High-voltage cable operation state reliability evaluation method and system

Abstract

The invention relates to the technical field of power equipment state monitoring and intelligent operation and maintenance, and discloses a high-voltage cable operation state reliability evaluation method and system. The method at least comprises the following steps: converting a standardized multi-modal data set into a standardized evaluation data set by referring to a pre-constructed integrity evaluation system, and generating a training sample set and a verification sample set based on the standardized evaluation data set; constructing a composite neural network architecture comprising a feature extraction layer, a nonlinear transformation layer and a parameter optimization layer, and training the composite neural network architecture based on the training sample set to obtain an evaluation result vector comprising a plurality of reliability dimensions; and comparing the evaluation result vector with a verification sample set from a plurality of dimensions including numerical value consistency check, trend goodness of fit analysis and anomaly detection capability evaluation to obtain a structured evaluation report reflecting the reliability of the running state of the high-voltage cable. The intelligent operation and maintenance response efficiency of the high-voltage cable is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and intelligent operation and maintenance technology, and in particular to a method and system for assessing the reliability of high-voltage cable operation status. Background Technology

[0002] The safe and stable operation of high-voltage cables is directly related to the reliability of power grid supply and the energy supply guarantee capability. Therefore, accurate monitoring and reliability assessment of their operating status have become key requirements in the field of smart grid operation and maintenance.

[0003] Currently, various online monitoring systems for high-voltage cables mainly focus on monitoring key parameters such as partial discharge, grounding current, temperature distribution, and channel environment. However, under complex operating conditions and long-term operation scenarios, existing monitoring and evaluation technologies still face multiple technical bottlenecks: First, the fusion of multi-source heterogeneous data is difficult, as the data formats and sampling frequencies of different monitoring modules vary significantly, making it difficult to form a high-quality evaluation data source; second, the evaluation model architecture has limitations, with traditional models often relying on single features or simple algorithms, failing to accurately map the correlation between the complex operating states of high-voltage cables and reliability indicators; third, the reliability evaluation dimension is singular, with existing methods often focusing on threshold judgments of single parameters, resulting in insufficient scientific rigor and practicality in the evaluation results. Summary of the Invention

[0004] To address the aforementioned technical bottlenecks, this invention provides a method and system for assessing the reliability of high-voltage cable operation status. This method and system can overcome the limitations of traditional technologies in data processing, model architecture, and assessment systems, providing scientific support for intelligent operation and maintenance of high-voltage cables and helping the power grid achieve a leap from "passive monitoring" to "precise optimization".

[0005] In a first aspect, embodiments of the present invention provide a method for assessing the reliability of high-voltage cable operation, including: Acquire multi-source heterogeneous data generated during the operation of high-voltage cables to obtain a standardized multimodal dataset; The standardized multimodal dataset is transformed into a standardized evaluation dataset according to a pre-built integrity evaluation system, and training sample sets and validation sample sets are generated based on the standardized evaluation dataset; A composite neural network architecture comprising a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer is constructed, and the composite neural network architecture is trained based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. The feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on a multi-objective loss function and update the network weights through backpropagation based on the loss value. The evaluation result vector is compared with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

[0006] Preferably, after comparing the evaluation result vector with the verification sample set from multiple dimensions including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status, the method further includes: Based on the dynamic matching relationship between multi-dimensional reliability weight configuration and the operating environment characteristics of high-voltage cables, the structured assessment report is transformed into reliability optimization suggestions for the operating status of high-voltage cables.

[0007] Preferably, the step of acquiring multi-source heterogeneous data generated during the operation of the high-voltage cable to obtain a standardized multimodal dataset includes: By deploying a distributed acquisition network at high-voltage cable monitoring terminals, communication quality indicators, location data, and multi-physical quantity monitoring data generated during the operation of high-voltage cables can be collected in real time. The communication quality indicators, the positioning data, and the multi-physical quantity monitoring data are preprocessed to obtain a standardized multimodal dataset.

[0008] Preferably, the step of transforming the standardized multimodal dataset into a standardized evaluation dataset with reference to a pre-constructed integrity evaluation system, and generating a training sample set and a validation sample set based on the standardized evaluation dataset, includes: Construct a completeness assessment system that includes data quality verification, time series alignment, feature dimension unification, and sample label annotation; Based on the aforementioned integrity assessment system, the standardized multimodal dataset is transformed into a standardized assessment dataset that meets the requirements of supervised training. A sliding window algorithm is used to extract time-related sample fragments from the standardized evaluation dataset, and the time-related sample fragments are then subjected to structured processing and data augmentation to generate a training sample set. The standardized evaluation dataset is stratified and sampled to obtain a sample set covering multiple typical operating conditions. Based on the training sample set, the sample set is temporally isolated without overlap to obtain a validation sample set.

[0009] Preferably, the construction of the composite neural network architecture comprising a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer includes: The feature extraction layer consists of multiple LSTM hidden layers and batch normalization layers. Each LSTM hidden layer is followed by a batch normalization layer, and each LSTM hidden layer is configured with a different number of feature units. The nonlinear transformation layer is constructed by multiple fully connected layers, wherein each fully connected layer is configured with a different number of feature units; The parameter optimization layer is constructed based on a multi-objective loss function that considers multiple dimensions, including the accuracy of output results, the interpretability of decisions, and the stability of reliability assessment results, combined with the gradient descent optimization mechanism. The feature extraction layer, the nonlinear transformation layer, and the parameter optimization layer are connected in series to form a composite neural network architecture.

[0010] Preferably, the process of constructing the multi-objective loss function includes: Based on the mean square error between the predicted and actual operating status values, and combined with the consistency constraint of the time series change trend, an accuracy loss function is constructed. By utilizing the KL divergence between the probability distribution of the significance of the output results and the probability distribution of the contribution of feature nodes, and combining the information entropy of feature nodes to constrain the transparency of model decision-making, an interpretable loss function is constructed. By analyzing the variance of the output results over a continuous time period and combining the robustness test of the reliability assessment results under different operating conditions, a stability loss function is constructed. The accuracy loss function, the interpretability loss function, and the stability loss function are dynamically weighted and fused to form a multi-objective loss function.

