the number of feature parameters.

[0010] Furthermore, the preset stage division rule in step S4 is as follows: based on the service life and the normalized value of the characteristic parameters, the cable status is divided into early failure period, random failure period and loss failure period through cluster analysis or threshold judgment; wherein, the early failure period corresponds to the initial service period, the random failure period corresponds to the stable index stage, and the loss failure period corresponds to the obvious deterioration of the index stage.

[0011] Furthermore, in step S5, the reinforcement learning algorithm employs the Q-learning algorithm. Its state space consists of the current weight vector, and its action space comprises the fine-tuning of the weights. The preset reward function consists of two parts, including: trend conformity. This is used to measure the consistency between the trend of the state score calculated under the current weight and the expected trend of the label of the current stage; the rationality of the weight distribution. This is used to penalize extreme cases where weights are excessively concentrated or dispersed. The reward function is: ,in and This is the harmonic coefficient.

[0012] Furthermore, the prompt in step S6 includes at least the following structured information: the role is set as a high-voltage cable insulation diagnosis expert, the current stage label, service life, normalized values ​​of each feature parameter, temporary optimized weight vector, and a clear instruction from the large model to provide weight adjustment suggestions based on the corrected bathtub curve theory; the generation of weight correction amount based on the expert's weight adjustment suggestions in step S7 includes: parsing the text suggestions returned by the large model, extracting the direction and magnitude of the increase or decrease of the weights of each feature parameter, and constructing the correction vector. Temporarily optimize the weight vector With correction vector The weights are superimposed and normalized to obtain the optimized weight vector.

[0013] Furthermore, step S8, which uses optimized weight vectors to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive and negative ideal solutions, includes the following formulas: ; ; ; ; in For the optimized weight vector, , , Let j be the dimensionless value of the j-th characteristic parameter in the high-voltage cable sample to be evaluated. The distance from the high-voltage cable sample to be evaluated to the ideal solution is... The distance from the high-voltage cable sample to be evaluated to the negative ideal solution; To evaluate the cosine similarity between the high-voltage cable sample to be evaluated and the positive ideal solution, The cosine similarity between the high-voltage cable sample to be evaluated and the negative ideal solution is given.

[0014] Furthermore, the preset fusion formula in step S9 is: ; in, The weighting factor (0≤ ≤1), S is the relative closeness. The calculated relative closeness S is used as the final quantitative score of the main insulation status of the high-voltage cable sample to be evaluated. The closer the score is to 1, the better the insulation status is, and the closer it is to 0, the worse the status is.

[0015] Secondly, embodiments of the present invention provide a high-voltage cable main insulation condition assessment system based on a large model, implemented using the aforementioned high-voltage cable main insulation condition assessment method based on a large model. The system includes: a feature parameter acquisition module, suitable for acquiring mechanical performance feature parameters, physicochemical performance feature parameters, and electrical performance feature parameters of the high-voltage cable sample to be assessed and its batch of comparative samples, forming an original data matrix, and recording the service life of the cable to be assessed; an initial weight calculation module, suitable for receiving the original data matrix, performing dimensionless processing using an extreme value normalization formula to obtain a normalized matrix; calculating the objective weight values ​​of each feature parameter of the high-voltage cable sample to be assessed using the entropy weight method based on the normalized matrix as the initial weights corresponding to each feature parameter; and a dynamic weight optimization module, suitable for determining the current stage of the modified bathtub curve of the cable as the current stage label of the high-voltage cable sample to be assessed according to a preset stage division rule, based on the service life and the feature parameter values ​​of the high-voltage cable sample to be assessed in the normalized matrix, wherein the modified bathtub curve stage includes the early failure period, the random failure period, and... During the loss and failure period; using the initial weights as the initial state and a preset reward function as the optimization objective, the initial weights are iteratively updated using a reinforcement learning algorithm to generate a temporary optimized weight vector; the reward function is used to evaluate the degree of conformity between the evaluation result under the current weights and the current stage label; according to a preset calling cycle, the current stage label, service life, normalized values ​​of feature parameters, and temporary optimized weight vector are encapsulated into prompt words, the large model is called, and the expert weight adjustment suggestions returned by the large model are obtained and parsed; a weight correction amount is generated according to the expert weight adjustment suggestions, and the weight correction amount is superimposed on the temporary optimized weight vector and re-normalized to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated; the state evaluation module is suitable for using the optimized weight vector to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive ideal solution and the negative ideal solution; based on the Euclidean distance and cosine similarity, a preset fusion formula is used to calculate the relative closeness as a quantitative score of the main insulation state of the high-voltage cable to be evaluated.

[0016] Thirdly, embodiments of the present invention also provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-described method for evaluating the main insulation status of high-voltage cables based on a large model.

[0017] Fourthly, embodiments of the present invention also provide a readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the above-described method for evaluating the main insulation status of high-voltage cables based on a large model.

