Cable defect detection method and system based on artificial intelligence

Abstract

The invention discloses a cable defect detection method and system based on artificial intelligence. The method comprises the following steps: acquiring cable detection data from a multi-modal data acquisition device; performing standardization processing and time dimension alignment operation on the cable detection data to obtain a normalized multi-modal data set; carrying out feature extraction on the multi-modal data set by adopting a convolutional neural network, and generating a preliminary feature set by alternately executing convolution operation and pooling operation for three times; obtaining a multi-modal feature vector according to the preliminary feature set, and generating dynamic weight distribution of each modal feature through an attention weight calculation module; performing weighted fusion on the multi-modal feature vectors based on dynamic weight distribution to obtain a fused feature set; and judging whether the fusion feature set contains a working condition parameter feature vector or not, and if the fusion feature set contains the working condition parameter feature vector, inputting the fusion feature set into the long-short-term memory network for time sequence modeling. The safety and reliability of the power system are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cable defect detection, and in particular discloses a cable defect detection method and system based on artificial intelligence. Background Art

[0002] Cable defect detection is an important guarantee for the safe operation of the power system, which is directly related to the stability of the power grid and the reliability of power supply.

[0003] With the deepening of urban power grid and rural power grid transformation, the accurate identification and prediction of cable defects has become a key technical problem that the power industry needs to solve urgently. Traditional detection methods mainly rely on manual inspections or single-mode equipment, such as infrared thermal imagers or ultrasonic detectors, but these methods often face the limitations of insufficient accuracy, low efficiency and poor adaptability in complex environments. Manual inspections are time-consuming and easily affected by subjective factors. Single-mode equipment is difficult to capture the multi-dimensional characteristics of cable defects, resulting in high missed detection rates and false detection rates, especially in hidden defects and complex geological environments.

[0004] The core challenge in the field of cable defect detection is how to effectively integrate multi-source heterogeneous data and achieve accurate defect feature modeling.

[0005] In the cable operating environment, defect characteristics involve multi-dimensional information such as geometric shape, material properties and operating parameters. A single data source is difficult to fully characterize the overall picture of the defects, resulting in incomplete feature extraction.

[0006] Although the introduction of multimodal data enriches the information dimension, the heterogeneity between data makes the fusion process complicated, and traditional fusion methods find it difficult to capture the correlation between deep features.

[0007] The lack of feature fusion further limits the dynamic prediction capability of defects, especially when the cable ages or the operating conditions change, the existing model is difficult to accurately simulate the expansion trend of defects. This lack of prediction capability directly affects the real-time performance and reliability of the detection system.

[0008] Therefore, how to efficiently fuse multimodal data to extract comprehensive defect features and accurately predict defect expansion based on dynamic working conditions has become a key issue in the field of cable defect detection. Summary of the invention

[0009] The present invention provides a cable defect detection method and system based on artificial intelligence, aiming to solve at least one defect existing in the above-mentioned prior art.

[0010] One aspect of the present invention relates to a cable defect detection method based on artificial intelligence, comprising the following steps: Obtain cable detection data from a multi-modal data acquisition device, where the multi-modal data acquisition device includes an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit; Perform standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multi-modal data set; Use a convolutional neural network to extract features from the multi-modal data set, and generate a preliminary feature set by alternately performing three convolution operations and pooling operations; Obtain a multi-modal feature vector based on the preliminary feature set, and generate a dynamic weight distribution of each modal feature through an attention weight calculation module; Perform weighted fusion on the multi-modal feature vector based on the dynamic weight distribution to obtain a fused feature set; Determine whether the fused feature set contains a working condition parameter feature vector. If the fused feature set contains a working condition parameter feature vector, input the fused feature set into a long short-term memory network for time series modeling; Output a defect expansion trend prediction sequence through the long short-term memory network. The defect expansion trend prediction sequence includes the feature change amounts in the next three detection cycles; Input the fused feature set and the defect expansion trend prediction sequence into a classification module, and calculate the feature matching degree using a preset defect threshold group; Determine the cable defect level according to the result of the feature matching degree calculation. When the matching degree exceeds the first threshold, it is determined that there is a defect. When the matching degree exceeds the second threshold, it is determined that the defect reaches a critical state.

