Virtual reality-based urban underground cable operation and maintenance method and system

Abstract

The application relates to the field of electric digital data processing, in particular to a virtual reality-based urban underground cable operation and maintenance method and system. The method comprises the following steps: collecting actual cable data and labeling corresponding operation states, training a data expansion model based on a generative adversarial network algorithm based on adaptive noise injection, training a classification model based on actual cable data and synthetic cable data, analyzing new actual cable data by using the trained classification model, obtaining the operation state of the cable, and completing operation and maintenance work. The existing urban underground cable operation and maintenance method and system have the problem that the judgment on the operation state of the underground cable is inaccurate. The virtual reality-based urban underground cable operation and maintenance method and system can make a relatively accurate judgment on the operation state of the underground cable.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, specifically to a method and system for the operation and maintenance of urban underground cables based on virtual reality. Background Technology

[0002] A utility tunnel is an underground passage structure whose main function is to provide space for and protect various urban pipelines, such as underground cables.

[0003] Problems with underground cables can disrupt the normal operation of a city. However, the scientific management of underground cables has always been a serious challenge because manual inspection is inefficient, some cable intersections are difficult to inspect, and underground cables are long, making it difficult for inspectors to keep track of their operation in real time.

[0004] With the gradual development of technologies such as artificial intelligence, machine vision, and the Internet of Things, a new type of urban underground cable operation and maintenance method and system has emerged—the smart utility tunnel operation and maintenance system—capable of all-weather remote monitoring of underground cables. This system deploys sensing devices, video surveillance, gas monitoring, personnel intrusion management, and cable joint temperature monitoring facilities within cable tunnels (tunnels specifically designed for underground cables). Combined with smart utility tunnel operation and maintenance methods, it uses 3D modeling technology to recreate the real-world scene of the tunnel, establishing an interactive virtual reality environment. Users can immerse themselves in browsing and operating the environment using head-mounted devices or projection systems, achieving comprehensive data collection and centralized monitoring of cables within the tunnel. The smart utility tunnel operation and maintenance system also features alarm and trend analysis query functions. Through this platform, maintenance personnel can fully grasp the internal environment of the tunnel and the operating status of the cables, realizing the informatization and intelligent management of cables and the tunnel.

[0005] However, existing methods and systems for the operation and maintenance of urban underground cables also have some problems. First, they mostly use basic generative adversarial network algorithms to augment the collected cable data, lacking specific optimization for rare state data. The resulting virtual samples have poor diversity and distribution consistency, making it difficult to improve the classification performance of low-frequency categories (such as fault states). Therefore, using such augmented data to train the classification model will affect the training effect of the classification model. Second, when training the classification model, traditional classification algorithms (such as support vector machines, random forests, etc.) have not been optimized for the class imbalance problem. Therefore, low-frequency categories are easily ignored, leading to a decrease in classification accuracy and insufficient ability to process high-dimensional feature data. It is difficult to effectively capture the complex nonlinear relationships in the cable state attributes, resulting in poor training effect of the classification model and inaccurate judgment of the operating status of underground cables.

[0006] Therefore, existing urban underground cable operation and maintenance methods and systems have the problem of inaccurate judgment of the operating status of underground cables. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a virtual reality-based method and system for the operation and maintenance of urban underground cables that can make a more accurate judgment on the operating status of underground cables.

[0008] To address the aforementioned technical problems, the present invention provides a virtual reality-based method for the operation and maintenance of urban underground cables, comprising:

[0009] S1. Obtain actual cable data for underground cables in various locations, and manually label the operating status of each actual cable data point;

[0010] The operating status includes normal, warning, and fault;

[0011] S2. Input the labeled actual cable data into the data augmentation model, train the data augmentation model using a generative adversarial network algorithm based on adaptive noise injection, obtain the trained data augmentation model, generate synthetic cable data using the trained data augmentation model, and use the actual cable data and the synthetic cable data as training data.

[0012] S3. Input the training data into the classification model, and train the classification model using the random forest algorithm based on relative entropy to obtain the trained classification model;

[0013] S4. Collect new actual cable data, input the new actual cable data into the trained classification model, and obtain the operating status corresponding to the new actual cable data. The operating status includes normal, warning, and fault. When the operating status is warning, perform inspection; when the operating status is fault, perform maintenance.