[0018] The advantages of this invention compared to the prior art are as follows: 1. Dynamic Adaptive Evaluation: Breaking through the limitations of traditional static weights, this evaluation achieves dynamic adjustment of weights as the cable ages through reinforcement learning and large-scale model co-optimization, significantly improving the timeliness and accuracy of the evaluation.

[0019] 2. Multi-dimensional integrated perception: Integrating 11 characteristic parameters across three major categories—mechanical, physical, chemical, and electrical—to comprehensively reflect the insulation aging mechanism, avoid misjudgments caused by single indicators, and improve the robustness of the assessment.

[0020] 3. Strong physical interpretability: The modified bathtub curve theory is introduced to divide the aging stages, providing a physical basis for weight optimization and making the evaluation results have clear engineering interpretability.

[0021] 4. Knowledge + Data Dual-Driven Approach: The large model provides expert knowledge guidance, and reinforcement learning enables data-driven optimization. The two work together to avoid "knowledge solidification" and "local optima", achieving a 1+1>2 effect. Attached Figure Description

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Figure 1 This is a flowchart of a method for evaluating the main insulation status of high-voltage cables based on a large model, provided in Embodiment 1 of this application.

[0024] Figure 2 This is a schematic diagram of a high-voltage cable main insulation status assessment system based on a large model, provided in Embodiment 2 of this application.

[0025] Figure 3 This is a partial block diagram of the electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] To facilitate understanding, before detailing the specific implementation of this embodiment, the overall inventive concept of the present invention is described here: The present invention discloses a method and system for assessing the main insulation status of high-voltage cables based on a large model. Addressing the problems of static weights and subjective dependence in traditional cable assessment methods, the present invention proposes a collaborative optimization scheme integrating entropy weighting, modified bathtub curve theory, a large model, and reinforcement learning. First, multi-dimensional feature parameters (mechanical, physicochemical, and electrical) are collected, and initial weights are obtained using the entropy weighting method. Then, based on the service life and feature data, the modified bathtub curve stage (early stage, random stage, loss and failure stage) of the cable is determined. Iterative optimization of weights is performed primarily using reinforcement learning, while the large model is periodically invoked. Stage labels, measured data, and current weights are encapsulated as prompts to obtain expert-level adjustment suggestions and correct the weights. Finally, Euclidean distance and cosine similarity are used to calculate relative proximity and output a status score. This system achieves dynamic adaptive weighting, multi-dimensional comprehensive assessment, strong physical interpretability, and high assessment accuracy, significantly improving the accuracy of status perception and the reliability of life prediction, providing effective support for intelligent operation and maintenance of high-voltage cables.

[0028] Example 1

[0029] like Figure 1 The flowchart shown is a method for evaluating the main insulation status of high-voltage cables based on a large model, provided in this application embodiment.

[0030] As an example, the method includes: Step S1, collecting mechanical performance characteristic parameters, physicochemical performance characteristic parameters, and electrical performance characteristic parameters of the high-voltage cable sample to be evaluated and its batch of comparative samples to form an original data matrix, and recording the service life of the cable to be evaluated; Step S2, receiving the original data matrix, performing dimensionless processing using the extreme value normalization formula to obtain a normalized matrix; Step S3, calculating the objective weight values ​​of each characteristic parameter of the high-voltage cable sample to be evaluated using the entropy weight method based on the normalized matrix as the initial weights corresponding to each characteristic parameter; Step S4, according to the service life and the characteristic parameter values ​​of the high-voltage cable sample to be evaluated in the normalized matrix, determining the current stage of the modified bathtub curve of the cable as the current stage label of the high-voltage cable sample to be evaluated according to the preset stage division rules, wherein the modified bathtub curve stage includes the early failure period, the random failure period, and the loss failure period; Step S5, using the initial weights as the initial state, and using the preset reward function as the optimization objective, performing reinforcement learning calculations... The method iteratively updates the initial weights to generate a temporary optimized weight vector; the reward function is used to evaluate the degree of conformity between the evaluation result under the current weight and the current stage label; step S6: according to the preset calling cycle, the current stage label, service life, normalized value of feature parameters and temporary optimized weight vector are encapsulated into prompt words, the large model is called, and the expert weight adjustment suggestions returned by the large model are obtained and parsed; step S7: according to the expert weight adjustment suggestions, a weight correction amount is generated, and the weight correction amount is superimposed on the temporary optimized weight vector and re-normalized to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated; step S8: the optimized weight vector is used to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive ideal solution and the negative ideal solution; step S9: based on the Euclidean distance and cosine similarity, the relative closeness is calculated using a preset fusion formula as a quantitative score of the main insulation state of the high-voltage cable to be evaluated.

[0031] In some feasible implementations, the mechanical performance characteristics include elongation at break and / or tensile strength; the physicochemical performance characteristics include melting temperature, crystallinity, and / or carbonyl index; the electrical performance characteristics include power frequency dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric constant, 0.1Hz dielectric loss factor, the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor, and / or power frequency breakdown field strength; the characteristics are divided into positive and negative characteristics; the positive characteristics are indicators where a larger value represents a better cable insulation condition, including elongation at break, tensile strength, melting temperature, crystallinity, and power frequency breakdown field strength; the negative characteristics are indicators where a smaller value represents a better cable insulation condition, including carbonyl index, power frequency dielectric constant, 0.1Hz dielectric constant, power frequency dielectric loss factor, and 0.1Hz dielectric loss factor. Dielectric loss factor, the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor.

