A method and system for predicting the life of an underground cable based on a mixture model

Abstract

The application discloses a kind of underground cable life prediction method and system based on mixed model, method includes following steps: S1: the historical detection data set of underground cable is preprocessed, obtains associated feature set, associated feature set is divided into training set and test set;S2: using Weibull distribution theory and training set constructs Weibull distribution prediction model, using BP neural network theory and training set constructs life prediction model;S3: according to probability distribution function and test set, the weight coefficient proportioning of Weibull distribution prediction model and life prediction model is calculated;S4: according to Weibull distribution prediction model, life prediction model and weight coefficient proportioning, construct fusion model, obtain current underground cable detection data, utilize fusion model and output underground cable life prediction data.The beneficial effects of the application: improve the accuracy and stability of underground cable life prediction, while reducing the demand for professional knowledge of model construction, reduce the difficulty of construction.

Description

Technical Field

[0001] This application relates to the field of underground cable life prediction technology, and in particular to an underground cable life prediction method and system based on a hybrid model. Background Technology

[0002] Cable lifespan testing is of great significance for ensuring the safe operation of the power grid system. Currently, the main testing methods are withstand voltage testing, physical model-based lifespan prediction, and data-driven lifespan prediction: (1) The withstand voltage testing method is based on cable usage regulations, conducting withstand voltage tests on cables to determine if they pass the test and can continue to function. Cables that fail the test are considered to have reached the end of their lifespan. (2) The physical model-based lifespan prediction method establishes a physical model through experiments, fits it with mathematical formulas, and establishes a relationship between aging parameters and cable lifespan to characterize the cable lifespan. (3) The data-driven lifespan prediction method uses big data mining and statistical analysis methods, employing machine learning or deep learning-based methods to continuously train and establish a prediction model from a large amount of aging data.

[0003] However, (1) the withstand voltage test has a long testing cycle, the cable testing process is a loss-based test, and can only obtain two rough test results: whether it can continue to operate or not. The assessment of cable life is not in-depth enough, and a relatively accurate lifespan cannot be determined, which increases the difficulty of operation and maintenance. (2) The physical model-based method requires a lot of existing system knowledge, and the development of complex equipment is expensive and time-consuming. Some cables use online monitoring devices to observe the cable operation status. The devices are expensive. When using the physical model-based method, expert experience is required to make judgments. The process of building a physical model requires the selection of data through professional knowledge. (3) Among the data-driven methods, the deep learning-based method can use different data analysis methods to analyze the data and deeply mine the data information to obtain feature information, and then build a model based on the feature information.

[0004] One commonly used method is the lifetime prediction method based on artificial neural networks. However, it has two main drawbacks: first, the initial solution is randomly generated, and the quality of the initial solution greatly affects the algorithm's performance, thus introducing instability; second, the algorithm uses gradient descent to determine the search direction, and because the search space contains flat regions and multiple local minima, this method is prone to slow convergence or failing to converge due to getting trapped in local minima.

[0005] Chinese patent "Method, Apparatus, and Non-volatile Storage Medium for Predicting Cable Lifespan", publication number CN115495985A, publication date: December 20, 2022, specifically discloses a method for measuring the partial discharge signal of a target cable; inputting the partial discharge signal into a pre-trained aging time prediction model, outputting the aging time of the target cable, whereby the aging time characterizes the current degree of aging of the target cable, and the aging time prediction model includes a convolutional neural network; obtaining the first service life of the target cable; and calculating the remaining service life of the target cable based on the first service life and the aging time. This method uses a convolutional neural network to predict the remaining service life of the cable, which suffers from the instability of artificial neural networks, resulting in low accuracy and stability.

