Cable state multi-source data fusion and prediction system

Abstract

The invention relates to the technical field of power cable intelligent monitoring, and aims to solve the technical problems that in an existing cable state monitoring system, multi-source heterogeneous data fusion is insufficient, the coupling relation between physical quantities is difficult to reveal, and insulation aging, current-carrying capacity and fault probability cannot be quantitatively predicted according to evaluation results. The invention discloses a cable state multi-source data fusion and prediction system. The system comprises a data acquisition module, a feature processing module, a prediction analysis module and a cloud management platform. The system collects partial discharge, surface temperature and vibration data of the cable through the distributed terminal; performing spectrogram decomposition and fusion of the multi-source data by adopting a non-subsampled contourlet transform algorithm to generate multi-scale fusion features; building a prediction model based on a particle swarm optimization least square support vector machine, and outputting an insulation aging state prediction value, a current-carrying capability evaluation value and a fault probability prediction value of the cable; and data management and full-life-cycle service are realized through the cloud platform.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power cables, specifically a multi-source data fusion and prediction system for cable status. Background Technology

[0002] Current power cable condition monitoring primarily relies on deploying multiple sensors to independently collect data. Existing solutions typically perform threshold alarms or simple statistical analyses on these multi-source monitoring data separately, lacking effective correlation and fusion between different data types. The overall system presents itself as a combination of decentralized data acquisition and independent alarm functions.

[0003] Such solutions have shortcomings. Isolated or superficial analysis of heterogeneous data such as ultrasound, temperature, and vibration makes it difficult to reveal the intrinsic correlations and coupling mechanisms between different physical quantities, resulting in insufficient ability to identify early-stage latent defects and complex aging states in cables. At the prediction level, existing methods mostly rely on threshold warnings for single indicators or traditional statistical models, with outputs often being binary judgments of "normal" or "abnormal." This fails to quantitatively assess the degree of insulation aging, real-time current-carrying capacity, and probability of failure, making it difficult to meet the decision-making needs of precise operation and maintenance and lifespan management.

[0004] The challenge lies in how to perform deep feature extraction and effective fusion of multi-source heterogeneous monitoring data for cables to form a comprehensive feature set that can fully characterize their health status. Simultaneously, it is necessary to establish an intelligent model capable of high-precision, multi-dimensional state quantification prediction based on the fused features, outputting specific insulation status, current-carrying capacity, and fault probability assessment values. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source data fusion and prediction system for cable status to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a multi-source data fusion and prediction system for cable status, the system comprising:

[0007] The data acquisition module collects a multi-source monitoring data set of the target cable by deploying multiple distributed intelligent monitoring terminals. The multi-source monitoring data set includes partial discharge ultrasonic signals, cable surface temperature distribution data, and vibration time series data.

[0008] The feature processing module performs spectral decomposition and fusion processing on the multi-source monitoring data set by running a non-subsampled contour wave transform algorithm to generate multi-scale fusion features of the target cable. The multi-scale fusion features include partial discharge energy spectrum, temperature spatial gradient spectrum and vibration harmonic component spectrum.

[0009] The predictive analysis module performs cable operation status prediction processing on the multi-scale fusion features based on the particle swarm optimization least squares support vector machine model, generating the insulation aging status prediction value, current carrying capacity assessment value, and fault occurrence probability prediction value of the target cable.

[0010] The cloud management platform module integrates a cloud data platform to store and manage the predicted values ​​of insulation aging status, current carrying capacity assessment values, and fault occurrence probability prediction values. The cloud data platform provides full life cycle management services for cables, and the full life cycle management services calculate the estimated remaining life of the cable based on the predicted values ​​of insulation aging status.

[0011] Preferably, the step of performing spectral decomposition and fusion processing on the multi-source monitoring data set by running a non-subsampled contour wave transform algorithm to generate multi-scale fusion features of the target cable includes:

[0012] The partial discharge ultrasonic signal is input into a non-subsampled contour wave transform algorithm, and multi-scale directional subband coefficients of the partial discharge signal are generated through a multi-level decomposition process. Energy statistical processing is performed on the multi-scale directional subband coefficients to generate the partial discharge energy spectrum.

[0013] The cable surface temperature distribution data is divided into spatial grids. Non-subsampled profile wave transform is performed on the temperature time series data of each spatial grid to extract multi-scale spatial frequency features of temperature change. The temperature change rate between adjacent spatial grids is calculated based on the multi-scale spatial frequency features to generate the temperature spatial gradient spectrum.

[0014] The vibration time series data is input into a non-subsampled contour wave transform algorithm to decompose the vibration signal into time-frequency representations of the dominant frequency component and each harmonic component. The amplitude and phase of each harmonic component are tracked and analyzed to generate the vibration frequency component spectrum.

[0015] The partial discharge energy spectrum, the temperature spatial gradient spectrum, and the vibration frequency component spectrum are subjected to time-domain alignment and feature layer fusion processing to generate the multi-scale fused feature containing the three-state correlation information of discharge, heat, and vibration.

[0016] Preferably, the step of performing time-domain alignment and feature layer fusion processing on the partial discharge energy spectrum, the temperature spatial gradient spectrum, and the vibration frequency component spectrum includes:

[0017] Based on the data acquisition timestamp, the time-frequency unit of the partial discharge energy spectrum, the spatial grid unit of the temperature spatial gradient spectrum, and the time sequence unit of the vibration frequency component spectrum are synchronized and matched to establish a multi-source data sequence with a unified time baseline.

[0018] Under a unified time baseline, the time delay correlation coefficient between the peak value of the partial discharge energy spectrum and the extreme value of the temperature spatial gradient spectrum is calculated to generate a discharge-thermal correlation delay index.

[0019] Under a unified time baseline, the correlation between the distortion rate change of the vibration frequency component spectrum and the energy accumulation of the partial discharge energy spectrum is analyzed to generate a vibration-discharge linkage index.

[0020] The discharge-thermal correlation delay index and the vibration-discharge linkage index are weighted and superimposed according to time windows to generate the multi-scale fusion feature that characterizes the multi-physics coupling relationship of the cable state.

[0021] Preferably, the process of predicting cable operating status based on the multi-scale fused features using a particle swarm optimization least squares support vector machine model includes:

[0022] The multi-scale fused features are input into the input layer of the particle swarm optimization least squares support vector machine model;

[0023] In the particle swarm optimization least squares support vector machine model, the particle swarm optimization algorithm is used to iteratively optimize the penalty parameter and kernel function width of the least squares support vector machine model to obtain the optimal combination of model parameters;

[0024] The prediction core of the least squares support vector machine model is configured using the optimal model parameter combination;

[0025] The multi-scale fusion features are input into the configured prediction core, regression prediction calculation is performed, and a continuous prediction sequence about the cable state is output.

[0026] The continuous prediction sequence is post-processed to obtain the predicted value of insulation aging state, the current carrying capacity assessment value, and the predicted value of fault occurrence probability.