[0011] Further, the steps of performing standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multi-modal data set include: Obtain the original cable detection data from the multi-modal data acquisition device, and use a data cleaning method to perform anomaly detection on the cable surface temperature distribution data, cable internal defect echo signals, and cable appearance image data in the original cable detection data to obtain a first data set; According to the first data set, use a format conversion method to convert data of different modalities into a unified structure. If the time stamp deviation between the cable surface temperature distribution data and the cable internal defect echo signal exceeds a preset threshold, align the time stamps through interpolation to obtain a second data set; For the second data set, use the z-score normalization method to standardize the cable surface temperature distribution data and the cable internal defect echo signal, and at the same time perform grayscale normalization on the cable appearance image data to obtain a third data set; Slice the third data set through a batch processing method, and use a multi-modal fusion algorithm to splice the sliced cable surface temperature distribution data, cable internal defect echo signals, and cable appearance image data in the time dimension to obtain a normalized multi-modal data set.

[0012] Further, the steps of using a convolutional neural network to extract features from the multimodal dataset and generating a preliminary feature set by alternately performing three convolution operations and pooling operations include: Obtain the cable surface temperature distribution data, cable internal defect echo signals, and cable appearance image data from the multimodal dataset. Use the sequence resampling method to convert the cable surface temperature distribution data and cable internal defect echo signals into two-dimensional matrices that match the resolution of the appearance images, and obtain the first intermediate dataset; For the first intermediate dataset, if the time stamps of the cable surface temperature distribution data or the cable internal defect echo signals deviate from the cable appearance image data by more than a preset threshold, align the time stamps by the linear interpolation method, and use the matrix splicing method to merge the aligned multi-channel matrices to obtain the second intermediate dataset; According to the second intermediate dataset, use a convolutional neural network to perform the first convolution operation, extract features from the multi-channel matrix using a preset convolution kernel, and downsample through the max pooling operation to obtain the first feature set; For the first feature set, alternately perform the second and third convolution and max pooling operations, and use the channel weighting method to fuse the results of multiple convolution and pooling operations to obtain the preliminary feature set.

[0013] Further, the steps of obtaining a multimodal feature vector based on the preliminary feature set and generating a dynamic weight distribution of each modal feature through an attention weight calculation module include: Obtain the multimodal feature vector based on the preliminary feature set, use the matrix decomposition method to split the multimodal feature vector into independent modal matrices, and convert the independent modal matrices into one-dimensional feature vectors through the vector mapping method to obtain the first feature vector set; For the first feature vector set, use the attention weight calculation module to perform weighted processing on each modal feature vector, and calculate the dynamic weight distribution through the softmax function to obtain the weighted feature vector set; If the weight value of any modality in the weighted feature vector set is lower than the preset threshold, adjust the weight distribution by the linear interpolation method, and use the vector splicing method to merge the adjusted weighted feature vectors to obtain the second feature vector set; According to the second feature vector set, use the principal component analysis method to perform dimensionality reduction processing on the high-dimensional feature vectors, and generate an optimized multimodal feature vector through the matrix reorganization method to obtain the final feature vector set.

[0014] Further, the steps of performing weighted fusion on the multimodal feature vector based on the dynamic weight distribution to obtain the fusion feature set include: Obtain the initial feature matrix from the multi-modal feature vectors, split the initial feature matrix into independent modal sub-matrices by using matrix decomposition method, and convert the independent modal sub-matrices into one-dimensional vectors by using vector mapping method to obtain the first vector set; For the first vector set, calculate the dynamic weights of each modal vector by using the attention mechanism, and normalize the dynamic weights through the softmax function to obtain the dynamic weight distribution; If the weight value of any modality in the dynamic weight distribution is lower than the preset threshold, adjust the dynamic weight distribution by using the linear interpolation method, and perform weighted processing on the first vector set by using the vector weighting method to obtain the weighted vector set; According to the weighted vector set, merge the weighted vectors of each modality by using the vector concatenation method, and perform dimensionality reduction processing on the merged vectors by using the principal component analysis method to obtain the fusion feature set.