[0011] Preferably, the step of training the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions includes: The training sample set is input into the composite neural network architecture, and the feature dimension is expanded through the input layer, transforming the training sample set into several three-dimensional evaluation tensors containing time batch size, feature dimension and number of monitoring points; All the three-dimensional evaluation tensors are input into the feature extraction layer, and temporal features are extracted using forward propagation; The time-series features are input into the nonlinear transformation layer for nonlinear mapping to obtain a high-dimensional feature vector that is adapted to the reliability assessment dimension. The parameter optimization layer calculates the loss value based on the high-dimensional feature vector using a multi-objective loss function, and then updates the network weights via backpropagation using a gradient descent optimization mechanism. After iterative training until the composite neural network architecture converges, the output is an evaluation result vector containing communication quality, positioning accuracy, and multi-physical quantity monitoring reliability.

[0012] Preferably, the step of comparing the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status includes: The evaluation result vector is compared with the validation sample set from multiple dimensions, including numerical consistency test, trend fit analysis and anomaly detection capability assessment, to obtain the comparison results of the corresponding dimensions. The comparison results for each dimension are weighted and integrated to obtain a comprehensive reliability score; Based on the aforementioned comprehensive reliability score, and at least incorporating the results of abnormal event diagnosis and performance trend prediction, a structured assessment report reflecting the reliability of the high-voltage cable's operating status is generated.

[0013] Preferably, the dynamic matching relationship between the multi-dimensional reliability weight configuration and the operating environment characteristics of the high-voltage cable transforms the structured assessment report into reliability optimization suggestions for the operating status of the high-voltage cable, including: Based on the deviation and historical trend of each reliability indicator in the structured report, the optimal weight combination for multi-dimensional reliability is calculated through dynamic weight allocation. Based on the optimal weight combination of the multi-dimensional reliability, identify the performance bottleneck of each reliability dimension; For each performance bottleneck in the reliability dimension, the dynamic matching relationship between the multi-dimensional reliability weight configuration and the operating environment characteristics of the high-voltage cable is invoked. Combined with the expert knowledge base rules and the prediction results of the machine learning model, reliability optimization suggestions corresponding to the performance bottleneck are generated.

[0014] Secondly, embodiments of the present invention provide a high-voltage cable operating status reliability assessment system, comprising: The data acquisition module is used to acquire multi-source heterogeneous data generated during the operation of high-voltage cables and obtain a standardized multimodal dataset. The sample set generation module is used to transform the standardized multimodal dataset into a standardized evaluation dataset with reference to a pre-built integrity evaluation system, and to generate training sample sets and validation sample sets based on the standardized evaluation dataset; An architecture training module is used to construct a composite neural network architecture that includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer, and to train the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. The feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on a multi-objective loss function and update the network weights through backpropagation based on the loss value. The report generation module is used to compare the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

[0015] Compared with the prior art, the reliability assessment method and system for high-voltage cable operation status according to embodiments of the present invention has the following advantages at least one point: First, the composite neural network architecture accurately captures the temporal correlation characteristics of high-voltage cable operation data through the combination design of multi-layer LSTM hidden layers and batch normalization layers. Combined with the nonlinear mapping of fully connected layers and parameter optimization of multi-objective loss functions, it reduces the numerical error of reliability assessment and improves the consistency of temporal trends. At the same time, it improves the transparency of model decision-making through KL divergence and information entropy constraints, thus solving the "black box" problem of traditional deep models. Second, by dynamically allocating weights, the evaluation results are transformed into targeted optimization suggestions, forming a full-process control of "data collection-model evaluation-parameter optimization-feedback closed loop". This significantly improves the response efficiency of intelligent operation and maintenance of high-voltage cables, while enhancing the reliability and security of power grid supply, providing technical support for the efficient and stable operation of smart grids. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for assessing the reliability of high-voltage cable operation status according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process of constructing a composite neural network architecture according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the process for obtaining the evaluation result vector according to an embodiment of the present invention; Figure 4 This is another flowchart illustrating a method for assessing the reliability of high-voltage cable operation status according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a high-voltage cable operation status reliability assessment system according to an embodiment of the present invention; Figure label: 01. Data Acquisition Module; 02. Sample Set Generation Module; 03. Architecture Training Module; 04. Report Generation Module. Detailed Implementation

[0017] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0018] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by those skilled in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0019] like Figure 1 The diagram shown is a flowchart illustrating a method for assessing the reliability of high-voltage cable operation status according to an embodiment of the present invention. (Refer to...) Figure 1 This invention provides a method for assessing the reliability of high-voltage cable operation, comprising the following steps: S1. Obtain multi-source heterogeneous data generated during the operation of high-voltage cables to obtain a standardized multimodal dataset; Specifically, step S1 includes: 1) By deploying a distributed acquisition network at the high-voltage cable monitoring terminal, communication quality indicators, location data, and multi-physical quantity monitoring data generated during the operation of the high-voltage cable are collected in real time; During the data acquisition process, a unified time-series coordination mechanism deployed in a distributed acquisition network is used to assign weight coefficients based on the importance of each data source. A unified timestamp is calculated using a time-series synchronization algorithm to achieve millisecond-level time alignment of communication quality indicators, location data, and multi-physical quantity monitoring data, thereby eliminating time-series deviations caused by different sampling frequencies.

[0020] In the collection of communication quality indicators, a multi-mode communication link status monitoring unit is deployed, covering mainstream communication standards such as 4G / 5G / NB-IoT. The unit uses the radio frequency front-end chipset to cyclically detect the received signal strength indication and collect core parameters such as signal strength, bit error rate, and handover success rate in real time. Based on the CRC check algorithm, the number of bit error frames is counted and a communication quality time series database is constructed.

[0021] During positioning data acquisition, a GPS / BeiDou dual-mode positioning receiver and a traveling wave positioning device are simultaneously connected to form a multi-source positioning acquisition array. Basic parameters such as latitude and longitude coordinates, altitude, and positioning accuracy factor are collected. Fault traveling wave waveforms are captured with the help of a high-frequency sampling card, and environmental compensation data such as temperature and humidity are recorded simultaneously for positioning correction.

[0022] In the acquisition of multi-physical quantity monitoring data, distributed sensor acquisition nodes are deployed. Grounding current is acquired through Rogowski coils, partial discharge signals are acquired through capacitive voltage dividers, temperature distribution is acquired through thermocouple arrays, and mechanical vibration spectrum is acquired through vibration sensors. Combined with a unified clock source, time synchronization of all sensor data is achieved, ensuring the comprehensiveness and accuracy of multi-dimensional data acquisition.