[0027] Preferably, the step of using the particle swarm optimization algorithm to iteratively optimize the penalty parameters and kernel width of the least squares support vector machine model to obtain the optimal combination of model parameters includes:

[0028] Initialize the particle population for the particle swarm optimization algorithm, where each particle represents a set of parameters including the penalty parameter and the kernel function width;

[0029] Calculate the prediction error of the parameter combination corresponding to each particle on the least squares support vector machine model validation set, and use the prediction error as the fitness value of each particle;

[0030] Compare the current particle fitness value with the individual's historical best fitness value, and update the individual historical best parameter combination for each particle;

[0031] Compare the current fitness values ​​of all particles with the swarm's historical best fitness values, and update the swarm's historical best parameter combination.

[0032] Calculate the particle's velocity and position for the next iteration based on the particle's current velocity, the individual's historical best parameter combination, and the group's historical best parameter combination.

[0033] Repeat the calculation of fitness value, update of individual historical best, update of group historical best and update of particle velocity and position until the maximum number of iterations is reached or the fitness value converges;

[0034] The final group of historically optimal parameter combinations is used as the optimal model parameter combination.

[0035] Preferably, the step of post-processing the continuous prediction sequence to obtain the predicted value of insulation aging state, the assessed value of current carrying capacity, and the predicted value of fault occurrence probability includes:

[0036] The prediction component corresponding to the insulation state is extracted from the continuous prediction sequence, and the prediction component is compared with the insulation aging grading threshold to be mapped to a specific insulation deterioration level, thereby generating the insulation aging state prediction value.

[0037] The predicted components corresponding to the current-carrying characteristics are extracted from the continuous prediction sequence. Combined with the cable conductor material and cross-sectional area parameters, the maximum safe current value under the predicted state is calculated to generate the current-carrying capacity assessment value.

[0038] The prediction component corresponding to the fault risk is extracted from the continuous prediction sequence, and the prediction component is converted into a probability value between 0 and 1 by applying the fault probability transformation function to generate the fault occurrence probability prediction value.

[0039] Preferably, the cloud data platform provides a panoramic status monitoring service, including:

[0040] It receives and visualizes the raw multi-source monitoring data set from the distributed intelligent monitoring terminal in real time;

[0041] The multi-scale fusion features generated by the non-subsampled contour wave transform algorithm are received and visualized in real time.

[0042] The predicted values ​​of insulation aging status, current carrying capacity assessment value, and fault occurrence probability are received and visualized in real time by the particle swarm optimization least squares support vector machine model.

[0043] The original multi-source monitoring data set, the multi-scale fusion features, the insulation aging state prediction value, the current carrying capacity assessment value, and the fault occurrence probability prediction value are overlaid on a unified geographic information view and a time axis view to generate a panoramic monitoring screen of cable status.

[0044] Preferably, the cloud data platform also provides a fault early warning service, including:

[0045] Set an early warning threshold for the predicted insulation aging state value, an early warning threshold for the current carrying capacity assessment value, and an early warning threshold for the predicted fault occurrence probability value.

[0046] The predicted value of the insulation aging state acquired in real time is compared with the corresponding early warning threshold. If the threshold is exceeded, an insulation aging early warning signal is generated.

[0047] The real-time current carrying capacity assessment value is compared with the corresponding warning threshold. If it is lower than the threshold, an overload risk warning signal is generated.

[0048] The predicted probability of failure is compared with the real-time acquired value and the corresponding warning threshold. If the value exceeds the threshold, a failure probability warning signal is generated.

[0049] The insulation aging warning signal, the overload risk warning signal, and the fault probability warning signal are aggregated and combined with cable location information to generate a graded fault warning report.

[0050] Preferably, the cloud data platform also provides micro-application services with predictive and extrapolation functions, including:

[0051] The system receives user-inputted simulation conditions, which include future load change scenarios, ambient temperature change scenarios, or maintenance plan scenarios.

[0052] Based on the prediction model corresponding to the predicted value of insulation aging state, the current carrying capacity assessment value, and the predicted value of fault occurrence probability, the inference conditions are used as the input boundary of the prediction model for forward simulation calculation.

[0053] Calculate the insulation aging state evolution trajectory, current carrying capacity change curve, and fault occurrence probability time-varying spectrum of the cable under the above simulation conditions;

[0054] The evolution trajectory of the insulation aging state, the change curve of the current carrying capacity, and the time-varying spectrum of the fault occurrence probability are visualized and displayed to generate a report on the future development trend of the cable state.

[0055] Preferably, the full lifecycle management service further includes:

[0056] Based on the estimated remaining life of the cable and the predicted probability of failure, a preventative maintenance schedule is developed.

[0057] Based on the historical trends in predicted insulation aging conditions, assess the insulation degradation rate of cables at different operational stages.

[0058] By integrating cable commissioning time, historical load records, historical fault records, and environmental parameters, a personalized full life-cycle profile for the cable is constructed.

[0059] Based on the individualized full life-cycle profile of the cable and its current predicted status, a comprehensive decision-making recommendation is made regarding cable replacement, repair, or continued operation.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] A non-subsampled contour wave transform algorithm is employed to perform unified spectral decomposition and fusion processing on partial discharge ultrasonic signals, cable surface temperature distribution data, and vibration time-series data. This algorithm possesses translation invariance and multi-directional selection capabilities, enabling the analysis of one-dimensional time-series signals and two-dimensional spatial distribution data within the same multi-scale framework. Through joint transformation and fusion of these heterogeneous data, multi-scale fused features are generated, including partial discharge energy spectra, temperature spatial gradient spectra, and vibration frequency component spectra. This process deeply explores the coupling relationships and intrinsic connections between different physical phenomena, transforming seemingly independent data into a highly condensed and strongly correlated set of state features. This provides high-quality input features that are unavailable through traditional methods, comprehensively reflecting the complex operating state of the cable, for subsequent accurate prediction.

[0062] The particle swarm optimization algorithm is used to automatically optimize the parameters of the least squares support vector machine model, constructing an optimized prediction model to handle the aforementioned multi-scale fusion features. The particle swarm optimization algorithm efficiently determines the optimal parameter combination of the least squares support vector machine, improving the model's convergence speed and generalization performance. This optimized model can simultaneously process fusion features and directly output three quantitative indicators with clear engineering significance: predicted insulation aging status, current-carrying capacity assessment, and predicted fault occurrence probability. This design achieves a precise mapping from composite features to multi-dimensional state quantities. The prediction results are no longer simple binary alarms, but rather a refined and digital description of the cable's current insulation condition, load-carrying capacity, and risk probability, enabling operation and maintenance decisions to be based on more specific and quantitative assessments. Attached Figure Description

[0063] Figure 1 This is a timing diagram of the cable status multi-source data fusion and prediction system described in this invention;

[0064] Figure 2A flowchart for generating multi-scale fusion features;

[0065] Figure 3 A flowchart for predicting cable operating status;

[0066] Figure 4 A time-series correlation analysis diagram of cable multi-source condition monitoring data;

[0067] Figure 5 A state classification analysis diagram of the multi-scale fusion characteristics of cables. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Please see Figure 1 This invention provides a multi-source data fusion and prediction system for cable status. The system includes: a data acquisition module deploying multiple distributed intelligent monitoring terminals installed at key monitoring points of the target cable to simultaneously acquire partial discharge ultrasonic signals, cable surface temperature distribution data, and vibration time-series data, thus forming a multi-source monitoring data set; a feature processing module receiving this multi-source monitoring data set and running a non-subsampled contour wave transform algorithm to perform spectral decomposition and fusion processing, generating multi-scale fused features of the target cable including partial discharge energy spectrum, temperature spatial gradient spectrum, and vibration frequency component spectrum; a prediction analysis module with a built-in particle swarm optimization least squares support vector machine model, receiving the multi-scale fused features as input, and calculating and generating predicted values ​​for the insulation aging status, current carrying capacity, and fault probability of the target cable; and a cloud management platform module integrating a cloud data platform, receiving and storing the predicted values ​​for insulation aging status, current carrying capacity, and fault probability, and calculating the estimated remaining lifespan of the cable based on the predicted insulation aging status, thereby providing full life-cycle management services for distribution cables.