[0015] Further, determine whether the fusion feature set contains the working condition parameter feature vector. If the fusion feature set contains the working condition parameter feature vector, the steps of inputting the fusion feature set into the long short-term memory network for time series modeling include: Obtain the working condition parameter feature vector containing the working condition parameters from the fusion feature set, split the working condition parameter feature vector into a time dimension sub-vector and a parameter dimension sub-vector by using the matrix decomposition method, and perform alignment processing on the time dimension sub-vector and the parameter dimension sub-vector by using the linear interpolation method to obtain the first feature set; If the standard deviation of the time dimension sub-vector in the first feature set is lower than the preset threshold, smooth the time dimension sub-vector by using the sliding window method, and standardize the smoothed sub-vector by using the vector normalization method to obtain the second feature set; According to the second feature set, perform time series modeling on the time dimension sub-vector by using the long short-term memory network, and perform weighted processing on the parameter dimension sub-vector by using the gated recurrent unit to obtain the third feature set; Perform dimensionality reduction processing on the third feature set by using the principal component analysis method, and merge the dimensionality-reduced time dimension features and parameter dimension features by using the vector concatenation method to obtain the fourth feature set.

[0016] Further, the steps of outputting the defect expansion trend prediction sequence by the long short-term memory network, where the defect expansion trend prediction sequence contains the feature change amounts in the next three detection cycles include: Obtain the original data set containing the defect features and the working condition parameters from the historical detection data, perform alignment processing on the time dimension data by using the linear interpolation method, and standardize the working condition parameter data by using the vector normalization method to obtain the first data set; Determine whether the standard deviation of the time - dimension data of the first data set is lower than a preset threshold. If it is lower than the preset threshold, use the sliding - window method to smooth the time - dimension data, and use the principal - component analysis method to reduce the dimension of the smoothed data to obtain a second data set; According to the second data set, use a gated recurrent unit to weight the working - condition parameter data, and merge the weighted working - condition parameter data with the time - dimension data through the feature - splicing method to obtain a third data set; Perform time - series modeling on the third data set through a long short - term memory network, and output a defect - expansion trend prediction sequence containing the next three detection cycles to obtain a feature - change amount sequence.

[0017] Furthermore, the steps of inputting the fusion feature set and the defect - expansion trend prediction sequence into the classification module and calculating the feature - matching degree using a preset defect - threshold group include: Obtain a fifth feature set containing multi - dimensional data from the fusion feature set and the defect - expansion trend prediction sequence, use the principal - component analysis method to reduce the dimension of the fifth feature set, and standardize the dimension - reduced data through the vector - normalization method to obtain a sixth feature set; According to the sixth feature set, compare each feature dimension in the sixth feature set with the preset threshold group. If the feature value of a feature dimension exceeds the upper limit of the corresponding dimension in the threshold group, mark it as an abnormal feature to obtain an abnormal - feature set; Perform clustering processing on the abnormal - feature set through the classification module, and group the abnormal - feature set using the K - means algorithm to obtain multiple feature subsets; For the feature subsets, calculate the matching degree between the feature subsets and the defect - threshold group, and comprehensively evaluate the matching degree through the weighted - average method to obtain a classification result.

[0018] Furthermore, the steps of determining the cable - defect level according to the feature - matching - degree calculation result, where it is determined that there is a defect when the matching degree exceeds the first threshold and it is determined that the defect reaches the critical state when the matching degree exceeds the second threshold include: Obtain a multi - dimensional information feature vector containing multi - dimensional information from the cable - state data, use principal - component analysis to reduce the dimension of the multi - dimensional information feature vector, and process the dimension - reduced data through the vector - normalization method to obtain a standardized feature set; According to the standardized feature set, compare each feature dimension with the preset first threshold and second threshold. If the feature value of a feature dimension exceeds the first threshold, mark it as an abnormal feature; if the feature value of a feature dimension exceeds the second threshold, mark it as a critical feature to obtain an abnormal - feature set and a critical - feature set; For the abnormal - feature set and the critical - feature set, use the K - means algorithm to cluster and group the features to obtain multiple feature subsets; The matching degree value between the feature subset and the defect level is calculated by a weighted evaluation method to determine the defect level and the critical state, and a determination result is obtained.