[0023] Meanwhile, the collected data is pre-labeled according to the classification rules of key data sources (which need to be processed in real time), auxiliary data sources (used for historical backtracking), and important data sources (used for redundancy verification), so as to clarify the processing priority and application scenarios of each type of data.

[0024] 2) Preprocess the communication quality indicators, positioning data, and multi-physical quantity monitoring data to obtain a standardized multimodal dataset.

[0025] The processing strategies used in the preprocessing stage are explained below: First, an outlier filtering algorithm is used to remove invalid data such as abnormal spikes in communication quality bit error rate, excessive drift in positioning data, and sudden changes in sensor monitoring data. Then, interpolation is used to fill in a small number of missing values ​​to ensure data integrity.

[0026] Secondly, based on the data standardization formula, heterogeneous data with different dimensions such as signal strength, positioning error, grounding current, and partial discharge signal are transformed into standardized data with a unified scale, eliminating the impact of differences in dimensions.

[0027] Finally, all processed data is transformed into a unified format that meets the input requirements of subsequent composite neural network architectures, and the stability of the data is verified through a standardized data interface, ultimately forming a standardized multimodal dataset with a well-structured structure and meeting quality standards.

[0028] S2. Transform the standardized multimodal dataset into a standardized evaluation dataset according to the pre-built integrity evaluation system, and generate training sample sets and validation sample sets based on the standardized evaluation dataset; Specifically, step S2 includes: 1) Construct a completeness assessment system that includes data quality verification, time series alignment, feature dimension unification, and sample label annotation; When constructing an integrity assessment system, a four-level architecture needs to be built: "Data Collection and Verification (Data Collection and Verification Layer) - Quality Calculation (Quality Indicator Calculation Layer) - Hierarchical Assessment (Integrity Hierarchical Assessment Layer) - Feedback and Optimization (Result Feedback and Optimization Layer)".

[0029] The data acquisition and verification layer performs preliminary verification of the effectiveness of standardized multimodal data acquisition through time-series consistency checks and data integrity screening; the quality index calculation layer conducts timeliness assessment, consistency verification, and integrity rate calculation to quantify the data quality level; the integrity grading assessment layer establishes a three-level assessment standard of excellent, good, and qualified based on the quantitative results, and clarifies the data compliance threshold; the result feedback and optimization layer automatically triggers a supplementary acquisition mechanism for data that does not meet the standards, diagnoses and locates abnormal data sources, and adaptively adjusts the assessment parameters.

[0030] Simultaneously, four major functional modules are embedded in each level: data quality verification (ensuring the basic validity of data through outlier detection, missing value determination, and data format compliance verification), time series alignment (achieving multi-dimensional data time series synchronization based on a unified clock source), feature dimension unification (eliminating heterogeneous data differences through normalization and dimension mapping), and sample label annotation (labeling data according to operating conditions and anomaly types), forming a complete closed-loop integrity assurance system. Among them, data quality verification is embedded in the quality indicator calculation layer, time series alignment is embedded in the data collection and verification layer, and feature dimension unification and sample label annotation are both embedded in the integrity grading assessment layer.

[0031] 2) Based on the integrity evaluation system, the standardized multimodal dataset is transformed into a standardized evaluation dataset that meets the requirements of supervised training; Specifically, based on the constructed integrity assessment system, the standardized multimodal dataset is first subjected to data quality verification. Invalid data is removed by outlier detection algorithms, and data meeting the standards is selected by combining integrity rate calculation. Then, time series alignment technology is used to correct data timestamp deviations based on a unified time series coordination mechanism to ensure the time series consistency of data from different sources. Next, feature dimensions are uniformly processed. Methods such as feature normalization and dimension mapping are used to transform heterogeneous features such as communication quality indicators, positioning data, and multi-physical quantity monitoring data into standardized features with unified dimensions. Finally, sample labeling is completed according to the supervised training requirements. Labels such as working condition type (normal / abnormal), abnormality level, and operating environment are added to each data sample, ultimately forming a standardized assessment dataset with regular features, clear labels, and that meets the input requirements for supervised training.

[0032] 3) The sliding window algorithm is used to extract time-related sample fragments from the standardized evaluation dataset, and the time-related sample fragments are subjected to structured processing and data augmentation to generate a training sample set; Specifically, a sliding window algorithm with a fixed window size and step size is used to extract time-series related samples within continuous time segments from the standardized evaluation dataset. Each sample segment contains time-series sequence data with multi-dimensional features. The extracted sample segments are then structured, including feature filtering (retaining key features strongly correlated with reliability assessment) and data format regularization (converting to a tensor format recognizable by the model). Data augmentation techniques are used to expand sample diversity by generating multiple sets of derived samples through time-series data flipping, Gaussian noise injection, and interpolation augmentation, effectively improving the model's generalization ability. Finally, the original and augmented samples after structured processing are integrated to form a training sample set covering different time dimensions and multiple operational features.

[0033] 4) Perform stratified sampling on the standardized evaluation dataset to obtain a sample set covering multiple typical operating conditions, and perform time-series non-overlapping isolation on the sample set based on the training sample set to obtain a validation sample set.

[0034] Based on the operating condition classification information in the standardized evaluation dataset, a stratified sampling technique is used to extract a representative sample set covering all operating conditions according to the sample proportion of different typical operating conditions (including normal operating conditions and various abnormal operating conditions such as signal strength drop, positioning drift, and sensor anomalies).

[0035] To prevent data leakage from affecting the accuracy of model validation, the extracted sample set is subjected to temporal non-overlap isolation processing based on the time range of the training sample set, ensuring that the time interval of the validation sample is completely independent of the training sample and there is no temporal overlap. At the same time, the completeness of the working condition coverage of the validation sample is verified, and missing typical working condition samples are supplemented, ultimately forming a validation sample set for model generalization capability verification and hyperparameter tuning.

[0036] S3. Construct a composite neural network architecture that includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer, and train the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. In the composite neural network architecture, the feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on the multi-objective loss function and update the network weights through backpropagation based on the loss value.