[0032] Preferably, the mechanical performance parameters are: elongation at break and tensile strength. These two parameters directly reflect the ductility and strength of the insulating material under mechanical stress and are important indicators for judging whether it has become embrittled or deteriorated. Physicochemical performance parameters are: melting temperature, crystallinity, and carbonyl index. Melting temperature and crystallinity reflect the microscopic thermodynamic state and crystal structure of the insulating material; the carbonyl index is a key indicator characterizing the degree of oxidative aging of polyolefin materials. Electrical performance parameters are: power frequency dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric constant, 0.1Hz dielectric loss factor, the ratio of 0.1Hz to power frequency dielectric loss factor, and power frequency breakdown field strength. These parameters reveal the dielectric and dielectric strength of the insulating material under the action of an electric field and are extremely sensitive to defects such as moisture dendrites and electrical aging. According to the indicative direction of each parameter on the quality of insulation, they can be classified as: positive indicators (the larger the value, the better the condition): elongation at break, tensile strength, melting temperature, crystallinity, and breakdown field strength. Negative indicators (the smaller the value, the better the condition): carbonyl index, power frequency dielectric constant, 0.1Hz dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric loss factor, 0.1Hz / power frequency dielectric loss factor ratio.

[0033] For a specific example, the raw data of 11 indicators of the high-voltage cable sample S0 (operating for 15 years) and 4 sets of historical comparison samples were collected, as shown in Table 1 below (all units conform to industry standards): Table 1:

[0034] Among them, the breaking elongation (j1), tensile strength (j2), melting temperature (j3), crystallinity (j4), carbonyl index (j5), power frequency dielectric constant (j6), power frequency dielectric loss factor (j7), 0.1Hz dielectric constant (j8), 0.1Hz dielectric loss factor (j9), dielectric loss ratio (j10), and power frequency breakdown field strength (j11).

[0035] In some feasible implementations, step S2 includes: ; ; ; In the formula, This represents the dimensionless value of the j-th feature parameter of the i-th sample in the positive feature parameters. Let j be the dimensionless value of the j-th feature parameter of the i-th sample in the negative feature parameters. Depend on and constitute, This indicates that the i-th sample corresponds to the original data value of the j-th feature parameter; m is the number of samples. and These are the maximum and minimum values ​​of the j-th feature parameter across all samples, respectively. It is the dimensionless value of the j-th feature parameter of the i-th sample after normalization.

[0036] For a specific example, taking sample S0 in Table 1 as an example, the maximum value of j1 (positive direction) is 420, and the minimum value is 250. Substituting these values ​​into the formula, we get: ; Taking sample S1 in Table 1 as an example, the maximum value of j5 (negative direction) is 1.05, and the minimum value is 0.28. Substituting these values ​​into the formula, we get: ; Similarly, the calculation method for the remaining samples is the same, and the final result is all... and It constitutes .

[0037] In some feasible implementations, step S3 includes: ; In the formula, This represents the information entropy corresponding to the j-th feature parameter; The objective weighting values ​​for each parameter are calculated based on the entropy value, including: ; In the formula, The entropy weight corresponding to the j-th feature parameter is used as the initial weight for each feature parameter, and n is the number of feature parameters (11 in this case).

[0038] For a specific example, suppose the initial weights calculated based on the sampled data in Table 1 above through steps S2 to S3 are as follows: =[0.1000(j1),0.0858(j2),0.0920(j3),0.0935(j4),0.1148(j5),0.0793 (j6),0.1070(j7),0.0817(j8),0.1110(j9),0.0633(j10),0.0717(j11)].

[0039] In some feasible implementations, the preset stage division rule in step S4 is as follows: based on the service life and the normalized value of the characteristic parameters, the cable status is divided into early failure period, random failure period and loss failure period by cluster analysis or threshold judgment; wherein, the early failure period corresponds to the initial service period, the random failure period corresponds to the stable index stage, and the loss failure period corresponds to the obvious deterioration of the index stage.

[0040] For a specific example, based on the cable's S0 operating history of 15 years (for XLPE cables, the design life is typically 30 years), and considering the data showing that indicators such as a carbonyl index of 0.65, a dielectric loss of 0.025 at 0.1Hz, and a breakdown field strength of 22kV / mm all indicate moderate aging, consistent with the characteristics of the loss failure period (late stage of the bathtub curve), subsequent optimization should increase the weighting of aging-sensitive indicators.