[0019] Another aspect of the present invention relates to an artificial intelligence-based cable defect detection system for implementing the above-mentioned artificial intelligence-based cable defect detection method. The artificial intelligence-based cable defect detection system includes: A first acquisition module for acquiring cable detection data from a multimodal data acquisition device, the multimodal data acquisition device including an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit; A second acquisition module for performing normalization processing and time dimension alignment operations on the cable detection data to obtain a normalized multimodal data set; A first generation module for extracting features from the multimodal data set by using a convolutional neural network, and generating a preliminary feature set by alternately performing three convolutional operations and pooling operations; A second generation module for obtaining a multimodal feature vector according to the preliminary feature set, and generating a dynamic weight distribution of each modal feature through an attention weight calculation module; A third acquisition module for performing weighted fusion on the multimodal feature vector based on the dynamic weight distribution to obtain a fusion feature set; A judgment module for judging whether the fusion feature set contains a working condition parameter feature vector. If the fusion feature set contains a working condition parameter feature vector, the fusion feature set is input into a long short-term memory network for time series modeling; An output module for outputting a defect expansion trend prediction sequence through the long short-term memory network, the defect expansion trend prediction sequence including the feature change amounts in the next three detection cycles; A calculation module for inputting the fusion feature set and the defect expansion trend prediction sequence into a classification module, and calculating the feature matching degree by using a preset defect threshold group; A determination module for determining the cable defect level according to the feature matching degree calculation result. When the matching degree exceeds the first threshold, it is determined that there is a defect. When the matching degree exceeds the second threshold, it is determined that the defect reaches the critical state.

[0020] The beneficial effects achieved by the present invention are: The present invention provides an artificial intelligence-based cable defect detection method and system. Cable detection data is obtained through a multi-modal data acquisition device, the data is subjected to standardization processing and time alignment, a convolutional neural network is used to extract features, and an attention mechanism is used to dynamically weight and fuse multi-modal features. If operating condition parameters are included, a long short-term memory network is used for time series modeling to predict the defect expansion trend in the next three detection cycles. Finally, the fused features and the prediction sequence are input into a classification module, and the defect level is determined based on a preset threshold. The present invention realizes the accurate detection of cable defects and the prediction of future development trends, effectively improves the safety and reliability of the power system, provides a scientific basis for cable maintenance decision-making, and has important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic flowchart of an embodiment of an artificial intelligence-based cable defect detection method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0023] As Figure 1 shown, a first embodiment of the present invention proposes an artificial intelligence-based cable defect detection method, including the following steps: Step S100: Obtain cable detection data from a multi-modal data acquisition device, which includes an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit.

[0024] A multi-modal data acquisition device refers to a system device that synchronously obtains different types of data (such as visual, auditory, physiological signals, etc.) by integrating multiple sensors or devices, aiming to achieve comprehensive capture and collaborative analysis of cross-modal information.

[0025] Cable detection data is a structured or unstructured information set collected through multi-modal technical means and reflecting the physical state and performance parameters of the cable. The core of cable detection data lies in quantitatively evaluating the health status and potential risks of the cable.

[0026] Step S200: Perform standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multi-modal data set.

[0027] Standardization is a data preprocessing method that transforms data through mathematical transformations into a unified dimension and distribution range. The core goal of standardization is to eliminate the bias caused by differences in dimension or numerical range of different features, and improve the comparability and computational efficiency of data in model training or analysis. It is usually achieved by adjusting the data distribution so that its mean is 0 and its standard deviation is 1.

[0028] Temporal Alignment refers to eliminating the deviation of multimodal data on the time axis through technical means, ensuring that the time series data obtained by different sensors or acquisition devices can accurately match the occurrence time of the same physical event. The core goal of temporal alignment is to solve the problem of time series misalignment caused by differences in sampling frequency, inconsistent clock references, or transmission delays, providing a basis for subsequent multimodal fusion and analysis with temporal consistency.