[0070] In one embodiment of the present invention, see [reference] Figure 2The feature processing module runs a non-subsampled profile wave transform algorithm to perform spectral decomposition and fusion processing on the multi-source monitoring data set, generating multi-scale fused features of the target cable. Partial discharge ultrasonic signals are input into the non-subsampled profile wave transform algorithm, which generates multi-scale directional sub-band coefficients of the partial discharge signal through a multi-level decomposition process. These multi-scale directional sub-band coefficients are then subjected to energy statistical processing to generate the local discharge energy spectrum. Cable surface temperature distribution data is divided into predefined spatial grids. A non-subsampled profile wave transform is performed on the temperature time-series data within each spatial grid to extract multi-scale spatial frequency features of temperature changes. Based on these multi-scale spatial frequency features, the temperature change rate between adjacent grids is calculated to generate a temperature spatial gradient spectrum. Vibration time-series data is input into the non-subsampled profile wave transform algorithm, which decomposes the fundamental component and time-frequency representations of each harmonic component of the vibration signal. The amplitude and phase of each harmonic component are tracked and analyzed to generate a vibration frequency component spectrum. The partial discharge energy spectrum, temperature spatial gradient spectrum, and vibration frequency component spectrum are subjected to time-domain alignment and feature-layer fusion processing to generate multi-scale fused features containing information on the three states of discharge, heat, and vibration. In the temporal alignment and feature layer fusion process, the time-frequency units of the partial discharge energy spectrum, the spatial grid units of the temperature spatial gradient spectrum, and the temporal units of the vibration frequency component spectrum are synchronously matched according to the data acquisition timestamps to establish a multi-source data sequence with a unified time baseline. Under the unified time baseline, the time delay correlation coefficient between the peak value of the partial discharge energy spectrum and the extreme value of the temperature spatial gradient spectrum is calculated, generating a discharge-thermal correlation delay index. Under the unified time baseline, the linkage between the distortion rate change of the vibration frequency component spectrum and the energy accumulation of the partial discharge energy spectrum is analyzed to generate a vibration-discharge linkage index. The discharge-thermal correlation delay index and the vibration-discharge linkage index are weighted and superimposed according to a preset time window to finally generate a multi-scale fused feature characterizing the multi-physics coupling relationship of the cable state.

[0071] In practical implementation, the feature processing module runs a non-subsampled profile wave transform algorithm to perform spectral decomposition and fusion processing on the multi-source monitoring data set, generating multi-scale fused features of the target cable. Partial discharge ultrasonic signals are input into the non-subsampled profile wave transform algorithm, which generates multi-scale directional sub-band coefficients of the partial discharge signal through a multi-level decomposition process. Energy statistical processing is then performed on these multi-scale directional sub-band coefficients to generate a partial discharge energy spectrum. The sum of squares of each directional sub-band coefficient is calculated as the energy representation. Cable surface temperature distribution data is divided into spatial grids. A non-subsampled profile wave transform is performed on the temperature time-series data of each spatial grid to extract multi-scale spatial frequency features of temperature changes. The rate of temperature change between adjacent grids is calculated based on these multi-scale spatial frequency features, generating a temperature spatial gradient spectrum. The rate of temperature change is obtained through differential calculation of the amplitudes of corresponding frequency components between adjacent grids. Vibration time-series data is input into a non-subsampled contour wave transform algorithm to decompose the fundamental component and each harmonic component of the vibration signal into time-frequency representations. The amplitude and phase of each harmonic component are tracked and analyzed to generate a vibration frequency component spectrum. The amplitude and phase curves of each harmonic over time are tracked and recorded. The partial discharge energy spectrum, temperature spatial gradient spectrum, and vibration harmonic component spectrum are time-domain aligned and feature layer fused to generate a multi-scale fused feature containing information on the three states of discharge and thermal vibration.

[0072] In some embodiments, the temporal alignment and feature layer fusion processing synchronizes and matches the time-frequency units of the partial discharge energy spectrum, the spatial grid units of the temperature spatial gradient spectrum, and the temporal units of the vibration frequency component spectrum based on the data acquisition timestamp, establishing a multi-source data sequence with a unified time baseline. The synchronization matching process uses interpolation to align data with different sampling rates to a common timestamp. Under the unified time baseline, the time delay correlation coefficient between the peak value of the partial discharge energy spectrum and the extreme value of the temperature spatial gradient spectrum is calculated to generate a discharge-thermal correlation delay index. The time delay correlation coefficient measures the linear correlation between the peak sequence and the extreme value sequence at different time offsets. Under the unified time baseline, the linkage between the distortion rate change of the vibration frequency component spectrum and the energy accumulation of the partial discharge energy spectrum is analyzed to generate a vibration-discharge linkage index. The linkage is quantified by calculating the mutual information between the distortion rate sequence and the energy accumulation sequence. It can be understood that the discharge-thermal correlation delay index and the vibration-discharge linkage index are weighted and superimposed according to time windows to generate a multi-scale fusion feature characterizing the multi-physics coupling relationship of the cable state. The weighted superposition formula is:

[0073]

[0074] in: This represents the value of the multi-scale fusion feature in the k-th time window. This represents the value of the discharge-thermal correlation delay index in the k-th time window. This represents the value of the vibration-discharge linkage index in the k-th time window. and Preset weighting coefficients and satisfying .

[0075] In practical implementation, the multi-level decomposition process of the non-subsampled profile wave transform algorithm includes multi-scale decomposition and multi-directional decomposition steps. After processing by the non-subsampled profile wave transform algorithm, the partial discharge ultrasonic signal outputs sub-band coefficient matrices at multiple scales and directions. It can be understood that the spatial grid division in the generation of the temperature spatial gradient spectrum is based on the physical location of the temperature measurement points on the cable surface, with each grid associated with a set of time-series data collected by a temperature sensor. In the generation of the vibration frequency component spectrum, the harmonic components cover the frequency range from the second harmonic to the preset highest harmonic, and amplitude and phase tracking analysis are performed using a sliding time window.

[0076] Optionally, before synchronization matching, the temporal alignment and feature layer fusion processing normalizes the partial discharge energy spectrum, temperature spatial gradient spectrum, and vibration frequency component spectrum. This normalization maps each spectrum to a zero-mean, unit-variance distribution. In some embodiments, the discharge-thermal correlation delay index is calculated using a cross-correlation function method, scanning a preset time delay range to find the time offset corresponding to the maximum correlation coefficient. The vibration-discharge linkage index is analyzed using a temporal convolution method, convolving the distortion rate change sequence with the energy accumulation sequence to extract the linkage pattern. In specific implementations, the generation of multi-scale fusion features also includes smoothing filtering of the weighted superposition results, using a moving average filter to eliminate the influence of random noise. Optionally, after weighted superposition, the feature layer fusion processing performs principal component analysis to reduce the feature dimensionality. The principal component analysis retains principal component components with a cumulative contribution rate exceeding a preset threshold.