[0041] In some feasible implementations, the reinforcement learning algorithm in step S5 adopts the Q-learning algorithm, whose state space consists of the current weight vector, the action space is the fine-tuning amount of the weights, and the preset reward function consists of two parts, including: trend conformity. This is used to measure the consistency between the trend of the state score calculated under the current weight and the expected trend of the label of the current stage; the rationality of the weight distribution. This is used to penalize extreme cases where weights are excessively concentrated or dispersed. The reward function is: ,in and This is the harmonic coefficient.

[0042] In some feasible implementations, the prompt in step S6 includes at least the following structured information: the role is set as a high-voltage cable insulation diagnosis expert, the current stage label, service life, normalized values ​​of each feature parameter, temporary optimized weight vector, and a clear instruction from the large model to provide weight adjustment suggestions based on the modified bathtub curve theory; the generation of weight correction amount based on the expert weight adjustment suggestions in step S7 includes: parsing the text suggestions returned by the large model, extracting the direction and magnitude of the increase or decrease of the weights of each feature parameter, and constructing the correction vector. Temporarily optimize the weight vector With correction vector The weights are superimposed and normalized to obtain the optimized weight vector.

[0043] Preferably, combining steps S4 to S7, the initial weights obtained in step S3 are... Building upon this foundation, a collaborative mechanism between reinforcement learning and a large model is employed to dynamically optimize the weights of feature parameters, enabling them to adaptively adjust as the cable aging process changes. The specific process is as follows: First, a reinforcement learning model is constructed, whose state space is composed of the measured values ​​of the currently collected multi-dimensional feature parameters and the weight vector at the current time step. Taking the initial weights calculated in step S3 as the starting point, the reinforcement learning agent fine-tunes the current weights by executing actions. Each adjustment must ensure that the sum of the weights of all feature parameters is 1 to ensure the normalization constraint of the weights.

[0044] Secondly, design the reward function. This is used to evaluate the merits of each weight adjustment. The reward function consists of two parts: one is the trend conformity. That is, based on the modified bathtub curve theory, the current state score is compared with the theoretically expected trend. If the score change conforms to the pattern of "lower early failure period, stable random failure period, and decreased wear and tear failure period", a positive reward is given; secondly, the rationality of the weight distribution. To avoid extreme cases where any weight is too large or too small during optimization, thus maintaining the physical interpretability of the evaluation, the reinforcement learning agent iteratively optimizes the weights to maximize the cumulative reward. The total reward function is... ,in and This is the harmonic coefficient.

[0045] Simultaneously, a large model is introduced for collaborative guidance. The system periodically encapsulates current measured data, current weights, and the corrected bathtub curve theory into structured prompts, which are then input into the large model. The large model acts as a cable insulation assessment expert, providing suggestions for adjusting the current weights based on its built-in knowledge of aging mechanisms. For example, it indicates which feature parameters should be prioritized or the general direction for weight adjustment. These suggestions are transformed into prior knowledge, guiding the reinforcement learning agent's action exploration direction in subsequent cycles and preventing it from getting trapped in local optima.

[0046] Through iterative optimization of reinforcement learning and periodic guidance from a large model, the final output is a dynamically optimized weight that highly matches the current aging state of the cable. , ,in It is a dynamically adjusted amount (determined by reinforcement learning and the large model).

[0047] For a specific example, based on the fact that cable S0 has been in service for 15 years (design life of 30 years) and that the carbonyl index (0.5195) and dielectric loss (0.4333) in the normalized values ​​have significantly deviated from the optimal values, according to preset rules (e.g., service life > 12 years and key aging indicators below the threshold), cable S0 is determined to be in the loss failure period. This current stage is labeled "loss failure period".

[0048] The reinforcement learning agent simulates one iteration, starting with the initial weights: Current state: Initial weights; Action: The reinforcement learning agent fine-tunes the weights to generate new weights; Reward function: ,in and This is the harmonic coefficient.

[0049] After several iterations (e.g., 10 steps), a temporary optimized weight vector is obtained. Assume that the weights at this time are [0.09, 0.08, 0.08, 0.09, 0.13, 0.07, 0.11, 0.07, 0.12, 0.06, 0.10].

[0050] The call cycle T is set to 10 steps. The cycle condition has been met. The system constructs a prompt and calls the large model API: The system constructs the following prompt and sends it to the large model (e.g., GPT-4): Role: You are a senior high-voltage cable insulation diagnostic expert; Current stage: Wear and tear period (15 years of service); Measured data: (normalized values) Elongation at break: 0.4118, Tensile strength: 0.5195, Melting temperature: 0.5000, Crystallinity: 0.4667, Carbonyl index: 0.5195, Power frequency dielectric constant: 0.5714, Power frequency dielectric loss: 0.6667, 0.1Hz dielectric constant: 0.5000, 0.1Hz dielectric loss: 0.4333, 0.1Hz / power frequency dielectric loss ratio: 0.0194, Breakdown field strength: 0.4583; Current temporary weights: [0.09, 0.08, 0.08, 0.09, 0.13, 0.07, 0.11, 0.07, 0.12, 0.06, 0.10]; Based on the characteristics of the wear and tear period, analyze whether the current weights are reasonable and provide specific adjustment suggestions (direction and magnitude of increase or decrease). The sum of all weights must remain at 1.