[0006] Chinese patent "A Method and Device for Predicting the Remaining Life of Tunnel Cables Based on Machine Learning," publication number CN116861797A, published on October 10, 2023, specifically discloses a method for predicting the remaining life of tunnel cables, including statistical classification, power transmission performance monitoring, insulation material performance monitoring, synthetic feature combination construction and analysis, and cable remaining life prediction. This method uses machine learning to predict the remaining life of tunnel cables, learning the complex relationships between cable life and various features from a large amount of historical data, thus achieving more accurate predictions. It can better capture the nonlinear relationship of cable life and the interaction between features. Simultaneously, by monitoring the degree of loss at various parts of the cable, it uses the results of the loss at different parts for maintenance optimization and decision support. However, it still suffers from the instability of artificial neural networks and the tendency for slow convergence. Summary of the Invention

[0007] This application addresses the problems of instability and slow convergence in existing technologies that mostly use artificial neural networks to predict cable life. It provides a hybrid model-based method and system for predicting the life of underground cables. This method constructs a Weibull distribution prediction model based on a physical model and a data-driven life prediction model, respectively. The weighting coefficients of the two models are calculated based on the probability distribution function to construct a fusion model. This balances the high demand for expertise in the physical model with the instability of the artificial neural network model. While improving the accuracy and stability of underground cable life prediction, it reduces the need for specialized expertise in model construction, lowers the construction difficulty, and thus expands the scope of application.

[0008] To achieve the above technical objectives, this application provides a technical solution: a method for predicting the lifespan of underground cables based on a hybrid model, comprising the following steps: S1: Preprocessing the historical inspection dataset of underground cables to obtain a set of associated features, and dividing the set of associated features into a training set and a test set; S2: Constructing a Weibull distribution prediction model using Weibull distribution theory and the training set, and constructing a lifespan prediction model using BP neural network theory and the training set; S3: Calculating the weight coefficient ratio of the Weibull distribution prediction model and the lifespan prediction model based on the probability distribution function and the test set; S4: Constructing a fusion model based on the Weibull distribution prediction model, the lifespan prediction model, and the weight coefficient ratio, obtaining current underground cable inspection data, and outputting underground cable lifespan prediction data using the fusion model.

[0009] Furthermore, S1 includes: using the Pearson correlation coefficient to obtain the life-related correlation features in the historical detection dataset of underground cables; standardizing the numerical variables corresponding to the correlation features and performing one-hot encoding on the typological variables corresponding to the correlation features, using the standardized numerical variables and the one-hot encoded typological variables as the correlation feature set; and dividing the correlation feature set into a training set and a test set.

[0010] Furthermore, the standardization process for the numerical variables corresponding to the associated features includes: standardizing the numerical variables based on their mean and variance.

[0011] Furthermore, the construction of the Weibull distribution prediction model using Weibull distribution theory and training set includes: constructing the initial structure of the Weibull distribution prediction model based on the probability density function of the Weibull distribution in Weibull distribution theory; and training the initial structure of the Weibull distribution prediction model using the training set to obtain the Weibull distribution prediction model.

[0012] Furthermore, the probability density function of the Weibull distribution is: Where x is the cable failure time, λ is the scale parameter, representing the failure time corresponding to a failure probability of 63.2%, and k is the shape parameter, representing the dispersion of the data.

[0013] Furthermore, the construction of the lifetime prediction model using BP neural network theory and training set includes: constructing an initial structure of the lifetime prediction model based on BP neural network theory with a BP neural network structure that includes at least an input layer, a hidden layer, and an output layer; and training the initial structure of the lifetime prediction model using the training set to obtain the lifetime prediction model.

[0014] Furthermore, the step of calculating the weight coefficient ratio of the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function and the test set includes: constructing a first remaining lifetime using the Weibull distribution prediction model and the probability distribution function, and presetting a first weight; constructing a second remaining lifetime using the lifetime prediction model and the probability distribution function, and presetting a second weight; constructing a third remaining lifetime using the test set and the probability distribution function; calculating the sum of the product of the first remaining lifetime and the first weight, and the product of the second remaining lifetime and the second weight; determining whether the ratio of the third remaining lifetime to the sum is 1; if so, using the ratio of the first weight to the second weight as the weight coefficient ratio; if not, updating the first weight and the second weight.