[0037] like Figure 2 The diagram shown is a flowchart illustrating step S3 of an embodiment of the present invention. (Refer to...) Figure 2 Step S3 in this embodiment of the invention includes: S301, a feature extraction layer consisting of multiple LSTM hidden layers and batch normalization layers; Each LSTM hidden layer is followed by a batch normalization layer, and each LSTM hidden layer is configured with a different number of feature units.

[0038] The feature extraction layer adopts a cascaded structure of "LSTM hidden layer + batch normalization layer". In this embodiment, a total of 3 LSTM hidden layers and 3 batch normalization layers are set. Each LSTM hidden layer is directly connected to a batch normalization layer, forming a progressive processing unit of "extraction-optimization". The first LSTM hidden layer is configured with 16 feature units to capture short-term temporal correlation features of high-voltage cable operation data; the second LSTM hidden layer is configured with 8 feature units to focus on the extraction of medium-term temporal dependencies; and the third LSTM hidden layer is configured with 4 feature units to achieve the refinement and integration of long-term temporal features.

[0039] The batch normalization layer optimizes the distribution of the LSTM output data using the following formula: in, This represents the output features after batch normalization. This represents the output features of the LSTM layer. This represents the average of the batch data. Indicates the variance of batch data. and Indicates learnable parameters, This indicates the avoidance of tiny constants with a denominator of 0, effectively alleviating the gradient vanishing problem and improving training stability.

[0040] S302. A nonlinear transformation layer is constructed by multiple fully connected layers; Each fully connected layer is configured with a different number of feature units.

[0041] In this embodiment, the nonlinear transformation layer consists of three fully connected layers connected in series. Each fully connected layer achieves nonlinear mapping of the feature space by configuring a different number of feature units. Specifically, the first fully connected layer is configured with 8 feature units, mapping the temporal features output by the feature extraction layer to a low-dimensional feature space; the second fully connected layer is configured with 16 feature units, expanding the dimensionality and enhancing the nonlinearity of the low-dimensional features by introducing nonlinear expressive power through the ReLU activation function; the third fully connected layer is configured with 4 feature units, shrinking the enhanced features to a space suitable for reliability evaluation, and outputting a high-dimensional feature vector with uniform dimensionality.

[0042] The computational logic for each fully connected layer is shown in the following formula: in, This represents the output characteristics of the fully connected layer. This represents the weight matrix of this layer. This indicates the bias term.

[0043] By dynamically optimizing the training process, we ensure that the feature mapping is adapted to the reliability assessment requirements such as communication quality and positioning accuracy.

[0044] S303. A multi-objective loss function is constructed based on multiple dimensions, including the accuracy of the output results, the interpretability of the decision, and the stability of the reliability assessment results. This is combined with a gradient descent optimization mechanism to form a parameter optimization layer. Specifically, the process of constructing a multi-objective loss function includes: 1) Based on the mean square error between the predicted and actual operating status values, and combined with the consistency constraint of the time series change trend, an accuracy loss function is constructed; The accuracy loss function is constructed using a dual objective of "mean squared error + time series trend consistency constraint", and is characterized by the following formula: in, Represents the accuracy loss function. Represents the total number of samples. Indicates the first The actual running state value of each sample. The first term represents the model's predicted value, and it uses the mean squared error to ensure that the predicted value is consistent with the actual value. Indicates the length of the time series. The time-series rate of change of the true value This represents the time-series rate of change of the predicted value. The first term represents the trend constraint weight (default value is 0.3, which can be adjusted according to the operating conditions of the high-voltage cable and the time series characteristics of the data). The second term ensures that the evaluation results match the actual operating trend by constraining the deviation of the two rates of change.

[0045] 2) By utilizing the KL divergence between the probability distribution of the significance of the output results and the probability distribution of the contribution of the feature nodes, and combining the information entropy of the feature nodes to constrain the transparency of the model decision, an interpretable loss function is constructed. The interpretability loss function is constructed based on "KL divergence + feature node information entropy" and is represented by the following formula: in, Represents the interpretability loss function. KL divergence is used to measure the probability distribution of the significance of the output results. Contribution probability distribution of feature nodes Consistency; Indicates the total number of feature nodes. Indicates the first The information entropy of each feature node constrains the transparency of the model's decision-making process; , Indicates the weighting coefficient ( The default value is 0.25. The default value is 0.2, which can be adjusted according to the intensity of the need for interpretability of model decisions and the scale of the number of feature nodes, balancing the importance of the two constraints.

[0046] 3) By analyzing the variance of the output results over a continuous time period and combining the robustness test of the reliability assessment results under different operating conditions, a stability loss function is constructed; The stability loss function is constructed through "analysis of variance + robustness test" and is characterized by the following formula: in, Represents the stability loss function. This represents the variance calculation function. This represents the model output within a continuous time window. The first term ensures short-term output stability through variance constraints. Indicates the total number of operating conditions. Indicates the first Evaluation results under various working conditions This represents the average evaluation result under all operating conditions. The second term represents the L2 norm, and it passes a robustness test to ensure the model's adaptability under different operating conditions. , Indicates the weighting coefficient ( The default value is 0.2. The default value is 0.25, which can be adjusted according to the complexity of the high-voltage cable's operating conditions and the magnitude of data fluctuations.

[0047] 4) Dynamically weight and fuse the accuracy loss function, interpretability loss function, and stability loss function to form a multi-objective loss function.

[0048] An adaptive weighting strategy is used to fuse the three loss functions, and the specific formula is as follows: in, Represents a multi-objective loss function. , , Represents dynamic weighting coefficients, satisfying The calculation logic is as follows: in, This represents the variance of the accuracy loss function during training. This represents the variance of the interpretability loss function during training. This represents the variance of the stability loss function.

[0049] This strategy assigns higher weights to loss functions with lower variance (more stable performance), dynamically balancing the evaluation's accuracy, interpretability, and stability.

[0050] S304. The feature extraction layer, nonlinear transformation layer and parameter optimization layer are connected in series to form a composite neural network architecture.