[0051] Example of response from a large model: Recommendations: The carbonyl index is a key indicator of oxidative aging. The current value of 0.5195 is already too high, and its weight should be further increased to 0.15 (+0.02). The dielectric loss at 0.1Hz reflects water tree aging, and it is recommended to increase it from 0.12 to 0.14 (+0.02). The dielectric loss ratio at 0.1Hz / power frequency is extremely low (0.0194), indicating severe water treeing, and it is recommended to increase it from 0.06 to 0.09 (+0.03). The breakdown field strength has decreased significantly, and it is recommended to keep it unchanged at 0.10. Mechanical properties contribute little during the wear period and can be appropriately reduced: elongation at break should be reduced to 0.06 (-0.03), tensile strength to 0.06 (-0.02), melting temperature to 0.07 (-0.01), and crystallinity to 0.07 (-0.02). The power frequency dielectric constant, power frequency dielectric loss, and 0.1Hz dielectric constant can be slightly reduced: to 0.06 (-0.01), 0.10 (-0.01), and 0.06 (-0.01), respectively. The total after adjustment is 1.00.

[0052] Analysis of large model recommendations, construction of dynamic adjustment quantities: =[-0.03,-0.02,-0.01,-0.02,+0.02,-0.01,-0.01,-0.01,+0.02,+0.03,0].

[0053] Obtain the optimized weight vector : =[0.06,0.06,0.07,0.07,0.15,0.06,0.10,0.06,0.14,0.09,0.10].

[0054] In some feasible implementations, step S8 uses optimized weight vectors to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive and negative ideal solutions. The calculation formulas include: ; ; ; ; in For the optimized weight vector, , , Let j be the dimensionless value of the j-th characteristic parameter in the high-voltage cable sample to be evaluated. The distance from the high-voltage cable sample to be evaluated to the ideal solution is... The distance from the high-voltage cable sample to be evaluated to the negative ideal solution; To evaluate the cosine similarity between the high-voltage cable sample to be evaluated and the positive ideal solution, The cosine similarity between the high-voltage cable sample to be evaluated and the negative ideal solution is given.

[0055] In some feasible implementations, the preset fusion formula in step S9 is: ; in, The weighting factor (0≤ ≤1 (preferably 0.5 here), S is the relative closeness. The calculated relative closeness S is used as the final quantitative score of the main insulation status of the high-voltage cable sample to be evaluated. The closer the score is to 1, the better the insulation status, and the closer it is to 0, the worse the status.

[0056] Preferably, the positive ideal solution is the normalized maximum value of each characteristic parameter, and the negative ideal solution is the normalized minimum value of each characteristic parameter.