[0014] As a further improvement of the present invention: the training of the data augmentation model using a generative adversarial network algorithm based on adaptive noise injection in S2 includes:

[0015] S201. Initialize the parameters of the generator and discriminator of the generative adversarial network;

[0016] S202. The loss function is designed based on the dynamic changes in the cable data distribution during training, and the calculation formula is as follows:

[0017]

[0018] In the formula, D is the loss function of the generator; c () represents the discriminator function; G c() represents the generator function; z represents the input noise, specifically a random vector sampled from the latent space; x is the loss function of the discriminator; ce This is a sample of actual cable data;

[0019] S203. The weight parameters of the generator and discriminator are dynamically adjusted based on the discriminator feedback. Simultaneously, the learning rate and noise patterns are controlled using mechanisms based on topological differences and adaptive noise injection, as expressed below:

[0020]

[0021] In the formula, ΔT c For topological differences; nc is the number of samples input into the generative adversarial network in the current batch;

[0022] Re() is the ReLU activation function used to extract structural information from samples. c () represents the generator function. For the i-th z A noise vector, For the i-th z A sample of actual cable data; i z It is the cable data sample index and is the same as the noise vector index;

[0023] S204. Adaptive noise is injected into the generator, and the noise intensity is determined by the feedback from the discriminator on the synthesized cable data. The calculation formula is as follows:

[0024] N c (z,α c )=z+α c ·∈ c ,

[0025]

[0026] In the formula, N c (z,α c ) represents the adaptive noise injection function; z represents the input noise; α c Noise intensity coefficient; ∈ c For normally distributed random noise, Let I be a normal distribution with a mean of 0 and a variance equal to the identity matrix. Represents a normal distribution;

[0027] S205. After obtaining the generator's output, the generator and discriminator are updated according to the dynamic learning rate. In the optimization of the generator and discriminator, a dynamic weight update strategy is used to balance the generator's generation diversity and the discriminator's discriminative ability. The calculation formula is as follows:

[0028]

[0029] In the formula, η represents the generator's weights, ← represents the parameter update sign; c To generate a dynamic learning rate for adversarial networks; The gradient of the generator loss function with respect to its weights. For the weights of the discriminator, This represents the gradient of the discriminator loss function with respect to its weights.

[0030] S206. Repeat S202 to S205 until the preset stopping iteration condition is met.

[0031] Preferably, in S202, the D c (G c The formula for calculating (z) is:

[0032]

[0033] In the formula, D c () represents the discriminator function; G c () represents the generator function; z represents the input noise; Sig() represents the Sigmoid activation function; For the weights of the discriminator, This is the bias of the discriminator.

[0034] Preferably, the noise intensity coefficient α in S204 c The calculation formula is:

[0035]

[0036] In the formula, β c To adjust the sensitivity parameter; τ ce The target threshold.

[0037] As a further improvement of the present invention: the step S3 of training the classification model using the relative entropy-based random forest algorithm includes:

[0038] S301. Initialize the number and depth of decision trees required for the random forest;

[0039] S302. Feature selection is performed using Gaussian curvature;

[0040] S303. In the process of constructing the decision tree, a relative entropy loss function dynamically adjusted based on the distribution characteristics of cable data is adopted. The calculation formula of the relative entropy loss function is as follows:

[0041]

[0042] In the formula, K is the relative entropy loss function; ue Number of categories; The probability of the k-th class predicted by the model; This represents the true probability distribution of the k-th class, where k is the class index;

[0043] S304. Train each decision tree based on the adjusted loss function, and select the optimal split point that maximizes the inter-class separation when splitting nodes. Determine the optimal splitting method during the decision tree growth process. The node splitting optimization method is expressed as follows:

[0044]

[0045] In the formula, The optimal splitting parameters; θ is the node splitting optimization function based on the adjusted loss function; ury For the candidate splitting parameter set;

[0046] S305. By combining several independent trained decision trees into a random forest and using majority voting for global prediction, the calculation formula for the synthesized global model is as follows:

[0047]

[0048] In the formula, The final prediction result of the random forest; the mode() function is the majority voting function; Tu i Let i be the decision tree; represents the input features of the decision tree in the random forest; i is the decision tree index. The number of decision trees in the random forest;

[0049] S306. The weights of each feature are updated by analyzing its contribution to the model's output entropy. The formula for calculating the entropy weight adjustment of dynamic feature points is as follows:

[0050]

[0051] In the formula, Here, j represents the weight of the j-th feature, where j is the feature index; update() is the weight update function. The contribution of the j-th feature to the change in the loss function;

[0052] S307. Repeat S302 to S306 until the preset stopping iteration condition is met.

[0053] Preferably, in S304, the... The calculation formula is:

[0054]

[0055] In the formula, Left(θ) ury ) represents the data point in the left sub-decision tree of the decision tree under the candidate splitting parameter set; i w Index the data points; For the i-th prediction of the decision tree in a random forest w The class probability of each data point; For the i-th w The true class probability of each data point; Right(θ) ury ) represents the data point in the right sub-decision tree of the decision tree under the candidate split parameter set.