[0077] In one embodiment of the present invention, see [reference] Figure 3 The predictive analysis module uses a particle swarm optimization (PSO) least squares support vector machine (LSVM) model to predict cable operating conditions based on multi-scale fused features. The multi-scale fused features are input into the input layer of the PSO least squares LSVM model. In the PSO least squares LSVM model, the Particle Swarm Optimization (PSO) algorithm iteratively optimizes the penalty parameters and kernel width to obtain the optimal model parameter combination. The obtained optimal model parameter combination is then used to configure the prediction kernel of the LSVM model. The multi-scale fused features are input into the configured prediction kernel, and regression prediction calculations are performed, outputting a continuous prediction sequence about the cable condition. This continuous prediction sequence is post-processed to obtain predicted values ​​for insulation aging status, current carrying capacity, and fault probability.

[0078] In practical implementation, the predictive analysis module uses a particle swarm optimization (PSO) least squares support vector machine (LSVM) model to predict cable operating status based on multi-scale fused features. These multi-scale fused features are transmitted as input vectors to the input layer of the PSO least squares LSVM model, where a standardization preprocessing is performed. Within the PSO least squares LSVM model, the Particle Swarm Optimization (PSO) algorithm iteratively optimizes the penalty parameters and kernel width to obtain the optimal model parameter combination. This iterative optimization process is completed within an independent parameter optimization loop. The optimal model parameter combination is then used to configure the prediction core of the LSVM model. This configuration involves loading the penalty parameters and kernel width into the LSVM model's solver. The multi-scale fused features are then input into the configured prediction core, and regression prediction calculations are performed, outputting a continuous prediction sequence regarding the cable state. The regression prediction calculations solve for the mapping relationship between input features and state variables based on the LSVM model. The continuous prediction sequence is post-processed to obtain the predicted values ​​of insulation aging status, current carrying capacity assessment value and fault occurrence probability prediction value. The post-processing operation separates the corresponding numerical components from the continuous prediction sequence according to the preset analysis rules.

[0079] In some embodiments, the input layer of the particle swarm optimization (PSO) least squares support vector machine (SVM) model contains neurons with the same dimension as the multi-scale fusion feature, and each neuron receives one feature component. When the PSO algorithm iteratively optimizes the penalty parameter and kernel width of the least squares SVM model, the search range for the penalty parameter and the kernel width is set to a positive real number interval. It can be understood that the configured prediction core is a least squares SVM model with fixed parameters, mathematically expressed as:

[0080]

[0081] in: This indicates that for the input feature vector The predicted output, This indicates the number of support vectors. Indicates the first The Lagrange multipliers corresponding to each support vector Represents the radial basis kernel function. Indicates the first Support vectors, This indicates the bias term.

[0082] In practice, the regression prediction calculation process invokes the built-in solution algorithm of the least squares support vector machine model. This algorithm processes the input multi-scale fusion feature matrix and outputs a continuous prediction sequence. This continuous prediction sequence is a sequence of real numbers that corresponds one-to-one with the input data time points; each value in the sequence represents the model's quantitative estimate of the cable's overall state at that moment. Post-processing uses a dedicated parser to decompose the continuous prediction sequence. Based on the mapping relationship established during the training phase, the parser assigns the sequence values ​​to different state dimensions. Essentially, the insulation aging state prediction, current-carrying capacity assessment, and fault probability prediction are encapsulated in a structured data format after being output from the parser.

[0083] Optionally, the particle swarm optimization least squares support vector machine model performs feature dimension verification before receiving multi-scale fused features. Feature dimension verification ensures that the length of the input vector is consistent with the feature dimension during model training. In some embodiments, regression prediction calculation supports batch processing mode. Batch processing mode inputs all multi-scale fused feature samples within a time window at once and outputs the corresponding continuous prediction sequence blocks. When the continuous prediction sequence is post-processed, the parsing rules are based on a predefined output channel order. The first output channel corresponds to the source data of the insulation aging state prediction value, the second output channel corresponds to the source data of the current carrying capacity assessment value, and the third output channel corresponds to the source data of the fault occurrence probability prediction value.

[0084] In practical implementation, the particle swarm optimization least squares support vector machine model stores the optimal combination of model parameters in a persistent configuration file, which is loaded during the initialization of the prediction core. During the regression prediction calculation performed by the prediction core, the radial basis kernel function calculates the Euclidean distance between the input features and all support vectors and transforms it into a kernel matrix. In the post-processing operation, the parser performs the same separation and assignment logic on the output of each time step of the continuous prediction sequence, generating a time-ordered sequence of insulation aging state predictions, current carrying capacity assessments, and fault probability predictions. Optionally, the continuous prediction sequence is smoothed before being output to the post-processing module. The smoothing filter uses a moving median filter of length five to suppress impulse interference.

[0085] In one embodiment of the present invention, in a particle swarm optimization (PSO) least squares support vector machine (LSVM) model, the PSO algorithm iteratively optimizes the penalty parameter and kernel width of the LSVM model to obtain the optimal model parameter combination. This process initializes a particle population for the PSO algorithm, where each particle represents a parameter combination including the penalty parameter and kernel width. The prediction error of the parameter combination corresponding to each particle on the LSVM model validation set is calculated, and this prediction error is used as the fitness value of each particle. The current particle fitness value is compared with the individual's historical best fitness value, and the individual's historical best parameter combination is updated for each particle. The current fitness values ​​of all particles are compared with the swarm's historical best fitness value, and the swarm's ... When post-processing the continuous prediction sequence to parse and obtain predicted values, the prediction component corresponding to the insulation state is extracted from the continuous prediction sequence. This prediction component is compared with a preset insulation aging grading threshold and mapped to a specific insulation degradation level, generating an insulation aging state prediction value. The prediction component corresponding to the current-carrying characteristics is extracted from the continuous prediction sequence. Combined with the cable conductor material and cross-sectional area parameters, the maximum safe current value under the predicted state is calculated, generating a current-carrying capacity assessment value. The prediction component corresponding to the fault risk is extracted from the continuous prediction sequence. A fault probability transformation function is applied to convert the prediction component into a probability value between 0 and 1, generating a fault occurrence probability prediction value.