[0057] The above implementation method addresses industry pain points in traditional high-voltage cable main insulation condition assessment, such as poor adaptability of static weights, large subjective weighting bias, coarse aging stage judgment, and low degree of quantitative assessment. It integrates four core technologies: data-driven objective weighting, staged judgment based on corrected bathtub curves, large-scale model + reinforcement learning closed-loop dynamic weight optimization, and multi-dimensional distance-similarity fusion quantitative scoring. This forms a fully automated cable insulation condition assessment solution that provides objective and accurate results and aligns with engineering realities. Compared to existing technologies, it achieves significant improvements in assessment objectivity, dynamic adaptability, quantitative accuracy, and engineering applicability. Simultaneously, this method provides a scientific and practical decision-making basis for the lean operation and maintenance of the power grid, specifically as follows: This method standardizes the raw data through extreme value normalization in step S2, and then objectively calculates the initial weights based on the dispersion of the indicator data using the entropy weight method in step S3. The entire process relies on measured mechanical, physicochemical, and electrical parameters across multiple dimensions, completely abandoning the subjective weighting model of experts in traditional methods. This avoids the distortion of evaluation results caused by human experience bias and differences in industry perception, ensuring that the weight allocation of each characteristic parameter is highly matched to its own distinguishability of cable aging status, thus guaranteeing the objectivity and fairness of the evaluation from the root. Step S4 of this method combines the cable's service life with the normalized characteristic parameter values, determining the cable aging stage and assigning stage labels according to the stage division rules of the modified bathtub curve. This breaks through the crude model of traditional methods that rely solely on the single indicator of service life to determine the aging stage, achieving precise division of early failure period, random failure period, and loss failure period. This stage label sets a clear direction for subsequent dynamic weight optimization, ensuring that weight adjustments always align with the current physical characteristics of cable aging, avoiding aimless weight optimization, and improving the relevance of subsequent evaluations. This method constructs a closed-loop weight optimization mechanism through steps S5-S7, combining reinforcement learning-based autonomous iteration with expert knowledge guidance from a large model. Starting with initial weights, it iteratively generates temporary optimized weights in a static environment composed of measured data using reinforcement learning. Every T iterations, the large model is invoked to obtain professional weight adjustment suggestions, correcting and normalizing the temporary weights. This mechanism ensures that the final optimized weights possess both data adaptability and professional scientific rigor: it iterates through reinforcement learning to match the measured data characteristics of the cable being evaluated, while expert advice from the large model prevents reinforcement learning from getting stuck in a data black box or local optima. This achieves personalized and dynamic weight adjustment according to the cable's aging stage, completely solving the industry pain point that traditional static fixed weights cannot adapt to the core characteristics of different aging stages of cables, ensuring that weight allocation is highly matched to the current insulation state of the cable. Steps S8-S9 of this method abandon the traditional evaluation mode that uses only distance or similarity methods. Instead, it calculates Euclidean distance (representing the absolute numerical difference between the measured data and the ideal solution) and cosine similarity (representing the trend consistency between the measured data and the ideal solution) through the optimized weight vector. Then, it calculates the relative closeness as a quantitative score of the insulation state through a preset fusion formula.This integrated approach considers both numerical differences and trend matching, offering a more comprehensive and accurate assessment compared to single-method evaluations. Furthermore, the quantitative scoring ranges from 0 to 1, with values ​​positively correlated with insulation condition, allowing maintenance personnel to directly and intuitively determine cable insulation status levels based on the scores. This addresses the issues of traditional assessment methods, which rely heavily on qualitative descriptions, have low quantification levels, and hinder rapid decision-making by maintenance personnel. The mechanical, physicochemical, and electrical characteristic parameters collected by this method are all standard testing indicators in the power industry, requiring no additional dedicated testing equipment and directly connecting to existing power grid testing instruments and data acquisition platforms. Moreover, the large-scale model, implemented through application programming interfaces (APIs), reinforcement learning iterative optimization, and end-to-end data processing, is all software-based, eliminating the need for hardware modifications to the existing power grid maintenance system and minimizing transformation costs. The entire process is automated, from data acquisition and weight calculation to status scoring, requiring minimal human intervention, resulting in high testing efficiency. This allows for rapid large-scale deployment and application across power grid companies at all levels, adapting to the maintenance and assessment needs of the vast number of cables in the power grid. Furthermore, the quantitative status score output by this method enables power grid maintenance personnel to formulate differentiated maintenance strategies for cables at different aging stages: for example, focusing on mechanical damage and manufacturing / installation defects for cables in the early stages of failure; implementing routine monitoring for cables in the random failure stage; and strengthening the retesting of oxidation / electrical aging indicators and formulating replacement plans in advance for cables in the loss failure stage. This differentiated maintenance model avoids the problems of blind repair, over-maintenance, or lack of testing and sudden failures in traditional maintenance, effectively reducing power grid maintenance costs, while avoiding the risks of power outages and equipment damage caused by cable insulation deterioration in advance, significantly improving the safety, stability, and economy of high-voltage power grid operation.

[0058] Example 2

[0059] Please see Figure 2 This embodiment provides a structural diagram of a high-voltage cable main insulation condition assessment system based on a large model.

[0060] As an example, the control system is implemented using the high-voltage cable main insulation status assessment method based on a large model as described in Example 1. The system includes: The feature parameter acquisition module 210 is suitable for acquiring the mechanical performance feature parameters, physicochemical performance feature parameters, and electrical performance feature parameters of the high-voltage cable sample to be evaluated and its batch of comparative samples, forming an original data matrix, and recording the service life of the cable to be evaluated.

[0061] The initial weight calculation module 220 is adapted to receive the original data matrix, perform dimensionless processing using the extreme value normalization formula, and obtain a normalized matrix; based on the normalized matrix, the entropy weight method is applied to calculate the objective weight values ​​of each feature parameter of the high-voltage cable sample to be evaluated as the initial weights corresponding to each feature parameter.

[0062] The dynamic weight optimization module 230 is suitable for determining the current stage of the modified bathtub curve of the high-voltage cable sample to be evaluated as the current stage label of the sample, based on the service life and the feature parameter values ​​of the high-voltage cable sample to be evaluated in the normalized matrix, according to a preset stage division rule. The modified bathtub curve stage includes early failure period, random failure period, and loss failure period. Using the initial weight as the initial state and a preset reward function as the optimization objective, the initial weight is iteratively updated using a reinforcement learning algorithm to generate a temporary optimized weight vector. The reward function is used to evaluate the degree of conformity between the evaluation result under the current weight and the current stage label. According to a preset calling cycle, the current stage label, service life, normalized feature parameter values, and temporary optimized weight vector are encapsulated as prompt words. The large model is called to obtain and parse the expert weight adjustment suggestions returned by the large model. A weight correction amount is generated based on the expert weight adjustment suggestions, and the weight correction amount is superimposed on the temporary optimized weight vector and re-normalized to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated.

[0063] The state assessment module 240 is suitable for calculating the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be assessed and the reference points representing the positive ideal solution and the negative ideal solution using the optimized weight vector; based on the Euclidean distance and cosine similarity, the relative closeness is calculated using a preset fusion formula as a quantitative score of the main insulation state of the high-voltage cable to be assessed.