[0056] Preferably, in S306, the... The calculation formula is:

[0057]

[0058] In the formula, update() is the weight update function; The contribution of the j-th feature to the change in the loss function; β is the weight of the j-th feature; is the natural constant; β uac This is the Ricci curvature adjustment factor; Let be the Ricci curvature of the j-th feature; Let be the learning rate for the weight update in the t-th iteration, where t is the iteration number. The sign of the partial derivative. The adjusted loss function, This represents the change in the adjusted loss function.

[0059] This invention also provides a virtual reality-based urban underground cable operation and maintenance system, which, when applied to the above-mentioned virtual reality-based urban underground cable operation and maintenance method, includes:

[0060] The data acquisition unit is used to collect actual cable data;

[0061] The data storage and expansion unit is used to store data and generate synthetic cable data based on actual cable data through a data expansion model;

[0062] The data processing unit is used to label the corresponding operating status of the input cable data using a classification model.

[0063] The beneficial effects of the present invention are as follows: The urban underground cable operation and maintenance method based on virtual reality provided by the present invention can make a relatively accurate judgment on the operating status of underground cables.

[0064] This method employs a generative adversarial network (GAN) algorithm based on adaptive noise injection to train a data augmentation model. Adaptive noise is injected into the generator to produce diverse synthetic cable data. The discriminator uses feedback on the quality of the synthetic cable data to determine the noise intensity, achieving dynamic adjustment of the noise intensity. This optimizes the generator's early diversity exploration capabilities, improving the realism and coverage of the synthetic cable data. This results in higher-quality data for training the classification model. Another method uses a random forest algorithm based on relative entropy to train the classification model. A dynamically adjusted loss function adapts to imbalanced data, strengthening the model's ability to distinguish rare classes, thereby improving its classification ability and ultimately enhancing the accuracy of its judgments on the operational status of underground cables. Therefore, the virtual reality-based urban underground cable operation and maintenance system applying this method also makes relatively accurate judgments on the operational status of underground cables. Attached Figure Description

[0065] Figure 1 This is a comparison chart of the convergence of the loss functions of the generator and discriminator in this invention;

[0066] Figure 2 This is a comparison chart of the data distribution of the synthetic cable in this invention;

[0067] Figure 3 This is a diagram illustrating the impact of the number of decision trees on the model accuracy in this invention.

[0068] Figure 4 This diagram illustrates the impact of different feature selection methods on model performance in this invention. Detailed Implementation

[0069] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0070] The method for operation and maintenance of urban underground cables based on virtual reality provided by this invention includes:

[0071] S1. Obtain actual cable data for underground cables in various locations, and manually label the operating status of each actual cable data point;

[0072] Actual cable data includes cable temperature (reflecting the thermal state of the cable during operation), cable humidity (reflecting the humidity level of the surrounding environment), cable voltage (reflecting the voltage the cable carries), cable current (reflecting the current carrying capacity of the cable), cable insulation condition (reflecting the insulation performance of the cable), the horizontal and vertical coordinates of the cable location (for pinpointing the cable's precise location on a city map), depth information (indicating the cable's burial depth), cable type (reflecting whether the cable is high-voltage or low-voltage), and maintenance records (reflecting the cable's historical maintenance information).

[0073] Manual annotation is performed by technical experts in the field.

[0074] Operating status includes normal, warning, and fault; normal indicates that the cable is in good condition; warning indicates that there is a potential risk to the cable and it is recommended to check it soon; fault indicates that the cable is damaged and needs to be repaired immediately.

[0075] S2. Input the labeled actual cable data into the data augmentation model, train the data augmentation model using the generative adversarial network algorithm based on adaptive noise injection, obtain the trained data augmentation model, use the trained data augmentation model to generate synthetic cable data, and use the actual cable data and synthetic cable data as training data.

[0076] Generative Adversarial Network (GAN) algorithms based on adaptive noise injection refer to the dynamic adjustment of noise intensity during training by utilizing the feedback of the discriminator. By improving the noise injection method, it can adaptively adjust the noise injection according to data characteristics or model state, thereby enhancing the generator's diversity exploration capabilities in the early stages of training. This achieves more stable and efficient adversarial learning and provides reliable support for generating high-quality cable data.

[0077] S201. Initialize the parameters of the generator and discriminator of the generative adversarial network; initialize the weights in the model using a random distribution to provide scalable initial conditions for subsequent optimization, expressed as:

[0078]

[0079] In the formula, Here, represents the weights of the generator, and ~ represents the distribution it follows. It follows a normal distribution. Let I be a normal distribution with zero mean and variance equal to the identity matrix, where I is the identity matrix. The weights of the discriminator;

[0080] S202. A loss function is designed based on the dynamic changes in cable data distribution during training, including the loss components of the generator and discriminator, to improve the network's sensitivity and robustness to changes in cable data distribution. The calculation formula is as follows:

[0081]