[0086] In practical implementation, in the Particle Swarm Optimization (PSO) least squares support vector machine (LSVM) model, the PSO algorithm iteratively optimizes the penalty parameters and kernel width of the LSVM model to obtain the optimal model parameter combination. The PSO algorithm's progress is independent of the forward prediction calculation of the LSVM model. A particle swarm for the PSO algorithm is initialized, containing a preset number of individual particles, each representing a parameter combination including the penalty parameters and kernel width. The prediction error of each particle's parameter combination on the LSVM model validation set is calculated. The validation set is a subset of samples reserved from historical cable state data, and the reciprocal or inverse of the prediction error is used as the fitness value of each particle. The current particle fitness value is compared with the individual's historical best fitness value. If the current fitness value is better, the individual's historical best parameter combination is updated to the current parameter combination. The current fitness values ​​of all particles are compared with the swarm's historical best fitness value. If a particle's current fitness value is better than the swarm's historical best fitness value, the swarm's historical best parameter combination is updated to the particle's parameter combination. Based on the particle's current velocity, the individual's historical best parameter combination, and the group's historical best parameter combination, calculate the particle's velocity and position for the next iteration. The position update formula is:

[0087]

[0088] in: Represents particles In the In the next iteration, at the... The location of 4-dimensional space, Represents particles In the In the next iteration, at the... Velocity in 1st-dimensional space, velocity The fitness value is jointly determined by inertial, individual, and social cognitive components. The process involves repeatedly calculating the fitness value, updating the individual historical best, updating the group's historical best, and updating the particle velocity and position until a preset maximum number of iterations is reached or the fitness value converges. The convergence criterion is that the change in the group's historical best fitness value is less than a threshold across multiple consecutive iterations. The final group historical best parameter combination is output as the optimal model parameter combination for configuring the least squares support vector machine model.

[0089] In some embodiments, when post-processing the continuous prediction sequence to parse and obtain predicted values, a predicted component corresponding to the insulation state is extracted from the continuous prediction sequence, which is a multi-output one-dimensional array. The predicted component corresponding to the insulation state is compared with a preset insulation aging grading threshold, which defines numerical boundaries corresponding to multiple levels from normal, caution, abnormal to severe. A mapping operation categorizes the specific values ​​of the predicted components to the corresponding insulation degradation levels, generating predicted insulation aging state values ​​represented by level identifiers or numerical scores. A predicted component corresponding to the current-carrying characteristics is extracted from the continuous prediction sequence. This predicted component reflects the conductivity reduction factor of the cable conductor in the current overall state. Combining the cable conductor material and cross-sectional area parameters, the maximum safe current value in the predicted state is calculated. The calculation process multiplies the reduction factor by the cable's nominal current-carrying capacity to generate a current-carrying capacity assessment value in amperes. A predicted component corresponding to the fault risk is extracted from the continuous prediction sequence. This fault risk predicted component is an unnormalized raw score. The failure probability transformation function is applied to convert the predicted component corresponding to failure risk into a probability value between 0 and 1. The failure probability transformation function uses the Sigmoid function to generate the predicted value of failure occurrence probability.

[0090] In particle swarm optimization (PSO), each particle's position is represented by a two-dimensional vector. The first dimension represents candidate values ​​for the penalty parameter, and the second dimension represents candidate values ​​for the kernel function width. During fitness calculation, a least-squares support vector machine (SVM) model is temporarily configured using the parameter combinations represented by the particles, and predictions are made on a fixed validation set to calculate the error. The individual's historical best parameter combination and the group's historical best parameter combination may be updated after each iteration, following the principle of replacing the particle with the better fitness value.

[0091] In practice, the insulation aging grading thresholds are stored in a lookup table, and the mapping operation is completed through table lookup and interpolation. The calculation of the current-carrying capacity assessment value also needs to refer to the ambient temperature correction factor, which is obtained from real-time monitoring data. The expression for the fault probability conversion function Sigmoid is:

[0092]

[0093] in: This represents the predicted probability of a failure occurring. This represents the predicted component corresponding to the failure risk extracted from the continuous prediction sequence. and These are the preset scaling and translation parameters.

[0094] Optionally, a velocity constraint is introduced in the particle velocity update calculation to prevent particles from moving too far in a single iteration and missing the optimal region. In some embodiments, the order of the predicted components corresponding to insulation state, current-carrying characteristics, and fault risk in the continuous prediction sequence is fixed, and the post-processing parser extracts them according to a fixed offset. In addition to a level identifier, the insulation aging state prediction value can also output a normalized health index compared to a threshold. In specific implementations, the iterative optimization process of the particle swarm optimization algorithm can be executed in parallel, with the fitness value evaluation of each particle performed simultaneously in different computation threads. The post-processing parsing operation is deterministic; the same continuous prediction sequence input always produces the same insulation aging state prediction value, current-carrying capacity assessment value, and fault occurrence probability prediction value output. It can be understood that if the calculation of the current-carrying capacity assessment value finds the current conductor temperature parameter to be abnormally high, an overload risk flag will be added to the final assessment value. After output, the fault occurrence probability prediction value can be compared with a dynamically adjusted threshold to directly generate a binary risk warning signal.

[0095] In one embodiment of the present invention, a cloud-based data platform provides a panoramic monitoring service. This service receives and visualizes in real-time the raw multi-source monitoring data set from distributed intelligent monitoring terminals. It receives and visualizes in real-time the multi-scale fusion features generated by a non-subsampled contourlet transform algorithm. It also receives and visualizes in real-time the predicted insulation aging status, current-carrying capacity assessment, and fault probability prediction values ​​generated by a particle swarm optimization least squares support vector machine model. The raw multi-source monitoring data set, multi-scale fusion features, and the predicted insulation aging status, current-carrying capacity assessment, and fault probability prediction values ​​are overlaid on a unified geographic information view and a timeline view to generate a panoramic cable status monitoring image. The cloud-based data platform also provides a fault early warning service. Early warning thresholds are set for the predicted insulation aging status, current-carrying capacity assessment, and fault probability prediction. The real-time acquired predicted insulation aging status is compared with the corresponding early warning threshold; if it exceeds the threshold, an insulation aging early warning signal is generated. Similarly, the real-time acquired current-carrying capacity assessment is compared with the corresponding early warning threshold; if it is below the threshold, an overload risk early warning signal is generated. The system compares the real-time predicted fault occurrence probability with the corresponding warning threshold. If the probability exceeds the threshold, a fault probability warning signal is generated. It aggregates insulation aging warning signals, overload risk warning signals, and fault probability warning signals, and combines these with cable location information to generate a tiered fault warning report.

[0096] In practical implementation, the cloud-based data platform provides a panoramic monitoring service. This service receives and visualizes raw multi-source monitoring data sets from distributed intelligent monitoring terminals in real time. The raw multi-source monitoring data sets are pushed to the cloud-based data platform's visualization engine in the form of a data stream. The platform also receives and visualizes multi-scale fusion features generated by a non-subsampled contourlet transform algorithm. These multi-scale fusion features are parsed by the visualization engine in the form of feature vectors along with timestamps. Furthermore, the platform receives and visualizes predicted insulation aging status, current-carrying capacity assessment, and fault probability predictions generated by a particle swarm optimization least squares support vector machine model. These predictions are pushed to the visualization interface as structured data packets. The raw multi-source monitoring data sets, multi-scale fusion features, and predicted insulation aging status, current-carrying capacity assessment, and fault probability predictions are overlaid on a unified geographic information view and a timeline view. The unified geographic information view is rendered based on the actual geographic coordinates of the cable line, while the timeline view provides synchronized time slider control, generating a panoramic cable status monitoring screen. This screen allows operators to simultaneously observe spatial distribution and temporal evolution information.