[0051] Specifically, the output of the feature extraction layer is directly connected to the input of the nonlinear transformation layer, ensuring that the temporal features are transmitted to the fully connected layer for mapping without loss. The output of the nonlinear transformation layer is connected to the loss calculation module of the parameter optimization layer, providing input data for the multi-objective loss function. The weight update signal of the parameter optimization layer is fed back to the fully connected layer of the nonlinear transformation layer and the LSTM hidden layer of the feature extraction layer through the backpropagation link, forming a closed-loop architecture of "forward propagation feature processing + backpropagation parameter optimization", ensuring that each layer can work together to adapt to the reliability assessment requirements of multimodal time series data of high-voltage cables.

[0052] It should be noted that steps S301-S304 correspond to the implementation process of constructing a composite neural network architecture that includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer.

[0053] like Figure 3 The diagram shown is another flowchart illustrating step S3 of an embodiment of the present invention. (Refer to...) Figure 3 Step S3 in this embodiment of the invention includes: S305. Input the training sample set into the composite neural network architecture, expand the feature dimension through the input layer, and transform the training sample set into several three-dimensional evaluation tensors containing time batch size, feature dimension and number of monitoring points. After the training sample set is input into the composite neural network architecture, the input layer transforms the two-dimensional feature matrix (feature dimension × time step) of each training sample into a three-dimensional evaluation tensor through the feature dimension expansion technique.

[0054] The three dimensions of a tensor are defined as follows: time batch size (Each tensor contains) (data from consecutive time steps), feature dimensions (Covering communication quality, positioning accuracy, and monitoring of multiple physical quantities, etc.) (Core features of the class) and number of monitoring points (High-voltage cable deployment along the entire line) (Key monitoring points), the final tensor format is as follows: This format preserves the temporal correlation of time-series data and integrates the spatial features of multiple monitoring points, thus adapting to the input requirements of the LSTM feature extraction layer.

[0055] S306. Input all three-dimensional evaluation tensors into the feature extraction layer and extract temporal features using forward propagation; All 3D evaluation tensors are batch-wise input to the feature extraction layer, and temporal feature extraction is completed through forward propagation. During propagation, the first LSTM hidden layer traverses the temporal dimension of the tensor, capturing short-term temporal correlations of features at each monitoring point; after the data distribution is optimized by the batch normalization layer, the output features are fed into the second LSTM hidden layer to further explore medium-term temporal dependencies; the third LSTM hidden layer fuses the features from the first two layers to extract long-term temporal features across time steps, and the final output dimension is... The time-series feature matrix (each monitoring point corresponds to 4 refined time-series features) provides feature input for subsequent nonlinear mapping.

[0056] S307. Input the time-series features into the nonlinear transformation layer for nonlinear mapping to obtain a high-dimensional feature vector that fits the reliability assessment dimension. After the temporal feature matrix is ​​input into the nonlinear transformation layer, it undergoes nonlinear mapping sequentially through three fully connected layers. The first fully connected layer will... The temporal feature matrix is ​​transformed into an 8-dimensional low-dimensional vector, and invalid features are filtered out by the ReLU activation function. The second fully connected layer expands the 8-dimensional vector to 16 dimensions to enhance the non-linear expressive power of the features. The third fully connected layer shrinks the 16-dimensional features to a 4-dimensional high-dimensional feature vector through a linear activation function. Each dimension of this vector corresponds to four reliability assessment dimensions: communication quality, positioning accuracy, multi-physical quantity monitoring, and device operation, ensuring that the features are accurately matched with the assessment targets.

[0057] S308. The loss value is calculated based on the high-dimensional feature vector and the multi-objective loss function through the parameter optimization layer, and the network weight is updated through backpropagation in combination with the gradient descent optimization mechanism. Specifically, the parameter optimization layer first uses a multi-objective loss function based on high-dimensional feature vectors and the true labels of the samples. Calculate the loss value for the current batch; then, using a stochastic gradient descent optimization mechanism, calculate the gradient of the loss value with respect to the parameters of each layer through backpropagation. During gradient calculation, first solve for the gradient of the parameter optimization layer, then backtrack sequentially to the weights of the fully connected layers in the nonlinear transformation layer. and bias And the LSTM weights of the feature extraction layer and the parameters of the batch normalization layer. , Finally, the formula is updated through parameters. Update all parameters, including Represents the set of network parameters. ( () represents the initial learning rate. express The parameter gradient at each time step is used to continuously reduce the loss value through iterative optimization.

[0058] S309. After iterative training until the composite neural network architecture converges, the output includes an evaluation result vector containing communication quality, positioning accuracy, and multi-physical quantity monitoring reliability.

[0059] Specifically, an early stopping mechanism is used during iterative training; when the multi-objective loss function value reaches its maximum value after 10 consecutive epochs... When the fluctuation amplitude is less than 1e−5, the composite neural network architecture is considered converged. After convergence, the model performs final inference on the input training samples, and outputs an evaluation result vector through processes such as feature extraction and nonlinear mapping. .in, This indicates the reliability score of communication quality. This indicates the reliability score for positioning accuracy. This indicates the reliability score for monitoring multiple physical quantities. The score represents the reliability of the device operation. The closer the score is to 1, the more reliable the corresponding dimension of the operation. The final vector comprehensively reflects the reliability of the high-voltage cable's multi-dimensional operation status.

[0060] It should be noted that steps S305-S309 correspond to the implementation process of training the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions.

[0061] S4. Compare the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency test, trend consistency analysis and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

[0062] Specifically, step S4 includes: 1) Compare the evaluation result vector with the validation sample set from multiple dimensions, including numerical consistency test, trend fit analysis and anomaly detection capability assessment, to obtain the comparison results for the corresponding dimensions; The following sections explain the aspects from the perspectives of numerical consistency testing, trend consistency analysis, and anomaly detection capability assessment: a) Numerical consistency check: Numerical consistency testing includes: calculating the absolute and relative errors between the predicted values ​​of each reliability index in the evaluation result vector and the actual values ​​in the validation sample set, establishing an error distribution histogram, and setting an error tolerance threshold based on historical data statistics. ( This represents the historical mean error. (Represents the historical error standard deviation), marking outliers that exceed the threshold.

[0063] b) Trend alignment analysis: Trend fit analysis includes: calculating the Pearson correlation coefficient between the time series of the evaluation result vector and the real time series of the validation sample set, and combining it with the consistency constraint of the time series rate of change to evaluate the degree of trend matching.