[0029] Normalization is a data preprocessing technique that maps data to a specific interval (such as [0, 1] or [-1, 1]) through mathematical transformations to eliminate the dimensional differences between features and improve the stability and convergence efficiency of model training. The core of normalization is to achieve comparability between different features by adjusting the numerical range of data, avoiding model bias caused by dimensional or scale differences.

[0030] A Multimodal Dataset refers to a set that contains multiple types (modalities) of data, such as text, images, audio, video, sensor signals (such as temperature, accelerometer), etc. The core value of a multimodal dataset lies in enhancing the model's understanding ability of complex scenarios by fusing information from different modalities and making up for the limitations of single-modal data. The construction and processing of multimodal datasets need to solve key problems such as modality alignment, heterogeneous data fusion, and semantic consistency.

[0031] Step S300: Use a convolutional neural network to extract features from the multimodal dataset, and generate a preliminary feature set by alternately performing three convolution operations and pooling operations.

[0032] A Convolutional Neural Network (CNN) is a deep learning model specifically designed to process grid-like data (such as images, audio, video). The core of a convolutional neural network automatically extracts local features and constructs hierarchical representations through convolution operations, pooling, and non-linear activation functions (such as ReLU), significantly reducing the parameter scale while retaining the spatial or temporal correlation of the data.

[0033] The convolution operation is the core operation of a convolutional neural network (CNN). By sliding a filter (convolution kernel) to calculate the weighted sum of local regions at each position in the input data, it extracts local features in space or time series and preserves the internal structure of the data (such as edges and textures of images). Its core idea is to significantly reduce the number of model parameters through local connection and parameter sharing, while capturing translation-invariant features.

[0034] The pooling operation is an important part of a convolutional neural network. By downsampling local regions of the input feature map, it compresses the data dimension and preserves key features, thereby reducing the computational amount, enhancing the model's robustness to small displacements, and alleviating overfitting. Its core idea is to aggregate local information and gradually abstract high-level semantic features.

[0035] The initial feature set refers to the basic feature set extracted or constructed from raw data in machine learning or deep learning tasks, which is used for the initial training and analysis of the model. It usually contains unoptimized raw features or features generated by simple rules. Its core goal is to quickly construct the key representation of the data and provide the basic input for subsequent feature screening, optimization, or high-order feature engineering.

[0036] Step S400: Obtain the multimodal feature vector based on the initial feature set, and generate the dynamic weight distribution of each modal feature through the attention weight calculation module.

[0037] The multimodal feature vector is a unified vector representation generated by fusing feature representations from different modalities (such as images, text, audio, sensor data, etc.). Its purpose is to capture cross-modal complementary information and association patterns, and support the joint analysis and reasoning of complex tasks. The core goal of the multimodal feature vector is to break the modal barrier and utilize the synergy of multi-source data to enhance the model's understanding ability of complex scenarios.

[0038] The attention weight calculation module is the core component of the attention mechanism, which is used to dynamically quantify the correlation or dependence strength between different elements in the input sequence. By generating a weight distribution, it guides the model to focus on key information. Its essence is to calculate the similarity through the interaction between the query and the key, and normalize it into a probability distribution (Softmax). Finally, it weighted aggregates the value vector to form a context-aware feature expression.

[0039] Dynamic weight distribution is a mechanism in deep learning models that adaptively adjusts the importance of features. By calculating the correlation weights of different elements in the input data in real time, it guides the model to dynamically focus on key information in a specific context or task. The core of dynamic weight distribution lies in that the weight assignment is not fixed, but flexibly adjusted according to the input content, task requirements, or temporal changes to enhance the model's expressive power and robustness.

[0040] Step S500: Perform weighted fusion on the multi-modal feature vectors based on the dynamic weight distribution to obtain a fused feature set.