[0082] In the formula, Let D be the loss function of the generator, representing the optimization objective of the generator against the discriminator; c () is the discriminator function, representing the probability of classifying the input cable data as real or generated; G c () is the generator function, representing the mapping relationship between the noise-synthesized cable data; z is the input noise, specifically a random vector sampled from the latent space; x is the loss function of the discriminator, representing the optimization objective of the discriminator in distinguishing between actual cable data and synthetic cable data; ce This is a sample of actual cable data;

[0083] D c (G c The formula for calculating (z) is:

[0084] The discriminator performs probability estimation on the composite cable data based on its output. The discriminator's discrimination process for the composite cable data is expressed as follows:

[0085]

[0086] In the formula, D c () represents the discriminator function; G c () represents the generator function; z represents the input noise; Sig() is the Sigmoid activation function, which maps the input to a nonlinear function in the (0,1) interval; For the weights of the discriminator, This is the bias of the discriminator.

[0087] The generator's output is based on adaptive noise calculation. When synthesizing cable data, the diversity and complexity of cable data attributes (such as humidity, voltage, and current) are considered. These attributes may vary due to different cable operating states and environmental conditions. Through dynamic control of noise intensity, the generated cable data can cover a wider distribution range. The calculation formula is as follows:

[0088]

[0089] In the formula, The generated cable data characterizes the generator's mapping of the input noise to the output; G c () represents the generator function, z represents the input noise, and tanh() is the hyperbolic tangent activation function. N represents the weights of the generator; c (z,α c ) is the adaptive noise injection function, representing the application of additional random perturbations to the original noise; α c The noise intensity coefficient represents the dynamic weights superimposed on the original noise during the generation process; This is the bias of the generator.

[0090] S203. To achieve self-balancing iterative updates, the weight parameters of the generator and discriminator are dynamically adjusted based on discriminator feedback. Simultaneously, mechanisms for topological differences and adaptive noise injection are used to control the learning rate and noise patterns, thereby achieving a more stable adversarial learning effect. The calculation of topological differences effectively assesses the distance between synthetic cable data and real data in the topological space (i.e., using topological differences to assess the spatial attribute consistency between synthetic cable data and real data). By controlling the topological differences between synthetic cable data and real data, the consistency of the synthetic cable data in spatial attributes is ensured, while simultaneously improving the reliability of synthetic cable data in cable condition monitoring tasks, as expressed below:

[0091]

[0092] In the formula, ΔT c The topological difference characterizes the distance between the synthetic cable data and the actual cable data in the topological space; nc is the number of samples input into the generative adversarial network in the current batch; ∥∥ is the L2 norm, Re() is the ReLU activation function used to extract structural information from the samples, and G... c () represents the generator function. For the i-th z A noise vector, For the i-th z A sample of actual cable data; i z It is the cable data sample index and is the same as the noise vector index;

[0093] The learning rate is dynamically adjusted based on the obtained topological differences to avoid overfitting or undertraining that may result from a fixed learning rate. For example, cables with different maintenance records may have significant differences in voltage or insulation status. A dynamic learning rate can better adapt to these complexities and accelerate convergence. The calculation formula is as follows:

[0094] η c =η0exp(-λ c ΔT c ),

[0095] In the formula, η c The dynamic learning rate of the generative adversarial network (GAN) represents the control over the parameter update step size at different training times and is set to 0.03; η0 is the initial learning rate of the GAN; λ c To adjust the intensity parameter and characterize its sensitivity to topological differences, it is set to 0.3; ΔT c This is due to topological differences.

[0096] S204. To enhance the generator's ability to explore diversity in the early stages of training and overcome fluctuations in generated quality, adaptive noise is injected into the generator. The noise intensity is determined by the discriminator's feedback on the synthesized cable data, improving the ability to reproduce the synthesized cable data at a detailed level. For example, in the case of minor fluctuations in the cable's thermal state, the generator can better capture the fine distribution of real data features for continuous attributes such as temperature and humidity, improving the accuracy of reflecting the actual operating state. The calculation formula is as follows:

[0097] N c (z,α c )=z+α c ·∈ c ,

[0098]

[0099] In the formula, N c (z,α c ) represents the adaptive noise injection function, which characterizes the addition of an extra random perturbation to the original noise; z is the input noise; α c The noise intensity coefficient represents the dynamic weights superimposed on the original noise during the generation process; ∈ c For normally distributed random noise, ~ indicates that it follows a specific distribution. Let I be a normal distribution with a mean of 0 and a variance equal to the identity matrix. Represents a normal distribution;

[0100] Noise intensity coefficient α c The discriminator dynamically adjusts the noise intensity based on the output of the synthesized cable data. By dynamically adjusting the threshold and sensitivity parameters, the discriminator can improve the quality control of the synthesized cable data. For example, for attributes with high dynamic range such as current and voltage, it can ensure that the synthesized cable data has both the distribution characteristics of real data and avoids generation distortion caused by excessive noise. The calculation formula is as follows:

[0101]

[0102] In the formula, exp() is the natural exponential function; β c To adjust the sensitivity parameter, which characterizes the amplification effect on the discriminator's recognition level, it is set to 0.1; τ ce The target threshold, representing the probability level that the expected synthetic cable data can be identified as real by the discriminator, is set to 0.5.