[0097] In some embodiments, the visualization of the original multi-source monitoring data set adopts a multi-curve comparison chart, with partial discharge ultrasonic signals, cable surface temperature distribution data, and vibration time series data plotted separately on the same time coordinate system. The visualization of multi-scale fusion features uses a spectrum chart or heat map; the partial discharge energy spectrum is displayed as a time-frequency energy map, the temperature spatial gradient spectrum as a color gradient map along the cable path, and the vibration component spectrum as a frequency amplitude bar chart. The visualization of the insulation aging state prediction, current carrying capacity assessment, and fault probability prediction combines a digital dashboard with a trend curve chart; the dashboard displays the current instantaneous value, and the trend curve chart displays historical changes. The overlay display of the unified geographic information view and the time axis view is achieved through layer management technology, allowing operators to select and hide the original data layer, feature layer, or prediction result layer.

[0098] In practical implementation, the cloud data platform provides fault early warning services. These services set early warning thresholds for insulation aging state predictions, current-carrying capacity assessments, and fault occurrence probability predictions. These thresholds are configured by the system administrator based on cable operation procedures or historical statistical data. The real-time acquired insulation aging state predictions are compared with the corresponding early warning thresholds. This comparison is triggered each time a new insulation aging state prediction is received; if the predicted value exceeds the threshold, an insulation aging early warning signal is generated. The real-time acquired current-carrying capacity assessment is compared with the corresponding early warning threshold, which is a minimum allowable current-carrying value. If the current-carrying capacity assessment is below the threshold, an overload risk early warning signal is generated. The real-time acquired fault occurrence probability prediction is compared with the corresponding early warning threshold; if it exceeds the threshold, a fault probability early warning signal is generated. The insulation aging early warning signal, overload risk early warning signal, and fault probability early warning signal are aggregated, and combined with cable location information to generate a graded fault early warning report. This report is coded according to the signal type, severity, and location.

[0099] It is understandable that the configuration of warning thresholds can be accomplished through a threshold management interface, which provides the ability to set independent thresholds for different cable lines or cable sections. Refer to Table 1, which shows a set of warning threshold configurations for a certain type of 10kV cross-linked polyethylene cable.

[0100] Table 1: Early Warning Threshold Configuration Table

[0101]

[0102] The process of generating a graded fault warning report by aggregating insulation aging warning signals, overload risk warning signals, and fault probability warning signals follows a predetermined aggregation logic. The generation logic of the graded fault warning report can be formally represented as follows:

[0103]

[0104] in: This indicates the overall rating value of the final warning report. Indicates the first The quantification value of the severity of the warning signal. Indicates the first Preset weighting coefficients for warning signals. Based on the calculated... If the value falls within a certain range, the warning report will be marked as a "general warning", "important warning" or "emergency warning".

[0105] Optionally, the unified geographic information view of the status panoramic monitoring screen supports clicking on cable segments to query details. The details pop-up window displays the original data curve, fused feature map, and predicted value dashboard for that location point side by side. In some embodiments, the warning signal generated by the fault warning service will trigger the platform's message push mechanism, which will notify relevant maintenance personnel via in-site messages, emails, or SMS. In addition to the comprehensive level, the graded fault warning report also includes a specific list of warning parameters, trigger time, cable identifier, and a summary of suggested handling measures. In specific implementation, the synchronization control of the timeline view ensures that when the user drags the time slider, all data display layers, feature display layers, and prediction result display layers on the geographic information view are synchronously updated to the corresponding state at that time. After the insulation aging warning signal, overload risk warning signal, and fault probability warning signal are generated, their historical records will be stored in the warning log database for subsequent auditing and trend analysis. It can be understood that after the graded fault warning report is generated, it will be automatically associated with the asset file of the cable line and a record will be formed under the warning history tab of the asset file.

[0106] See Figure 4 This is a time-series correlation analysis chart of cable multi-source condition monitoring data. The three data types show a synchronous trend of "increasing-fluctuating-increasing," reflecting the interconnected cable condition pattern of "increased load → increased temperature → enhanced partial discharge → increased vibration amplitude." Around 8:00 AM, all three data types simultaneously reach their peak values, marking a key change point in the cable load / condition. After 10:00 PM, partial discharge, temperature, and vibration all rise rapidly, indicating a deterioration in the cable's operating condition. This chart serves the "multi-source data correlation analysis stage" of cable condition monitoring. Its value lies in verifying the physical field coupling relationship of "heat-discharge-vibration" through the synchronous trend of multiple data sources. The synchronous changes in multi-source data provide correlation characteristics for subsequent predictions of insulation aging and fault probability.

[0107] In one embodiment of the present invention, a cloud data platform provides a micro-application service with predictive simulation capabilities. This service receives simulation conditions input by the user, including future load change scenarios, ambient temperature change scenarios, or maintenance plan scenarios. Based on the prediction model corresponding to the predicted insulation aging state, current carrying capacity assessment value, and fault probability prediction value, the simulation conditions are used as the model input boundary for forward simulation calculation. The evolution trajectory of the cable's insulation aging state, the current carrying capacity change curve, and the time-varying spectrum of the fault probability are calculated under the simulation conditions. The evolution trajectory of the insulation aging state, the current carrying capacity change curve, and the time-varying spectrum of the fault probability are visualized and displayed, generating a report on the future development trend of the cable's condition. The full life cycle management service also includes other functions, such as developing a preventative maintenance schedule based on the estimated remaining cable life and the predicted fault probability value; assessing the insulation degradation rate of the cable at different operating stages based on the changing trend of historical insulation aging state prediction values; and integrating cable commissioning time, historical load records, historical fault records, and environmental parameters to construct an individualized full life cycle profile for the cable. Based on the individualized full life-cycle profile of the cable and its current predicted status, comprehensive decision-making recommendations are made regarding cable replacement, repair, or continued operation.

[0108] In practical implementation, the cloud data platform provides micro-application services for predictive simulation. These services receive simulation conditions input by the user, provided through a graphical interface or configuration files. These conditions include scenarios such as future load changes, ambient temperature changes, or maintenance plans. Based on the prediction model corresponding to the predicted values ​​of insulation aging status, current carrying capacity assessment, and fault probability, the simulation conditions are used as the model input boundaries for forward simulation calculations. These calculations are performed in an isolated computing sandbox created by the cloud data platform. The system calculates the cable's insulation aging status evolution trajectory, current carrying capacity change curve, and fault probability time-varying graph under the simulation conditions. The insulation aging status evolution trajectory describes the path of the predicted insulation status over time, the current carrying capacity change curve describes the path of the assessed current carrying capacity over time, and the fault probability time-varying graph describes the path of the predicted fault probability over time. The evolution trajectory of insulation aging, the curve of change in current carrying capacity, and the time-varying spectrum of fault occurrence probability are visualized and displayed. The visualization is presented in the form of dynamic curves and color gradient graphs, generating a report on the future development trend of cable condition. The report includes status data tables and trend analysis summaries for key time points.

[0109] In some embodiments, the user-input projection conditions for future load change scenarios define the percentage change curve of cable load current relative to the current load over a future period. The ambient temperature change scenario defines a seasonal or diurnal periodic change model of future ambient temperature. The maintenance plan scenario defines the time points for planned power outage maintenance and the assumed cable condition recovery parameters after maintenance. The forward simulation calculation calls the same particle swarm optimization least squares support vector machine model as the real-time prediction, but uses the current multi-scale fusion features as the initial state and feeds the projection conditions as an additional time-varying input sequence into the model. The calculation results of the insulation aging state evolution trajectory, current-carrying capacity change curve, and fault occurrence probability time-varying spectrum are stored in memory as time-series arrays for use by the visualization module.