[0015] Furthermore, S5 also includes: validating the fusion model with a test set, outputting the optimal solution for the weight coefficient ratio, and updating the fusion model based on the optimal solution for the weight coefficient ratio.

[0016] Furthermore, the historical inspection dataset of the underground cable includes at least: cable number, years of operation, partial discharge value, visual defects, neutral corrosion value, overload conditions, and health index.

[0017] Another technical solution provided in this application is a hybrid model-based underground cable life prediction system for implementing the above-mentioned method, comprising: a data acquisition and processing model for acquiring and preprocessing underground cable inspection data; a physical model construction module for constructing a Weibull distribution prediction model based on the data output by the data acquisition and processing model; a data-driven model construction module for constructing a life prediction model based on the data output by the data acquisition and processing model; a weight coefficient allocation module for calculating the weight coefficient allocation of the Weibull distribution prediction model and the life prediction model based on the probability distribution function; a fusion construction module for outputting a fusion model based on the weight coefficient allocation, the Weibull distribution prediction model, and the life prediction model; and a prediction analysis module for outputting underground cable life prediction data using the fusion model and the current underground cable inspection data.

[0018] The beneficial effects of this application are as follows: a Weibull distribution prediction model is constructed using Weibull distribution theory, a lifetime prediction model is constructed using BP neural network theory, and the weight coefficient ratio of the two prediction models is calculated using the probability distribution function to construct a fusion model. This model combines the physical model method based on Weibull distribution and the data-driven method based on BP neural network to predict the lifetime of underground cables. This not only makes up for the instability caused by the random generation of initial solutions and the defects of convergence errors in artificial neural network models, but also compensates for the errors caused by the reliance of physical models on expert experience. This improves the accuracy of prediction results while reducing the requirements for model construction and expanding the applicability of the model. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the underground cable lifetime prediction method based on a hybrid model proposed in this application.

[0020] Figure 2 This is a schematic diagram of the historical detection dataset of underground cables for the underground cable life prediction method based on a hybrid model proposed in this application.

[0021] Figure 3 This is a heatmap showing the correlation analysis of a hybrid model-based method for predicting the lifespan of underground cables according to this application.

[0022] Figure 4 This is a schematic diagram of the BP neural network structure of the underground cable lifetime prediction method based on a hybrid model proposed in this application.

[0023] Figure 5 This diagram illustrates a comparison of the output results of the fusion model for the underground cable lifetime prediction method based on a hybrid model proposed in this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] like Figure 1 As shown in the first embodiment of this application, a method for predicting the lifetime of underground cables based on a hybrid model includes the following steps: S1: Preprocess the historical detection dataset of underground cables to obtain the associated feature set, and divide the associated feature set into a training set and a test set; S2: Construct a Weibull distribution prediction model using Weibull distribution theory and training set, and construct a lifetime prediction model using BP neural network theory and training set; S3: Calculate the weighting ratios of the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function and the test set; S4: Construct a fusion model based on the Weibull distribution prediction model, the life prediction model, and the weighting coefficient ratio, obtain the current underground cable inspection data, and use the fusion model to output the underground cable life prediction data.

[0026] In this embodiment, a Weibull distribution prediction model is constructed using Weibull distribution theory, and a lifetime prediction model is constructed using BP neural network theory. The weight coefficient ratio of the two prediction models is calculated using the probability distribution function to construct a fusion model. Combining the physical model method based on Weibull distribution and the data-driven method based on BP neural network to predict the lifetime of underground cables, this approach can compensate for the instability caused by the random generation of initial solutions and the defects of convergence errors in artificial neural network models, as well as the errors caused by the reliance on expert experience in physical models. This improves the accuracy of prediction results while reducing the requirements for model construction and expanding the applicability of the model.

[0027] Specifically, step S1 includes: Pearson correlation coefficient was used to obtain life-related features in historical inspection data of underground cables. The numerical variables corresponding to the association features are standardized, and the typological variables corresponding to the association features are one-hot encoded. The standardized numerical variables and the one-hot encoded typological variables are used as the association feature set. The associated feature set is divided into a training set and a test set.