[0103] like Figure 1 As shown, by comparing the convergence of the loss function, it can be seen that the adaptive noise injection generative adversarial network and the traditional generative adversarial network show a decreasing trend in the loss function during training, demonstrating the advantages of the adaptive noise injection technique in terms of faster convergence speed and smaller fluctuations.

[0104] S205. After obtaining the generator's output, the generator and discriminator are updated according to the dynamic learning rate to achieve balanced and effective adversarial training. In the optimization of the generator and discriminator, a dynamic weight update strategy is used to balance the diversity generated by the generator and the discriminant's discriminative ability. For example, for attributes with imbalanced characteristics (such as high-voltage and low-voltage cable types), the generator can better learn the rare high-voltage cable state distribution, thereby avoiding the uniformity or class imbalance of the synthesized cable data. The calculation formula is as follows:

[0105]

[0106] In the formula, η represents the generator's weights, ← represents the parameter update sign; c The dynamic learning rate of the generative adversarial network represents the control of the parameter update step size at different training times; The gradient of the generator loss function with respect to its weights. For the weights of the discriminator, This represents the gradient of the discriminator loss function with respect to its weights.

[0107] S206. Repeat S202 to S205 until the preset stopping iteration condition is met, indicating that the data augmentation model training is complete. The preset stopping iteration condition is reaching the preset maximum number of iterations, which is set to 1000. Assuming there are 800 actual cable data points and the data augmentation model generates 200 synthetic cable data points, the training data contains 1000 samples.

[0108] like Figure 2 As shown, to verify the effectiveness, a comparison chart of the distribution of synthesized cable data is used to compare the distribution of real data, synthesized cable data from adaptive noise injection generative adversarial network, and synthesized cable data from traditional generative adversarial network. The chart shows that the data distribution generated by adaptive noise injection technology is closer to the real data.

[0109] S3. Input the training data into the classification model, and use the random forest algorithm based on relative entropy to train the classification model to obtain the trained classification model;

[0110] The Random Forest algorithm based on relative entropy adapts to class imbalance by using a dynamically adjusted relative entropy loss function, while using regularization to reduce the risk of overfitting. When splitting nodes, it optimizes the splitting method by maximizing information gain and improves classification stability by fusing multiple decision tree models for majority voting.

[0111] S301. Initialize the number and depth of decision trees required for the random forest, setting the number of decision trees in the random forest to be... D u The dimension of the cable dataset input to the random forest represents the total number of input features. The number of decision trees is set to 200.

[0112] like Figure 3 As shown, to illustrate the rationality of the numerical settings for decision trees, the model performance was verified by varying the number of decision trees. Furthermore, the performance of the random forest based on relative entropy improved faster as the number of trees increased, demonstrating the advantages of the random forest based on relative entropy as an optimization split point and dynamic adjustment of the loss function in classification algorithms.

[0113] To improve the diversity of cable state training data, a subset of cable data is randomly selected for each decision tree in the random forest. The subset of cable data used for the i-th decision tree represents the training samples for the i-th decision tree model; D u This is a complete cable dataset, representing training data for all available cable states.

[0114] By randomly selecting a subset of cable data for each decision tree, the diversity of training data is increased. For example, when dealing with data with different spatial characteristics (such as cable location attributes), the model can learn different state patterns of deep-buried and shallow-buried cables through subset sampling regarding cables with different burial depths (cable location z-axis attributes).

[0115] S302. By using Gaussian curvature to filter features, and by calculating the local curvature of cable data features in high-dimensional space, features that are highly sensitive to classification boundaries are selected to help the model better capture the complex nonlinear relationships in cable data. Let the feature set selected by Gaussian curvature be... The selection method is to calculate the Gaussian curvature of the feature vector. If it is greater than a preset threshold, it is selected. The selected feature represents a more discriminative feature in high-dimensional cable data.

[0116] Gaussian curvature feature screening technology can be used to select key features that are highly sensitive to classification boundaries (such as the combination of humidity and temperature) to improve the ability to model complex attribute relationships. For example, the combination of cable temperature and cable humidity attributes in the data may be identified as an important feature through Gaussian curvature screening, revealing the unique impact of hot and humid environments on cable conditions.

[0117] like Figure 4 As shown, to verify the effectiveness of using Gaussian curvature for feature selection, the impact of different feature selection methods on performance was compared. The results show that features selected by Gaussian curvature significantly improve model performance, highlighting the superiority of Gaussian curvature selection in complex nonlinear relationships.