[0110] In practical implementation, the full lifecycle management service also includes other functions. Based on the estimated remaining cable life and predicted failure probability, a preventative maintenance schedule is developed, and operational plans are formulated to correlate recommended maintenance times with predicted cable inflection points. The insulation degradation rate of cables at different operational stages is assessed based on the changing trends of historical insulation aging state predictions. The insulation degradation rate is obtained by linearly or nonlinearly fitting a sequence of historical insulation aging state predictions. Cable commissioning time, historical load records, historical fault records, and environmental parameters are integrated to construct an individualized full lifecycle profile for each cable. This profile is a structured database record associated with a unique cable asset code. Based on the individualized full lifecycle profile and the current predicted state, comprehensive decision-making recommendations are made for cable replacement, repair, or continued operation. The generation of these comprehensive recommendations utilizes a rule engine to weightedly evaluate multiple assessment dimensions.

[0111] Understandably, when developing a preventative maintenance schedule, the estimated remaining cable life provides a rough timeframe for the end of its life, while the predicted failure probability indicates the level of near-term risk. The maintenance schedule attempts to schedule maintenance activities before the risk significantly increases. The insulation degradation rate of the cable is assessed at different operational stages, which can be divided by years of operation, such as initial, mid-term, and late-term. The insulation degradation rate is the average slope of the curve showing the predicted insulation aging state over time within each stage. An individualized full-lifecycle profile of the cable is constructed, with the timeline as the main thread. The commissioning time is used as the starting event, and historical load records, historical failure records, environmental parameters, and all historical state predictions are stored sequentially as event-related data points.

[0112] Optionally, a comprehensive decision recommendation for cable replacement, repair, or continued operation is provided based on a preset set of evaluation rules. The logical formula for generating the comprehensive decision recommendation is as follows:

[0113]

[0114] in: This indicates the urgency index for maintenance. Indicates the cable's operating time. This represents the estimated remaining lifespan of the cable. This represents the predicted probability of the current failure occurring. This indicates the normalized degree of degradation corresponding to the current predicted insulation aging condition. These are the weighting coefficients for the corresponding items. When maintaining the urgency index... When the "Repair" threshold is exceeded, a repair suggestion is generated; when the "Replacement" threshold is exceeded, a replacement suggestion is generated; otherwise, a suggestion to continue operation and enhance monitoring is generated.

[0115] In some embodiments, the visualization interface provides a timeline controller, allowing users to drag the controller to observe the predicted insulation aging evolution trajectory, current-carrying capacity change curve, and time-varying probability graph of fault occurrence at any future point in time. The cable condition future development trend report supports exporting to a structured document format. The preventative maintenance plan schedule is integrated into the visualization interface of the cable's individualized life-cycle profile in the form of a Gantt chart. The assessment results of the insulation degradation rate are displayed in chart form, comparing the average degradation rate of the cable at different operating stages.

[0116] In practical implementation, forward simulation calculations support both step-by-step extrapolation and one-click extrapolation to a specified future time. The data structure of the individualized cable lifecycle archive is expandable to accommodate new monitoring data types or evaluation indicators. The thresholds and weighting coefficients in the rule set for generating comprehensive decision recommendations can be configured and adjusted by domain experts through the management interface. It can be understood that after the micro-application service of the predictive extrapolation function generates a report on the future development trend of the cable status, this report can be associated with the corresponding individualized cable lifecycle archive and saved as a historical record of an extrapolation analysis. Optionally, multiple maintenance plan scenarios can be set in the extrapolation conditions, and the micro-application service supports parallel simulation calculations and result comparisons for multiple maintenance plan scenarios. In some embodiments, the assessment of insulation degradation rate will eliminate abnormal fluctuations in predicted values ​​caused by sudden faults or abnormal events to obtain a more stable long-term trend. When generating maintenance or replacement recommendations, the comprehensive decision recommendations will simultaneously list the main basis, such as indicating whether it is based on insufficient remaining life, excessively high current fault probability, or excessively rapid insulation degradation rate.

[0117] See Figure 5This is a state classification analysis chart of the multi-scale fusion characteristics of cables. The fault precursor state characteristic value (approximately 0.68) is significantly higher than other states, making it a key indicator for fault prediction; the characteristic values ​​of the three states are relatively similar, serving as supplementary features to assist in state verification. The "fault precursor" characteristic value (0.68) of the vibration frequency component spectrum is a high-priority signal for risk warning. This chart serves the "state classification and risk identification stage" of cable state fusion characteristics. Its value lies in the fact that the partial discharge energy spectrum and the electrical vibration component spectrum are key bases for state classification; and the characteristic differences of different states can strengthen the weight of core features in the state prediction model.

[0118] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0119] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-source data fusion and prediction system for cable status, characterized in that, The system includes: The data acquisition module collects a multi-source monitoring data set of the target cable by deploying multiple distributed intelligent monitoring terminals. The multi-source monitoring data set includes partial discharge ultrasonic signals, cable surface temperature distribution data, and vibration time series data. The feature processing module performs spectral decomposition and fusion processing on the multi-source monitoring data set by running a non-subsampled contour wave transform algorithm to generate multi-scale fusion features of the target cable. The multi-scale fusion features include partial discharge energy spectrum, temperature spatial gradient spectrum and vibration frequency component spectrum. The predictive analysis module performs cable operation status prediction processing on the multi-scale fusion features based on the particle swarm optimization least squares support vector machine model, generating the insulation aging status prediction value, current carrying capacity assessment value, and fault occurrence probability prediction value of the target cable. The cloud management platform module integrates a cloud data platform to store and manage the predicted values ​​of insulation aging status, current carrying capacity assessment values, and fault occurrence probability prediction values. The cloud data platform provides full life cycle management services for cables, and the full life cycle management services calculate the estimated remaining life of the cable based on the predicted values ​​of insulation aging status.

2. The cable condition multi-source data fusion and prediction system according to claim 1, characterized in that, The process of performing spectral decomposition and fusion processing on the multi-source monitoring data set by running a non-subsampled contour wave transform algorithm to generate multi-scale fused features of the target cable includes: The partial discharge ultrasonic signal is input into a non-subsampled contour wave transform algorithm, and multi-scale directional subband coefficients of the partial discharge signal are generated through a multi-level decomposition process. Energy statistical processing is performed on the multi-scale directional subband coefficients to generate the partial discharge energy spectrum. The cable surface temperature distribution data is divided into spatial grids. Non-subsampled profile wave transform is performed on the temperature time series data of each spatial grid to extract multi-scale spatial frequency features of temperature change. The temperature change rate between adjacent spatial grids is calculated based on the multi-scale spatial frequency features to generate the temperature spatial gradient spectrum. The vibration time series data is input into a non-subsampled contour wave transform algorithm to decompose the vibration signal into time-frequency representations of the dominant frequency component and each harmonic component. The amplitude and phase of each harmonic component are tracked and analyzed to generate the vibration frequency component spectrum. The partial discharge energy spectrum, the temperature spatial gradient spectrum, and the vibration frequency component spectrum are subjected to time-domain alignment and feature layer fusion processing to generate the multi-scale fused feature containing the three-state correlation information of discharge, heat, and vibration.