[0064] It is not difficult to see that this embodiment is a system implementation corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0065] It is worth mentioning that all modules involved in this embodiment are logical units. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0066] Example 3

[0067] Please see Figure 3The present invention also provides an electronic device, including: a memory and a processor; the memory stores at least one program instruction; the processor loads and executes the at least one program instruction to implement the high-voltage cable main insulation status assessment method based on a large model provided in Embodiment 1.

[0068] The memory 702 and processor 701 are connected via a bus, which may include any number of interconnecting buses and bridges, connecting various circuits of one or more processors 701 and memory 702 together. The bus may also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 701 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 701.

[0069] Processor 701 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 702 can be used to store data used by processor 701 during operation.

[0070] Example 4

[0071] This invention also proposes a storage medium storing a high-voltage cable main insulation condition assessment method based on a large model. When the high-voltage cable main insulation condition assessment program based on the large model is executed by a processor, it implements the steps of the high-voltage cable main insulation condition assessment method based on the large model described above. Since this storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0072] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A large model-based high-voltage cable main insulation state evaluation method, characterized by, The method includes: Step S1: Collect the mechanical performance characteristics, physicochemical performance characteristics, and electrical performance characteristics of the high-voltage cable sample to be evaluated and the comparative samples of the same batch to form an original data matrix, and record the service life of the cable to be evaluated. Step S2: Receive the original data matrix and perform dimensionless processing using the extreme value normalization formula to obtain the normalized matrix; Step S3: Based on the normalized matrix, apply the entropy weight method to calculate the objective weight values ​​of each feature parameter of the high-voltage cable sample to be evaluated as the initial weights corresponding to each feature parameter. Step S4: Based on the service life and the characteristic parameter values ​​of the high-voltage cable sample to be evaluated in the normalized matrix, and according to the preset stage division rules, determine the current stage of the modified bathtub curve of the cable as the current stage label of the high-voltage cable sample to be evaluated. The modified bathtub curve stage includes the early failure period, the random failure period, and the loss failure period. Step S5: Using the initial weights as the initial state and a preset reward function as the optimization objective, the initial weights are iteratively updated using a reinforcement learning algorithm to generate a temporary optimized weight vector; the reward function is used to evaluate the degree of consistency between the evaluation result under the current weights and the label of the current stage. Step S6: According to the preset calling cycle, encapsulate the current stage label, service years, normalized value of feature parameters and temporary optimized weight vector into prompt words, call the large model, and obtain and parse the expert weight adjustment suggestions returned by the large model; Step S7: Generate a weight correction amount based on the expert weight adjustment suggestion, and superimpose the weight correction amount onto the temporary optimized weight vector and re-normalize it to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated. Step S8: Calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive and negative ideal solutions using the optimized weight vector; Step S9: Based on the Euclidean distance and cosine similarity, the relative closeness is calculated using a preset fusion formula, which serves as a quantitative score for the main insulation status of the high-voltage cable to be evaluated.

2. The large model-based high-voltage cable main insulation state evaluation method according to claim 1, characterized by, The mechanical property characteristics include elongation at break and / or tensile strength; the physicochemical property characteristics include melting temperature, crystallinity and / or carbonyl index; the electrical property characteristics include power frequency dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric constant, 0.1Hz dielectric loss factor, the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor and / or power frequency breakdown field strength; The characteristic parameters are divided into positive and negative characteristic parameters according to their positive and negative directions. The positive characteristic parameters are indicators that the larger the value, the better the cable insulation condition, including elongation at break, tensile strength, melting temperature, crystallinity, and power frequency breakdown field strength. The negative characteristic parameters are indicators that the smaller the value, the better the cable insulation condition, including carbonyl index, power frequency dielectric constant, 0.1Hz dielectric constant, power frequency dielectric loss factor, 0.1Hz dielectric loss factor, and the ratio of 0.1Hz dielectric loss factor to power frequency dielectric loss factor.

3. The large model-based high-voltage cable main insulation condition assessment method according to claim 2, characterized in that, Step S2 includes: ； ； ； In the formula, This represents the dimensionless value of the j-th feature parameter of the i-th sample in the positive feature parameters. Let j be the dimensionless value of the j-th feature parameter of the i-th sample in the negative feature parameters. Depend on and constitute, This indicates that the i-th sample corresponds to the original data value of the j-th feature parameter; m is the number of samples. and These are the maximum and minimum values ​​of the j-th feature parameter across all samples, respectively. It is the dimensionless value of the j-th feature parameter of the i-th sample after normalization.

4. The method for evaluating the main insulation condition of high-voltage cables based on a large model according to claim 3, characterized in that, Step S3 includes: ； In the formula, This represents the information entropy corresponding to the j-th feature parameter; The objective weighting values ​​for each parameter are calculated based on the entropy value, including: ； In the formula, The entropy weight of the j-th feature parameter is used as the initial weight for each feature parameter, and n is the number of feature parameters.