[0118] S303. In the process of constructing the decision tree, a relative entropy loss function that is dynamically adjusted based on the distribution characteristics of cable data is adopted to adapt to the inherent distribution characteristics of different categories of cable data, especially in the case of class imbalance. This allows the classifier to more accurately distinguish between different categories during the learning process. For example, low-voltage cables may appear much more frequently than high-voltage cables. By dynamically adjusting the relative entropy, the model focuses more on the state classification of high-voltage cables, thereby improving classification accuracy. The formula for calculating the relative entropy loss function is as follows:

[0119]

[0120] In the formula, K is the relative entropy loss function, representing the overall classification error; ue Number of categories; The probability of the k-th class predicted by the model represents the confidence level of the model's output for the k-th class; log represents the logarithmic function, with a default base of 10; The true probability distribution of the k-th class represents the true class distribution of the cable data; k is the class index.

[0121] To balance the complexity of the model structure with the risk of overfitting, a dynamically adjusted loss function is used to optimize the growth of the decision tree. The calculation formula is as follows:

[0122]

[0123] In the formula, The adjusted loss function represents the evaluation value after comprehensively considering classification error and regularization term; α u The first adjustment coefficient, β is the relative entropy loss function. u The second adjustment coefficient is set to 0.1; R ur (u) is the regularization term for the random forest.

[0124] To achieve dynamic loss function, the formula for calculating the first adjustment coefficient during training is as follows:

[0125]

[0126] In the formula, k u The parameter for adjusting the slope, which characterizes the sensitivity of the first adjustment coefficient to changes in the degree of imbalance, is set to 0.3; δ represents the entropy value of the cable dataset. u The entropy threshold is set to 0.2.

[0127] Regularization terms constrain model complexity, preventing overfitting to noisy data. For example, for potential extreme noise in Ia (cable current), regularization suppresses its adverse effects on model weights, thereby improving generalization performance and enhancing the convergence speed and stability of the classification model. The formula for calculating the regularization term in random forests is:

[0128]

[0129] In the formula, R ur (u) is the regularization term for the random forest, m u This represents the total number of sample features input into the random forest. The regularization weight for the j-th feature represents the degree of influence of the feature on the regularization term. It is obtained by calculating the L2 norm of each feature vector in the cable dataset. The weight of the j-th feature represents the importance of that feature in the model.

[0130] S304. By training each decision tree based on the adjusted loss function and selecting the optimal split point that maximizes inter-class separation when splitting nodes, the optimal partitioning method is determined during the decision tree growth process, fully exploiting the information gain under the relative entropy criterion. The node splitting optimization method is expressed as follows:

[0131]

[0132] In the formula, The optimal splitting parameter represents the splitting threshold or splitting dimension; Let θ be the node splitting optimization function based on the adjusted loss function, representing the evaluation value for different splitting schemes; ury The candidate splitting parameter set represents all feasible splitting methods; This represents the set of candidate splitting parameters for obtaining the minimum value of the function;

[0133] To measure the classification error of the left and right sub-decision trees after a node split, the calculation formula for the node split optimization function based on the adjusted loss function is as follows:

[0134]

[0135] In the formula, Left(θ) ury Let i be the data point i in the left sub-decision tree of the decision tree under the candidate split parameter set. w Index the data points; For the i-th prediction of the decision tree in a random forest w The class probability of a data point represents the prediction confidence that the data point belongs to a certain class; For the i-th wThe true class probability of a data point represents the true classification of that data point; Right(θ) ury ) represents the data point in the right sub-decision tree of the decision tree under the candidate split parameter set.

[0136] S305. By combining several independently trained decision trees into a random forest and using majority voting for global prediction, the aggregation of output results when multiple decision tree models are fused improves classification accuracy and stability. The calculation formula for the synthesized global model is as follows:

[0137]

[0138] In the formula, The final prediction result of the random forest represents the model's final classification of the input cable data; the `mode()` function is the majority voting function, representing the selection of the most frequently occurring category from the outputs of each decision tree; Tu i Let be the i-th decision tree, and let represent the classification function of each decision tree; is the input feature, representing the cable data to be classified; i is the decision tree index. The number of decision trees in the random forest;

[0139] S306. By analyzing the contribution of each feature to the model's output entropy, the weights are updated to improve feature sensitivity and classification accuracy in the later stages of random forest model training. The formula for calculating the dynamic feature point entropy weight adjustment is as follows:

[0140]

[0141] In the formula, Here, represents the weight of the j-th feature, characterizing the impact of this feature on the model's prediction; j is the feature index, and update() is the weight update function. The contribution of the j-th feature to the change in the loss function represents the importance of this feature in model tuning.