[0028] The Pearson correlation coefficient is used to measure the correlation (linear correlation) between two variables X and Y, and its value ranges between -1 and 1. ρ X，Y Let be the correlation coefficient between X and Y. When {|ρ}_{X,Y}| ≥ 0.8, the two variables are considered highly correlated; when 0.5 ≤ {|ρ}_{X,Y}| < 0.8, the two variables are considered moderately correlated; when 0.3 ≤ {|ρ}_{X,Y}| < 0.5, the two variables are considered poorly correlated; and when {|ρ}_{X,Y}| < 0.3, the two variables are considered uncorrelated. The correlation between various types of data in the historical underground cable monitoring dataset is calculated based on the Pearson correlation coefficient. Where, ρ X，Y Let X be the correlation coefficient between X and Y, where X is the first type of data used for correlation measurement, and Y is the second type of data used for correlation measurement. i Let y be the i-th sample point in X. i Let be the i-th sample point in Y.

[0029] like Figure 2As shown, in this embodiment, the historical inspection dataset for underground cables is a publicly available dataset – 20kV underground cable inspection dataset, containing at least seven types of data: cable number, service life, partial discharge value, visible defects, neutral corrosion value, overload conditions, and health index. Service life is related to the lifespan of the underground cable; that is, the remaining lifespan of the underground cable = normal operating lifespan of the underground cable - service life. Therefore, the value of service life is defined as the lifespan of the underground cable. Partial discharge value, visible defects, neutral corrosion value, overload conditions, and health index are aging characteristic quantities of the underground cable under the corresponding lifespan. These seven types of data can be collected visually or using professional inspection equipment during the actual operation of the underground cable.

[0030] like Figure 3 As shown, the Pearson correlation heatmap analysis, constructed using the Pearson correlation coefficient to calculate the correlation between seven types of data, reveals a strong positive correlation between partial discharge value, neutral corrosion value, and service life; a strong negative correlation between health index and service life; and a weak correlation between cable number, maximum current carrying capacity, and service life. Cable number and maximum current carrying capacity were then removed, and partial discharge value, visualized defects, neutral corrosion value, and health index were selected as the centralized correlation features related to service life in the historical inspection data of underground cables.

[0031] Since cable number, partial discharge value, neutral corrosion value, overload condition, and health index are numerical variables, while visual defects are categorical variables, data preprocessing is required for both numerical and categorical variables to facilitate later model construction. After filtering the correlation features using the Pearson correlation coefficient, the numerical variables corresponding to the correlation features are partial discharge value, neutral corrosion value, and health index, while the categorical variable corresponding to the correlation features is visual defects.

[0032] Standardization of the numerical variables corresponding to the associated features involves standardizing the partial discharge value, neutral corrosion value, and health index, including standardizing the numerical variables based on their mean and variance. Where x is the initial value of the numerical variable, x new x is the standardized value of the numerical variable. mean Let x be the mean of the numerical variable. std Let V be the variance of the numerical variable. This will change the mean of the characteristic values ​​to 0 and the standard deviation to 1, ensuring that the overall interval for all values ​​falls between [-1, 1].

[0033] As shown in Table 1, one-hot encoding of the type variables corresponding to the associated features is equivalent to one-hot encoding of the visualization defects.

[0034] Table 1. Type Variables: "One-hot Encoding of Visual Defects"

[0035] In this embodiment, the last 20% of the associated feature set is used as the test set, and the remaining 80% is used as the training set. It is understood that in other embodiments, the ratio of the test set to the training set can be adjusted according to the actual size of the associated feature set.

[0036] Constructing a Weibull distribution prediction model using Weibull distribution theory and a training set includes: The initial structure of the Weibull distribution prediction model is constructed based on the probability density function of the Weibull distribution in the theory of Weibull distribution. The initial structure of the Weibull distribution prediction model is trained using the training set to obtain the Weibull distribution prediction model.