[0064] The following formula is used to characterize the consistency constraint of the time series rate of change: in, This represents the tolerance threshold for trend deviation, with a default value of 0.05.

[0065] c) Anomaly detection capability assessment: Anomaly detection capability evaluation includes: constructing a multi-class confusion matrix based on the annotated anomaly condition data of multiple classes in the validation samples, as shown below: in, Represents a multi-class confusion matrix. This represents the true number of instances where communication quality is an anomaly. This represents the actual number of instances where the positioning accuracy dimension is abnormal. This represents the true number of instances where the sensor detected dimensional anomalies. This represents the number of false positives in the communication quality dimension. This represents the number of false positives in the positioning accuracy dimension. This represents the number of false positives in the dimension monitored by the sensor. The number of false negatives represents the communication quality dimension. This represents the number of false negatives in the positioning accuracy dimension. This represents the number of false negatives in the dimension monitored by the sensor.

[0066] The accuracy of the early warning is calculated using the following formula: in, Indicates the accuracy of abnormal warnings. This represents the total number of anomaly samples labeled in the validation sample set. Indicates an indicator function, This indicates the preset time tolerance window, set to 5-10 minutes. Indicates the first Predicted warning time for each abnormal sample, Indicates the first The actual occurrence time of each outlier sample was also analyzed. Precision, recall, and other metrics were also calculated.

[0067] 2) The comparison results for each dimension are weighted and integrated to obtain a comprehensive reliability score; The weighted integral is calculated using a multi-level adaptive weighting mechanism. First, the importance weights of the indicators are determined. (e.g., communication quality reliability 0.25, positioning accuracy reliability 0.2, multi-physical quantity monitoring reliability 0.3, device operation reliability 0.25), and then combined with time decay weight and environmental adaptation weight, a comprehensive weight is obtained.

[0068] Specifically, the formula for calculating the time decay weight is as follows: in, Indicates the time decay weight. Indicates the attenuation coefficient. Indicates the current time. Indicates the data collection time.

[0069] The formula for calculating the environmental adaptation weight is as follows: in, Indicates the environmental adaptation weight. , A coefficient representing the environmental adaptability weight, used to balance the impact of environmental severity and equipment criticality on the weight. Indicates the severity of the environment. This indicates the criticality coefficient of the equipment.

[0070] Overall weight The calculation formula is as follows: Then, the score for each indicator is calculated using the following formula: in, This represents the score for a single reliability metric. This indicates the actual comparison results of the indicators. This represents the maximum value within the normal range of the indicator. This indicates the minimum value within the normal range of the indicator.

[0071] Finally, the overall reliability score is obtained through a weighted summation formula, as shown below: in, This represents the overall reliability score. This indicates the number of reliability indicators used in the scoring.

[0072] The rating is divided into four levels: ≥90 (excellent), 75-89 (good), 60-74 (average), and <60 (poor), corresponding to different reliability statuses.

[0073] 3) Based on the comprehensive reliability score, and at least in combination with the results of abnormal event diagnosis and performance trend prediction, a structured assessment report reflecting the reliability of the high-voltage cable's operating status should be generated.

[0074] Specifically, based on the comprehensive reliability score and level, the abnormal event diagnosis results are integrated, and the confusion matrix analysis and error labeling are combined to clarify the abnormal operating condition type (such as sudden drop in signal strength, positioning drift, etc.), occurrence time, scope of impact and severity, and the warning level is divided into Level 1 (urgent), Level 2 (important) and Level 3 (advice).

[0075] Furthermore, by fitting time-series data and extrapolating trends, performance trend prediction is performed, the changing patterns of various reliability indicators (such as communication quality degradation trends, changes in positioning accuracy stability, etc.) are analyzed, and future operational risks are predicted.

[0076] Finally, by integrating key elements such as comprehensive scoring, level determination, anomaly details, and trend analysis, and supplementing this with explanations of key indicator deviations, preliminary diagnoses of anomaly causes, and targeted monitoring recommendations, a well-structured and comprehensive structured assessment report is formed. It should be noted that the report must clearly specify data sources, comparison methods, and scoring criteria to ensure traceability and usability, thereby providing direct data support for subsequent optimization decisions.

[0077] like Figure 4 The diagram shown is another flowchart illustrating a method for assessing the reliability of high-voltage cable operation status according to an embodiment of the present invention. (Refer to...) Figure 4 According to an embodiment of the present invention, a method for assessing the reliability of high-voltage cable operation status, after step S4, further includes the following step: S5. Based on the dynamic matching relationship between multi-dimensional reliability weight configuration and the operating environment characteristics of high-voltage cables, the structured assessment report is transformed into reliability optimization suggestions for the operating status of high-voltage cables.

[0078] Specifically, step S5 includes: 1) Based on the deviation and historical trend of each reliability indicator in the structured report, the optimal weight combination for multi-dimensional reliability is calculated through dynamic weight allocation; Based on the deviation between the actual and ideal values ​​of each reliability indicator in the structured assessment report, and combined with the historical trends of the indicators, an adaptive dynamic weight allocation algorithm is used to calculate the optimal weight combination.

[0079] Specifically, the adaptive dynamic weight adjustment algorithm is characterized by the following formula: in, Indicates the first The optimal weights for each reliability dimension Indicates the first The variance of the historical loss function for each indicator. Indicates the first The weighted fusion value of the deviation of each indicator and its historical trend.

[0080] Furthermore, we have: in, , The balance coefficient representing the trend and the degree of deviation ( The default value is 0.6. The default value is 0.4. Indicates the first The actual value of each indicator Indicates the first The ideal value of each indicator Indicates the first Historical trends of each indicator Indicates the first Each indicator Rate of change over time.

[0081] This formula assigns higher weights to indicators with large deviations and unstable trends, ultimately outputting the optimal weight combination for dimensions such as communication quality and positioning accuracy.

[0082] 2) Identify the performance bottleneck of each reliability dimension based on the optimal weight combination of multiple reliability dimensions; By utilizing the optimal weight combination of multi-dimensional reliability, performance bottlenecks are identified through a performance bottleneck identification function. Specifically, the performance bottleneck identification function is characterized by the following formula: in, This represents a function for identifying performance bottlenecks. The function is used to filter out the dimension with the largest product of "optimal weight × relative deviation", which is the most critical performance bottleneck at present; for the other dimensions, they are sorted by product value to determine the secondary bottlenecks.