[0142] The weight update function represents the specific implementation method for modifying feature weights. It dynamically updates feature weights by combining Ricci curvature and the feature's contribution to entropy, enabling the model to focus more on key features in the later stages of training. For example, for important historical maintenance information of cables, the model can gradually increase its weight during training, improving its ability to identify cable conditions with maintenance records. The formula for calculating the weight update function is:

[0143]

[0144] In the formula, update() is the weight update function; The contribution of the j-th feature to the change in the loss function represents the importance of this feature in model tuning. Let be the weight of the j-th feature, representing the influence of this feature on the model's prediction; e is the natural constant; β uac The Ricci curvature adjustment coefficient represents the strength of the influence of Ricci curvature on weight updates; Let Ricci curvature be the j-th feature, which characterizes the degree of bending of the data manifold for that feature under a specific category; Let be the learning rate for the weight update in the t-th iteration. The sign of the partial derivative. The adjusted loss function, This represents the change in the adjusted loss function.

[0145] To adapt to real-time changes in model performance, the learning rate is dynamically adjusted. This adjustment is based on real-time performance metrics, ensuring the model can quickly adapt to performance changes. For example, in the early stages of model training, when classification performance improves rapidly, the learning rate is higher; later, as performance approaches convergence, the learning rate is gradually reduced to improve the model's convergence stability. The calculation formula is as follows:

[0146]

[0147] In the formula, Let be the learning rate for the (t+1)th iteration. δ is the learning rate for the weight update in the t-th iteration, where t is the number of iterations; ucg is the learning rate decay factor, representing the strength of adjusting the learning rate based on model performance, and is set to 0.3; perf(t) is the performance metric of the random forest model after the t-th iteration, specifically the training accuracy metric.

[0148] S307. Repeat S302 to S306 until the preset stopping iteration condition is met, indicating that the classification model training is complete. The preset stopping iteration condition is reaching the preset maximum number of iterations, which is set to 1000.

[0149] S4. Collect new actual cable data and input the new actual cable data into the trained classification model to obtain the operating status corresponding to the new actual cable data. The operating status includes normal, warning, and fault. When the operating status is warning, perform inspection; when the operating status is fault, perform maintenance.

[0150] The virtual reality-based urban underground cable operation and maintenance system provided by this invention applies the aforementioned virtual reality-based urban underground cable operation and maintenance method, including:

[0151] The data acquisition unit is used to collect actual cable data;

[0152] The data storage and expansion unit is used to store data and generate synthetic cable data based on actual cable data through a data expansion model;

[0153] The data processing unit is used to label the corresponding operating status of the input cable data through a classification model;

[0154] The control unit is used to receive the corresponding operating status of the cable data, output instructions based on the operating status, and send the instructions to the execution unit.

[0155] The execution unit is used to receive and execute instructions, including instructions for staff inspection and maintenance.

Claims

1. A method for operation and maintenance of urban underground cables based on virtual reality, characterized in that, include: S1. Obtain actual cable data for underground cables in various locations, and manually label the operating status of each actual cable data point; The operating status includes normal, warning, and fault; S2. Input the labeled actual cable data into the data augmentation model, train the data augmentation model using a generative adversarial network algorithm based on adaptive noise injection, obtain the trained data augmentation model, generate synthetic cable data using the trained data augmentation model, and use the actual cable data and the synthetic cable data as training data. S3. Input the training data into the classification model, and train the classification model using the random forest algorithm based on relative entropy to obtain the trained classification model; The step S3, which involves training the classification model using the relative entropy-based random forest algorithm, includes: S301. Initialize the number and depth of decision trees required for the random forest; S302. Feature selection is performed using Gaussian curvature; S303. In the process of constructing the decision tree, a relative entropy loss function dynamically adjusted based on the distribution characteristics of cable data is adopted. The calculation formula of the relative entropy loss function is as follows: In the formula, K is the relative entropy loss function; ue Number of categories; The probability of the k-th class predicted by the model; This represents the true probability distribution of the k-th class, where k is the class index; S304. Train each decision tree based on the adjusted loss function, and select the optimal split point that maximizes the inter-class separation when splitting nodes. Determine the optimal splitting method during the decision tree growth process. The node splitting optimization method is expressed as follows: In the formula, The optimal splitting parameters; θ is the node splitting optimization function based on the adjusted loss function; ury For the candidate splitting parameter set; S305. By combining several independent trained decision trees into a random forest and using majority voting for global prediction, the calculation formula for the synthesized global model is as follows: In the formula, The final prediction result of the random forest; the mode() function is the majority voting function; Tu i Let i be the decision tree; represents the input features of the decision tree in the random forest; i is the decision tree index. The number of decision trees in the random forest; S306. The weights of each feature are updated by analyzing its contribution to the model's output entropy. The formula for calculating the entropy weight adjustment of dynamic feature points is as follows: In the formula, Here, j represents the weight of the j-th feature, where j is the feature index; update() is the weight update function. The contribution of the j-th feature to the change in the loss function; S307. Repeat S302 to S306 until the preset stopping iteration condition is met; S4. Collect new actual cable data, input the new actual cable data into the trained classification model, and obtain the operating status corresponding to the new actual cable data. The operating status includes normal, warning, and fault. When the operating status is warning, perform inspection; when the operating status is fault, perform maintenance.