[0037] The Weibull distribution, also known as the Weibull distribution or Weibull distribution, has the following probability density function: Where x is the cable failure time, λ is the scale parameter, representing the failure time corresponding to a failure probability of 63.2%, and k is the shape parameter, representing the data dispersion; the larger this parameter is, the smaller the data dispersion.

[0038] The cumulative probability P(x) of the Weibull distribution is obtained by integrating the probability density function of the Weibull distribution: Taking the logarithm twice on both sides of the cumulative probability P(x) of the Weibull distribution, we get In[-In(1-P)]=k·In(x|λ). Plotting a scatter plot of In[-In(1-P)] on k·In(x|λ) reveals the linear relationship between the two, with a slope of k. When P=1-1 / e=0.632, x=λ.

[0039] In this embodiment, the initial structure of the Weibull distribution prediction model is trained using the training set, and the slope k of the training set that satisfies the linear relationship is the shape parameter k of the Weibull distribution. Then, the Weibull distribution prediction model is constructed based on the shape parameter k.

[0040] Building lifespan prediction models using BP neural network theory and training sets includes: The initial structure of the lifetime prediction model is constructed based on BP neural network theory; The initial structure of the lifespan prediction model is trained using the training set to obtain the lifespan prediction model.

[0041] like Figure 4 As shown, the BP neural network, belonging to artificial neural networks, can use multiple hidden layers between input and output to model complex nonlinear relationships. A BP neural network structure containing at least an input layer, hidden layers, and an output layer is used as the output structure of the lifetime prediction model. As a lifetime prediction model, the lifetime of the underground cable should be the output; therefore, the cable's service life is used as the output value, and partial discharge value, visible defects, neutral corrosion value, and health index are used as input features. The predicted service life is defined as Y. i ={y1,y2,…,y n The input feature quantity, as a parameter input to the neural network, is defined as the input vector X. i ={x1,x2,…,x n}, where n is the number of samples accepted by the neural network, i = 1, 2, ..., n, and the actual operating years are defined as the expected output vector as O. i ={o1,o2,…,o n The hidden layer output vector is B = {b1, b2, ..., b}. n The connection weight of each neuron from the input layer to the hidden layer is defined as W. ji ={w j1 ,w j2 ,…,w jn}, j = 1, 2, ..., p; the connection weights of each neuron from the hidden layer to the output layer are defined as V. mi ={v m1 ,v m2 ,…,v mp}, m = 1, 2, ..., n. During training, the entire forward propagation process of the BP (back propagation) neural network is performed first. This includes calculating the activation values ​​s of each neuron in the hidden layer. j : The sum of the products of the input vector and its corresponding weights is then subtracted from the hidden layer threshold θj to determine whether a neuron is activated. The sigmoid function is chosen as the activation function as follows: t is the independent variable, f(t) is the dependent variable, and its value range is (0,1).

[0042] Activation value s j Substituting the value into the Sigmoid function yields the hidden layer output value b. j for: Similarly, the transmission method from the hidden layer to the output layer is as follows: where m is the slope parameter of the Sigmoid function, which has the advantage of being differentiable, and when m approaches infinity, the Sigmoid function is equivalent to a threshold. y m is the actual output value. s m is the activation value of each neuron in the output layer, θ m is the hidden layer threshold.

[0043] According to the entire forward propagation process of the BP (back propagation) neural network, the output value y m is obtained. Then the difference between the output value and the expected value is judged, that is, the error between the predicted operating life and the actual operating life is judged. If there is a large error, the BP (back propagation) neural network will perform correction from back to front (output layer - hidden layer - input layer): Output layer correction error: Hidden layer correction error: where is the differential of m, and d m and e j reverse error adjustment, adjust the inter-layer weights V and W: a is the learning rate and 0<a<1. Reverse adjustment is performed according to the error between the output value and the expected value, and the gradient descent method is used to update the inter-layer weights: W=W ji -ΔW ji ; Until the error threshold is reached, that is, convergence, the life prediction model is constructed according to the finally updated inter-layer weights.