3. The cable condition multi-source data fusion and prediction system according to claim 2, characterized in that, The temporal alignment and feature layer fusion processing of the partial discharge energy spectrum, the temperature spatial gradient spectrum, and the vibration frequency component spectrum includes: Based on the data acquisition timestamp, the time-frequency unit of the partial discharge energy spectrum, the spatial grid unit of the temperature spatial gradient spectrum, and the time sequence unit of the vibration frequency component spectrum are synchronized and matched to establish a multi-source data sequence with a unified time baseline. Under a unified time baseline, the time delay correlation coefficient between the peak value of the partial discharge energy spectrum and the extreme value of the temperature spatial gradient spectrum is calculated to generate a discharge-thermal correlation delay index. Under a unified time baseline, the correlation between the distortion rate change of the vibration frequency component spectrum and the energy accumulation of the partial discharge energy spectrum is analyzed to generate a vibration-discharge linkage index. The discharge vibration-thermal correlation delay index and the vibration-discharge linkage index are weighted and superimposed according to a time window to generate the multi-scale fusion feature that characterizes the multi-physics coupling relationship of the cable state.

4. The cable condition multi-source data fusion and prediction system according to claim 1, characterized in that, The process of predicting cable operating status using the multi-scale fused features based on the particle swarm optimization least squares support vector machine model includes: The multi-scale fused features are input into the input layer of the particle swarm optimization least squares support vector machine model; In the particle swarm optimization least squares support vector machine model, the particle swarm optimization algorithm is used to iteratively optimize the penalty parameter and kernel function width of the least squares support vector machine model to obtain the optimal combination of model parameters; The prediction core of the least squares support vector machine model is configured using the optimal model parameter combination; The multi-scale fusion features are input into the configured prediction core, regression prediction calculation is performed, and a continuous prediction sequence about the cable state is output. The continuous prediction sequence is post-processed to obtain the predicted value of insulation aging state, the current carrying capacity assessment value, and the predicted value of fault occurrence probability.

5. The cable condition multi-source data fusion and prediction system according to claim 4, characterized in that, The step of using particle swarm optimization to iteratively optimize the penalty parameters and kernel width of the least squares support vector machine model to obtain the optimal combination of model parameters includes: Initialize the particle population for the particle swarm optimization algorithm, where each particle represents a set of parameters including the penalty parameter and the kernel function width; Calculate the prediction error of the parameter combination corresponding to each particle on the least squares support vector machine model validation set, and use the prediction error as the fitness value of each particle; Compare the current particle fitness value with the individual's historical best fitness value, and update the individual historical best parameter combination for each particle; Compare the current fitness values ​​of all particles with the swarm's historical best fitness values, and update the swarm's historical best parameter combination. Calculate the particle's velocity and position for the next iteration based on the particle's current velocity, the individual's historical best parameter combination, and the group's historical best parameter combination. Repeat the calculation of fitness value, update of individual historical best, update of group historical best and update of particle velocity and position until the maximum number of iterations is reached or the fitness value converges; The final group of historically optimal parameter combinations is used as the optimal model parameter combination.

6. The cable condition multi-source data fusion and prediction system according to claim 4, characterized in that, The step of post-processing the continuous prediction sequence to obtain the predicted value of insulation aging state, the assessed value of current carrying capacity, and the predicted value of fault occurrence probability includes: The prediction component corresponding to the insulation state is extracted from the continuous prediction sequence, and the prediction component is compared with the insulation aging grading threshold to be mapped to a specific insulation deterioration level, thereby generating the insulation aging state prediction value. The predicted components corresponding to the current-carrying characteristics are extracted from the continuous prediction sequence. Combined with the cable conductor material and cross-sectional area parameters, the maximum safe current value under the predicted state is calculated to generate the current-carrying capacity assessment value. The prediction component corresponding to the fault risk is extracted from the continuous prediction sequence, and the prediction component is converted into a probability value between 0 and 1 by applying the fault probability transformation function to generate the fault occurrence probability prediction value.

7. The cable condition multi-source data fusion and prediction system according to claim 1, characterized in that, The cloud-based data platform also provides a comprehensive status monitoring service, including: It receives and visualizes the raw multi-source monitoring data set from the distributed intelligent monitoring terminal in real time; The multi-scale fusion features generated by the non-subsampled contour wave transform algorithm are received and visualized in real time. The predicted values ​​of insulation aging status, current carrying capacity assessment value, and fault occurrence probability are received and visualized in real time by the particle swarm optimization least squares support vector machine model. The original multi-source monitoring data set, the multi-scale fusion features, the insulation aging state prediction value, the current carrying capacity assessment value, and the fault occurrence probability prediction value are overlaid on a unified geographic information view and a time axis view to generate a panoramic monitoring screen of cable status.

8. The cable condition multi-source data fusion and prediction system according to claim 7, characterized in that, The cloud data platform also provides fault early warning services, including: Set an early warning threshold for the predicted insulation aging state value, an early warning threshold for the current carrying capacity assessment value, and an early warning threshold for the predicted fault occurrence probability value. The predicted value of the insulation aging state acquired in real time is compared with the corresponding early warning threshold. If the threshold is exceeded, an insulation aging early warning signal is generated. The real-time current carrying capacity assessment value is compared with the corresponding warning threshold. If it is lower than the threshold, an overload risk warning signal is generated. The predicted probability of failure is compared with the real-time acquired value and the corresponding warning threshold. If the value exceeds the threshold, a failure probability warning signal is generated. The insulation aging warning signal, the overload risk warning signal, and the fault probability warning signal are aggregated and combined with cable location information to generate a graded fault warning report.

9. The cable condition multi-source data fusion and prediction system according to claim 1, characterized in that, The cloud data platform also provides micro-application services with predictive and extrapolation capabilities, including: The system receives user-inputted simulation conditions, which include future load change scenarios, ambient temperature change scenarios, or maintenance plan scenarios. Based on the prediction model corresponding to the predicted value of insulation aging state, the current carrying capacity assessment value, and the predicted value of fault occurrence probability, the inference conditions are used as the input boundary of the prediction model for forward simulation calculation. Calculate the insulation aging state evolution trajectory, current carrying capacity change curve, and fault occurrence probability time-varying spectrum of the cable under the above simulation conditions; The evolution trajectory of the insulation aging state, the change curve of the current carrying capacity, and the time-varying spectrum of the fault occurrence probability are visualized and displayed to generate a report on the future development trend of the cable state.

10. The cable condition multi-source data fusion and prediction system according to claim 1, characterized in that, The full lifecycle management service also includes: Based on the estimated remaining life of the cable and the predicted probability of failure, a preventative maintenance schedule is developed. Based on the historical trends in predicted insulation aging conditions, assess the insulation degradation rate of cables at different operational stages. By integrating cable commissioning time, historical load records, historical fault records, and environmental parameters, a personalized full life-cycle profile for the cable is constructed. Based on the individualized full life-cycle profile of the cable and its current predicted status, a comprehensive decision-making recommendation is made regarding cable replacement, repair, or continued operation.

Images