2. The method for operation and maintenance of urban underground cables based on virtual reality according to claim 1, characterized in that, The method described in S2 for training the data augmentation model using a generative adversarial network algorithm based on adaptive noise injection includes: S201. Initialize the parameters of the generator and discriminator of the generative adversarial network; S202. The loss function is designed based on the dynamic changes in the cable data distribution during training, and the calculation formula is as follows: In the formula, D is the loss function of the generator; c () represents the discriminator function; G c () represents the generator function; z represents the input noise, specifically a random vector sampled from the latent space; x is the loss function of the discriminator; ce This is a sample of actual cable data; S203. The weight parameters of the generator and discriminator are dynamically adjusted based on the discriminator feedback. Simultaneously, the learning rate and noise patterns are controlled using mechanisms based on topological differences and adaptive noise injection, as expressed below: In the formula, ΔT c For topological differences; nc is the number of samples input into the generative adversarial network in the current batch; Re() is the ReLU activation function used to extract structural information from samples. c () represents the generator function. For the i-th z A noise vector, For the i-th z A sample of actual cable data; i z It is the cable data sample index and is the same as the noise vector index; S204. Adaptive noise is injected into the generator, and the noise intensity is determined by the feedback from the discriminator on the synthesized cable data. The calculation formula is as follows: N c (z,α c )=z+α c ·∈ c , In the formula, N c (z,α c ) represents the adaptive noise injection function; z represents the input noise; α c Noise intensity coefficient; ∈ c For normally distributed random noise, Let I be a normal distribution with a mean of 0 and a variance equal to the identity matrix. Represents a normal distribution; S205. After obtaining the generator's output, the generator and discriminator are updated according to the dynamic learning rate. In the optimization of the generator and discriminator, a dynamic weight update strategy is used to balance the generator's generation diversity and the discriminator's discriminative ability. The calculation formula is as follows: In the formula, η represents the generator's weights, ← represents the parameter update sign; c To generate a dynamic learning rate for adversarial networks; The gradient of the generator loss function with respect to its weights. The weights of the discriminator, This represents the gradient of the discriminator loss function with respect to its weights. S206. Repeat S202 to S205 until the preset stopping iteration condition is met.

3. The method for operation and maintenance of urban underground cables based on virtual reality according to claim 2, characterized in that, D in S202 c (G c The formula for calculating (z) is: In the formula, D c () represents the discriminator function; G c () represents the generator function; z represents the input noise; Sig() represents the Sigmoid activation function; The weights of the discriminator, This is the bias of the discriminator.

4. The method for operation and maintenance of urban underground cables based on virtual reality according to claim 2, characterized in that, The noise intensity coefficient α in S204 c The calculation formula is: In the formula, β c To adjust the sensitivity parameter; τ ce The target threshold.

5. The method for operation and maintenance of urban underground cables based on virtual reality according to claim 1, characterized in that, The S304 described The calculation formula is: In the formula, Left(θ) ury ) represents the data point in the left sub-decision tree of the decision tree under the candidate splitting parameter set; i w Index the data points; For the i-th prediction of the decision tree in a random forest w The class probability of each data point; For the i-th w The true class probability of each data point; Right(θ) ury ) represents the data point in the right sub-decision tree of the decision tree under the candidate split parameter set.

6. The method for operation and maintenance of urban underground cables based on virtual reality according to claim 1, characterized in that, The S306 described The calculation formula is: In the formula, update() is the weight update function; The contribution of the j-th feature to the change in the loss function; The weight of the j-th feature; e is the natural constant; β uac This is the Ricci curvature adjustment factor; Let be the Ricci curvature of the j-th feature; Let be the learning rate for the weight update in the t-th iteration, where t is the iteration number. The sign of the partial derivative. The adjusted loss function, This represents the change in the adjusted loss function.

7. A virtual reality-based urban underground cable operation and maintenance system, characterized in that, The method for operation and maintenance of urban underground cables based on virtual reality, as described in any one of claims 1 to 6, includes: The data acquisition unit is used to collect actual cable data; The data storage and expansion unit is used to store data and generate synthetic cable data based on actual cable data through a data expansion model; The data processing unit is used to label the corresponding operating status of the input cable data using a classification model.

Images