[0044] Calculating the weight coefficient ratio of the Weibull distribution prediction model and the life prediction model according to the probability distribution function and the test set includes: Constructing a first remaining useful life with the Weibull distribution prediction model and the probability distribution function, and presetting a first weight; Constructing a second remaining useful life with the life prediction model and the probability distribution function, and presetting a second weight; Constructing a third remaining useful life with the test set and the probability distribution function; Calculate the sum of the product of the first remaining useful life and the first weight, and the product of the second remaining useful life and the second weight. Determine if the ratio of the third remaining useful life to the sum is 1. If it is, use the ratio of the first weight to the second weight as the weight coefficient. If not, update the first weight and the second weight.

[0045] In this embodiment, the cable life is predicted, using the Remaining Usable Life (RUL) as the standard. Partial discharge values, visible defects, neutral corrosion values, and health indices from the test set are used as inputs. A Weibull distribution prediction model and a life prediction model are used to calculate the probability distribution function representing the RUL, respectively. The output is the first Remaining Usable Life (RUL) based on the physical model. WD (μ p ,σ p ) and data-driven second remaining useful life RUL BP (μ N ,σ N The third remaining useful life (RUL) is constructed using the output values ​​of the test set, i.e., the number of years the device has been in operation, and the probability distribution function. ansys (μ0,σ0). μ p Let σ be the mean of the RUL probability distribution function constructed in the Weibull distribution prediction model. p Let μ be the variance in the RUL probability distribution function constructed in the Weibull distribution prediction model. Similarly, μ N Let σ be the mean of the RUL probability distribution function constructed in the lifetime prediction model. N μ represents the variance in the RUL probability distribution function constructed in the lifetime prediction model. 0 Let σ be the mean of the RUL probability distribution function constructed from the output values ​​of the test set. 0 Let Dif represent the variance in the RUL probability distribution function constructed from the output values ​​of the test set. Let Dif denote the validity of the mixed RUL prediction results, then: λ w As the first weight, λ B As the second weight, λ w ,λ B ∈[0,1]. Continuously update the first weight and the second weight until Dif reaches 1, and construct a fusion model using the first and second weights at this time, the Weibull distribution prediction model, and the lifetime prediction model.

[0046] In this embodiment, the weighting coefficient ratio is λ. w ,λ B =[0.5,0.9]、λ w ,λ B =[0.25,0.75]、λ w ,λB = any one of [0.5, 0.5].

[0047] In some other embodiments, step S5 further includes: The fusion model is validated using a test set, the optimal solution for weight coefficient allocation is output, and the fusion model is updated based on the optimal solution for weight coefficient allocation.

[0048] like Figure 5 As shown, the optimal solution for the weighting coefficient ratio is λ. w ,λ B =[0.5,0.9].

[0049] As a second embodiment of this application, an underground cable lifetime prediction system based on a hybrid model includes: A data acquisition and processing model is used to acquire underground cable inspection data and perform preprocessing. The physical model building module is used to build a Weibull distribution prediction model based on the data output from the data acquisition and processing model. The data-driven model building module is used to build a lifetime prediction model based on the data output from the data acquisition and processing model. The weighting coefficient allocation module is used to calculate the weighting coefficient allocation of the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function. The fusion construction module is used to output a fusion model based on the weighting coefficient ratio, Weibull distribution prediction model, and life prediction model; the prediction analysis module uses the fusion model and current underground cable inspection data to output underground cable life prediction data.

[0050] In this embodiment, by constructing the Weibull distribution prediction model and the lifetime prediction model respectively, and calculating the weight coefficient ratio of the two, a fusion model that balances the dependence defects and error defects of the two is obtained. This not only improves the accuracy of underground cable lifetime prediction, but also reduces the need for professional knowledge of physical models, making the model construction simpler and the scope of application wider.

[0051] The specific embodiments described above are preferred embodiments of the underground cable life prediction method and system based on a hybrid model of this application, and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.