5. The method for assessing the main insulation condition of high-voltage cables based on a large model according to claim 1, characterized in that, The preset stage division rule in step S4 is as follows: based on the service life and the normalized value of the characteristic parameters, the cable status is divided into early failure period, random failure period and loss failure period through cluster analysis or threshold judgment; wherein, the early failure period corresponds to the initial service period, the random failure period corresponds to the stable index stage, and the loss failure period corresponds to the obvious deterioration of the index stage.

6. The method for assessing the main insulation condition of high-voltage cables based on a large model according to claim 1, characterized in that, In step S5, the reinforcement learning algorithm uses the Q-learning algorithm. Its state space consists of the current weight vector, and its action space consists of the fine-tuning of the weights. The preset reward function consists of two parts, including: trend conformity. This is used to measure the consistency between the trend of the state score calculated under the current weight and the expected trend of the label of the current stage; the rationality of the weight distribution. This is used to penalize extreme cases where weights are excessively concentrated or dispersed. The reward function is: ,in and This is the harmonic coefficient.

7. The method for assessing the main insulation condition of high-voltage cables based on a large model according to claim 6, characterized in that, The prompt in step S6 includes at least the following structured information: the role is set as a high-voltage cable insulation diagnosis expert, the current stage label, service life, normalized values ​​of each feature parameter, temporary optimization weight vector, and a clear instruction based on the corrected bathtub curve theory that the large model should give a weight adjustment suggestion. Step S7, generating the weight correction amount based on the expert weight adjustment suggestions, includes: parsing the text suggestions returned by the large model, extracting the direction and magnitude of the increase or decrease of the weights of each feature parameter, and constructing the correction vector. The temporary optimization weight vector will be used. With correction vector The weights are superimposed and normalized to obtain the optimized weight vector.

8. The method for evaluating the main insulation condition of high-voltage cables based on a large model according to claim 3, characterized in that, Step S8 uses the optimized weight vector to calculate the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be evaluated and the reference points representing the positive and negative ideal solutions. include: ； ； ； ； in For the optimized weight vector, , , Let j be the dimensionless value of the j-th characteristic parameter in the high-voltage cable sample to be evaluated. The distance from the high-voltage cable sample to be evaluated to the ideal solution is... The distance from the high-voltage cable sample to be evaluated to the negative ideal solution; To evaluate the cosine similarity between the high-voltage cable sample to be evaluated and the positive ideal solution, The cosine similarity between the high-voltage cable sample to be evaluated and the negative ideal solution is given.

9. The method for assessing the main insulation condition of high-voltage cables based on a large model according to claim 8, characterized in that, The preset fusion formula in step S9 is: ； in, The weighting coefficient (0≤ ≤1), S is the relative closeness. The calculated relative closeness S is used as the final quantitative score of the main insulation status of the high-voltage cable sample to be evaluated. The closer the score is to 1, the better the insulation status is, and the closer it is to 0, the worse the status is.

10. A high-voltage cable main insulation condition assessment system based on a large model, implemented using the high-voltage cable main insulation condition assessment method based on a large model as described in any one of claims 1-9, characterized in that, The system includes: The feature parameter acquisition module is suitable for collecting mechanical performance feature parameters, physicochemical performance feature parameters, and electrical performance feature parameters of the high-voltage cable sample to be evaluated and its batch of comparative samples, forming an original data matrix, and recording the service life of the cable to be evaluated. The initial weight calculation module is adapted to receive the original data matrix, perform dimensionless processing using the extreme value normalization formula to obtain a normalized matrix; and apply the entropy weight method to calculate the objective weight values ​​of each feature parameter of the high-voltage cable sample to be evaluated as the initial weights corresponding to each feature parameter based on the normalized matrix. The dynamic weight optimization module is applicable to determining the current stage of the modified bathtub curve of the high-voltage cable sample to be evaluated, based on the service life and the feature parameter values ​​of the sample in the normalized matrix, according to a preset stage division rule. This stage includes early failure period, random failure period, and loss failure period. Using the initial weights as the initial state and a preset reward function as the optimization objective, the module iteratively updates the initial weights using a reinforcement learning algorithm to generate a temporary optimized weight vector. The reward function is used to evaluate the degree of conformity between the evaluation result under the current weights and the current stage label. According to a preset calling cycle, the module encapsulates the current stage label, service life, normalized feature parameter values, and temporary optimized weight vector into prompt words, calls the large model, obtains and parses the expert weight adjustment suggestions returned by the large model, generates a weight correction amount based on the expert weight adjustment suggestions, and superimposes the weight correction amount onto the temporary optimized weight vector and re-normalizes it to obtain the optimized weight vector corresponding to each feature parameter in the high-voltage cable sample to be evaluated. The state assessment module is suitable for calculating the Euclidean distance and cosine similarity between the actual state data points of the high-voltage cable sample to be assessed and the reference points representing the positive and negative ideal solutions using optimized weight vectors; based on the Euclidean distance and cosine similarity, the relative closeness is calculated using a preset fusion formula as a quantitative score of the main insulation state of the high-voltage cable to be assessed.

Images