[0083] 3) For the performance bottleneck of each reliability dimension, the dynamic matching relationship between the multi-dimensional reliability weight configuration and the operating environment characteristics of the high-voltage cable is invoked. Combined with the expert knowledge base rules and the prediction results of the machine learning model, reliability optimization suggestions for the corresponding performance bottleneck are generated.

[0084] Specifically, for the identified performance bottlenecks in each dimension, the dynamic mapping relationship between multi-dimensional reliability weight configuration and runtime environment characteristics is invoked. ( This represents an environmental feature vector, including parameters such as temperature, humidity, and signal coverage, and a weight adjustment strategy that is matched to adapt to the current environment.

[0085] Furthermore, by combining the expert knowledge base with the machine learning model fusion mechanism, targeted expert experience is extracted through the expert knowledge base rule matching formula. The expert knowledge base rule matching formula is shown below: in, This represents the score for matching expert knowledge base rules. This represents the total number of expert rules. Indicates the first The confidence weight of each rule This indicates an indicator function that takes the value 1 when the rule is met.

[0086] The optimization direction prediction is output through the machine learning model prediction formula, which is shown below: in, This represents the predicted direction of optimization for the machine learning model. This represents the Sigmoid activation function. Indicates the first A feature transformation function, This represents the input feature vector. Indicates the first The weight coefficients of the feature transformation function, This represents the total number of characteristic transformation functions. This represents the bias term of the machine learning model.

[0087] Finally, by integrating the results through a weighted fusion formula, specific optimization suggestions are generated, including parameter adjustment ranges (such as the communication module transmission power adjustment range), implementation priorities (prioritizing key bottlenecks), and expected effects (such as a 10% reduction in bit error rate). Power adjustment suggestions can be generated for communication quality bottlenecks, and algorithm parameter optimization suggestions can be generated for positioning accuracy bottlenecks, ensuring that the suggestions are targeted and feasible.

[0088] The weighted fusion formula is as follows: in, This represents the weighted fusion of expert experience and model predictions to optimize the decision-making process. This indicates the fusion weight (the default value is 0.5, which can be adjusted based on historical fusion results).

[0089] This invention discloses a method for assessing the reliability of high-voltage cable operation status. Its composite neural network architecture, through a combination of multi-layer LSTM hidden layers and batch normalization layers, accurately captures the temporal correlation characteristics of high-voltage cable operation data. Combined with nonlinear mapping of fully connected layers and parameter optimization of multi-objective loss functions, it reduces numerical errors in reliability assessment and improves the consistency of temporal trends. Simultaneously, it enhances the transparency of model decision-making through KL divergence and information entropy constraints, solving the "black box" problem of traditional deep learning models. Dynamic weight allocation transforms assessment results into targeted optimization suggestions, forming a complete process control of "data acquisition - model evaluation - parameter optimization - feedback closed loop." This significantly improves the intelligent operation and maintenance response efficiency of high-voltage cables, while enhancing the reliability and security of power grid supply, providing technical support for the efficient and stable operation of smart grids.

[0090] like Figure 5 The diagram shown is a structural schematic of a high-voltage cable operating status reliability assessment system according to an embodiment of the present invention. (Refer to...) Figure 5 An embodiment of the present invention provides a high-voltage cable operating status reliability assessment system, comprising: Data acquisition module 01 is used to acquire multi-source heterogeneous data generated during the operation of high-voltage cables and obtain a standardized multimodal dataset; The sample set generation module 02 is used to transform the standardized multimodal dataset into a standardized evaluation dataset with reference to a pre-built integrity evaluation system, and to generate training sample sets and validation sample sets based on the standardized evaluation dataset; Architecture training module 03 is used to construct a composite neural network architecture that includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer, and to train the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. The feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on the multi-objective loss function and update the network weights through backpropagation based on the loss value. The report generation module 04 is used to compare the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

[0091] It should be noted that each module in the aforementioned high-voltage cable operational status reliability assessment system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module. For specific limitations regarding the high-voltage cable operational status reliability assessment system, please refer to the limitations of the high-voltage cable operational status reliability assessment method described above; both have the same function and role, and will not be repeated here.

[0092] In summary, the present invention provides a method and system for assessing the reliability of high-voltage cable operation status. Its composite neural network architecture, through a combination of multi-layer LSTM hidden layers and batch normalization layers, accurately captures the temporal correlation characteristics of high-voltage cable operation data. Combined with nonlinear mapping of fully connected layers and parameter optimization of multi-objective loss functions, it reduces numerical errors in reliability assessment and improves the consistency of temporal trends. Simultaneously, it enhances the transparency of model decision-making through KL divergence and information entropy constraints, solving the "black box" problem of traditional deep learning models. Dynamic weight allocation transforms the assessment results into targeted optimization suggestions, forming a complete process control of "data acquisition - model evaluation - parameter optimization - feedback closed loop," significantly improving the intelligent operation and maintenance response efficiency of high-voltage cables, while enhancing the reliability and security of power grid supply, providing technical support for the efficient and stable operation of smart grids.

[0093] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0094] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for assessing the reliability of high-voltage cable operation, characterized in that, include: Acquire multi-source heterogeneous data generated during the operation of high-voltage cables to obtain a standardized multimodal dataset; The standardized multimodal dataset is transformed into a standardized evaluation dataset according to a pre-built integrity evaluation system, and training sample sets and validation sample sets are generated based on the standardized evaluation dataset; A composite neural network architecture comprising a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer is constructed, and the composite neural network architecture is trained based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. The feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on a multi-objective loss function and update the network weights through backpropagation based on the loss value. The evaluation result vector is compared with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

2. The high-voltage cable operating status reliability assessment method according to claim 1, characterized in that, After comparing the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status, the process further includes: Based on the dynamic matching relationship between multi-dimensional reliability weight configuration and the operating environment characteristics of high-voltage cables, the structured assessment report is transformed into reliability optimization suggestions for the operating status of high-voltage cables.

3. The high-voltage cable operating status reliability assessment method according to claim 1, characterized in that, The acquisition of multi-source heterogeneous data generated during the operation of high-voltage cables to obtain a standardized multimodal dataset includes: By deploying a distributed acquisition network at high-voltage cable monitoring terminals, communication quality indicators, location data, and multi-physical quantity monitoring data generated during the operation of high-voltage cables can be collected in real time. The communication quality indicators, the positioning data, and the multi-physical quantity monitoring data are preprocessed to obtain a standardized multimodal dataset.