Claims

1. A method for predicting the lifetime of underground cables based on a hybrid model, characterized in that: Includes the following steps: S1: Preprocess the historical detection dataset of underground cables to obtain the associated feature set, and divide the associated feature set into a training set and a test set; S2: Construct a Weibull distribution prediction model using Weibull distribution theory and training set, and construct a lifetime prediction model using BP neural network theory and training set; S3: Calculate the weighting ratios of the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function and the test set; S4: Construct a fusion model based on the Weibull distribution prediction model, the life prediction model, and the weighting coefficient ratio, obtain the current underground cable inspection data, and use the fusion model to output the underground cable life prediction data.

2. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 1, characterized in that: S1 includes: Pearson correlation coefficient was used to obtain life-related features in historical inspection data of underground cables. The numerical variables corresponding to the association features are standardized, and the typological variables corresponding to the association features are one-hot encoded. The standardized numerical variables and the one-hot encoded typological variables are used as the association feature set. The associated feature set is divided into a training set and a test set.

3. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 2, characterized in that: The standardization process for the numerical variables corresponding to the associated features includes: The numerical variables are standardized based on their mean and variance.

4. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 1, characterized in that: The construction of a Weibull distribution prediction model using Weibull distribution theory and training set includes: The initial structure of the Weibull distribution prediction model is constructed based on the probability density function of the Weibull distribution in the theory of Weibull distribution. The initial structure of the Weibull distribution prediction model is trained using the training set to obtain the Weibull distribution prediction model.

5. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 4, characterized in that: The probability density function of the Weibull distribution is: ； Where x is the cable failure time, λ is the scale parameter, representing the failure time corresponding to a failure probability of 63.2%, and k is the shape parameter, representing the dispersion of the data.

6. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 1, characterized in that: The method of constructing a lifetime prediction model using BP neural network theory and training set includes: Based on the BP neural network theory, the initial structure of the lifetime prediction model is constructed using a BP neural network structure that includes at least an input layer, a hidden layer, and an output layer. The initial structure of the lifespan prediction model is trained using the training set to obtain the lifespan prediction model.

7. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 1, characterized in that: The calculation of the weight coefficient ratios for the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function and the test set includes: The first remaining useful life is constructed using a Weibull distribution prediction model and a probability distribution function, with a preset first weight; A second remaining useful life is constructed using a life prediction model and a probability distribution function, with a preset second weight; Construct the third remaining lifetime using the test set and probability distribution function; Calculate the sum of the product of the first remaining useful life and the first weight, and the product of the second remaining useful life and the second weight. Determine if the ratio of the third remaining useful life to the sum is 1. If it is, use the ratio of the first weight to the second weight as the weight coefficient. If not, update the first weight and the second weight.

8. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 7, characterized in that: S4 further includes: The fusion model is validated using a test set, the optimal solution for weight coefficient allocation is output, and the fusion model is updated based on the optimal solution for weight coefficient allocation.

9. The method for predicting the lifetime of underground cables based on a hybrid model as described in claim 1, characterized in that: The historical monitoring dataset for underground cables includes at least: cable number, years of operation, partial discharge value, visual defects, neutral corrosion value, overload conditions, and health index.

10. A hybrid model-based underground cable lifetime prediction system, used to implement the method as described in any one of claims 1 to 9, characterized in that: include: A data acquisition and processing model is used to acquire underground cable inspection data and perform preprocessing. The physical model building module is used to build a Weibull distribution prediction model based on the data output from the data acquisition and processing model. The data-driven model building module is used to build a lifetime prediction model based on the data output from the data acquisition and processing model. The weighting coefficient allocation module is used to calculate the weighting coefficient allocation of the Weibull distribution prediction model and the lifetime prediction model based on the probability distribution function. The fusion building module is used to output a fusion model based on the weighting coefficient ratio, Weibull distribution prediction model, and lifetime prediction model. The predictive analysis module uses a fusion model and current underground cable inspection data to output predicted underground cable lifespan data.

Images