4. The high-voltage cable operating status reliability assessment method according to claim 1, characterized in that, The step of transforming the standardized multimodal dataset into a standardized evaluation dataset according to a pre-constructed integrity evaluation system, and generating training and validation sample sets based on the standardized evaluation dataset, includes: Construct a completeness assessment system that includes data quality verification, time series alignment, feature dimension unification, and sample label annotation; Based on the aforementioned integrity assessment system, the standardized multimodal dataset is transformed into a standardized assessment dataset that meets the requirements of supervised training. A sliding window algorithm is used to extract time-related sample fragments from the standardized evaluation dataset, and the time-related sample fragments are then subjected to structured processing and data augmentation to generate a training sample set. The standardized evaluation dataset is stratified and sampled to obtain a sample set covering multiple typical operating conditions. Based on the training sample set, the sample set is temporally isolated without overlap to obtain a validation sample set.

5. The high-voltage cable operating status reliability assessment method according to claim 1, characterized in that, The construction of the composite neural network architecture, which includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer, includes: The feature extraction layer consists of multiple LSTM hidden layers and batch normalization layers. Each LSTM hidden layer is followed by a batch normalization layer, and each LSTM hidden layer is configured with a different number of feature units. The nonlinear transformation layer is constructed by multiple fully connected layers, wherein each fully connected layer is configured with a different number of feature units; The parameter optimization layer is constructed based on a multi-objective loss function that considers multiple dimensions, including the accuracy of output results, the interpretability of decisions, and the stability of reliability assessment results, combined with the gradient descent optimization mechanism. The feature extraction layer, the nonlinear transformation layer, and the parameter optimization layer are connected in series to form a composite neural network architecture.

6. The method for assessing the reliability of high-voltage cable operation status according to claim 5, characterized in that, The process of constructing the multi-objective loss function includes: Based on the mean square error between the predicted and actual operating status values, and combined with the consistency constraint of the time series change trend, an accuracy loss function is constructed. By utilizing the KL divergence between the probability distribution of the significance of the output results and the probability distribution of the contribution of feature nodes, and combining the information entropy of feature nodes to constrain the transparency of model decision-making, an interpretable loss function is constructed. By analyzing the variance of the output results over a continuous time period and combining the robustness test of the reliability assessment results under different operating conditions, a stability loss function is constructed. The accuracy loss function, the interpretability loss function, and the stability loss function are dynamically weighted and fused to form a multi-objective loss function.

7. The method for assessing the reliability of high-voltage cable operation status according to claim 1, characterized in that, The process of training the composite neural network architecture based on the training sample set yields an evaluation result vector containing multiple reliability dimensions, including: The training sample set is input into the composite neural network architecture, and the feature dimension is expanded through the input layer, transforming the training sample set into several three-dimensional evaluation tensors containing time batch size, feature dimension and number of monitoring points; All the three-dimensional evaluation tensors are input into the feature extraction layer, and temporal features are extracted using forward propagation; The time-series features are input into the nonlinear transformation layer for nonlinear mapping to obtain a high-dimensional feature vector that is adapted to the reliability assessment dimension. The parameter optimization layer calculates the loss value based on the high-dimensional feature vector using a multi-objective loss function, and then updates the network weights via backpropagation using a gradient descent optimization mechanism. After iterative training until the composite neural network architecture converges, the output is an evaluation result vector containing communication quality, positioning accuracy, and multi-physical quantity monitoring reliability.

8. The method for assessing the reliability of high-voltage cable operation status according to claim 1, characterized in that, The evaluation result vector is compared with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis, and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status, including: The evaluation result vector is compared with the validation sample set from multiple dimensions, including numerical consistency test, trend fit analysis and anomaly detection capability assessment, to obtain the comparison results of the corresponding dimensions. The comparison results for each dimension are weighted and integrated to obtain a comprehensive reliability score; Based on the aforementioned comprehensive reliability score, and at least incorporating the results of abnormal event diagnosis and performance trend prediction, a structured assessment report reflecting the reliability of the high-voltage cable's operating status is generated.

9. The method for assessing the reliability of high-voltage cable operation status according to claim 2, characterized in that, The dynamic matching relationship between the multi-dimensional reliability weight configuration and the operating environment characteristics of the high-voltage cable transforms the structured assessment report into reliability optimization suggestions for the operating status of the high-voltage cable, including: Based on the deviation and historical trend of each reliability indicator in the structured report, the optimal weight combination for multi-dimensional reliability is calculated through dynamic weight allocation. Based on the optimal weight combination of the multi-dimensional reliability, identify the performance bottleneck of each reliability dimension; For each performance bottleneck in the reliability dimension, the dynamic matching relationship between the multi-dimensional reliability weight configuration and the operating environment characteristics of the high-voltage cable is invoked. Combined with the expert knowledge base rules and the prediction results of the machine learning model, reliability optimization suggestions corresponding to the performance bottleneck are generated.

10. A reliability assessment system for the operating status of high-voltage cables, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data generated during the operation of high-voltage cables and obtain a standardized multimodal dataset. The sample set generation module is used to transform the standardized multimodal dataset into a standardized evaluation dataset with reference to a pre-built integrity evaluation system, and to generate training sample sets and validation sample sets based on the standardized evaluation dataset; An architecture training module is used to construct a composite neural network architecture that includes a feature extraction layer, a nonlinear transformation layer, and a parameter optimization layer, and to train the composite neural network architecture based on the training sample set to obtain an evaluation result vector containing multiple reliability dimensions. The feature extraction layer is configured to extract temporal features using forward propagation, the nonlinear transformation layer is configured to perform nonlinear mapping on the extracted temporal features, and the parameter optimization layer is configured to calculate the loss value based on a multi-objective loss function and update the network weights through backpropagation based on the loss value. The report generation module is used to compare the evaluation result vector with the verification sample set from multiple dimensions, including numerical consistency verification, trend consistency analysis and anomaly detection capability assessment, to obtain a structured evaluation report reflecting the reliability of the high-voltage cable's operating status.

Images