[0041] Weighted fusion is a core fusion method in multi-modal data processing. By assigning dynamic weights to the features or decision results of different modalities, it realizes a linear combination to generate a comprehensive representation. Its core lies in adjusting the contribution degree of each modality through weights to adapt to the information complementarity and reliability differences between modalities in different scenarios, thereby improving the robustness and accuracy of the model.

[0042] The fused feature set is a joint representation generated by integrating the feature representations of different modalities in multi-modal data processing, aiming to eliminate the heterogeneity between modalities and capture complementary information, thereby improving the model's semantic understanding and reasoning ability. The core of the fused feature set is to transform the feature vectors from modalities such as vision, language, and sensors into a unified high-dimensional space expression through feature-level fusion strategies (such as concatenation, weighted combination, or non-linear mapping) to support joint modeling and decision-making for downstream tasks.

[0043] Step S600: Determine whether the fused feature set contains the working condition parameter feature vector. If the fused feature set contains the working condition parameter feature vector, input the fused feature set into the long short-term memory network for time series modeling.

[0044] The working condition parameter feature vector is a numerical representation that describes the key parameters of a device or system in a specific working state. By encoding multi-dimensional monitoring data (such as temperature, pressure, vibration frequency, etc.) into a structured vector, it realizes a machine-readable expression of the working condition state. Its core function is to provide a unified high-dimensional input space for machine learning models to support tasks such as device state classification, fault prediction, and performance optimization.

[0045] The long short-term memory network (LSTM) is a special type of recurrent neural network (RNN). By introducing a gating mechanism and a cell state, it solves the problem of gradient vanishing / explosion in traditional RNNs when processing long sequence data and realizes efficient modeling of long-term dependencies. The core design of the long short-term memory network is to dynamically regulate the information flow through the forget gate, input gate, and output gate, balancing the relevance between historical memory and current input, and is widely used in fields such as time series prediction and natural language processing.

[0046] Time series modeling is a modeling process that targets data sequences observed in chronological order (such as stock prices, meteorological data, sensor signals, etc.), constructs models through mathematical or statistical methods to reveal their internal laws (trends, periodicity, randomness), and realizes prediction, classification, or anomaly detection. Its core is to capture the time dependency of data, that is, the correlation between the current observed value and historical values, so as to provide a quantitative analysis framework for dynamic systems.

[0047] Step S700: Output a defect expansion trend prediction sequence through a long short-term memory network. The defect expansion trend prediction sequence includes the feature change amounts in the next three detection cycles.

[0048] The defect expansion trend prediction sequence is a quantitative modeling of the development process of defects (such as cracks, wear, corrosion, etc.) in equipment, materials, or systems based on time series analysis methods. A time series model is constructed through historical defect data (such as size, location, frequency, etc.) to predict the future defect expansion rate, critical state, and failure risk. Its core goal is to provide a dynamic decision-making basis for maintenance strategy optimization and life assessment by capturing the time dependency of defect evolution (such as the accelerated expansion stage, periodic fluctuations, etc.).

[0049] Step S800: Input the fused feature set and the defect expansion trend prediction sequence into the classification module, and calculate the feature matching degree using a preset defect threshold group.

[0050] The classification module is a functional component in a machine learning or deep learning system responsible for mapping input data to preset class labels. Its core role is to achieve group identification of data (such as text classification, image recognition, fault diagnosis, etc.) based on feature learning or rule judgment, and complete class determination through the model output probability distribution or decision boundary. The classification module usually consists of a feature extraction layer, a classifier, and decision logic, and needs to be designed according to the task requirements (such as binary classification, multi-class classification) and data characteristics (such as structured, time series).

[0051] The defect threshold group is a set of predefined key parameters or conditions in a defect management or predictive maintenance system, used to dynamically evaluate the severity, expansion risk of defects, or trigger intervention actions (such as early warning, repair). Its core role is to divide the defect state levels through quantitative criteria (such as defect quantity threshold, growth rate threshold, time persistence threshold), providing a basis for classification response for decision-makers.

[0052] Feature matching degree calculation is a key technology in image processing and pattern recognition, which is used to quantify the similarity or difference between two features (such as key points, regions or templates). The core objective of feature matching degree calculation is to evaluate the consistency of feature pairs through mathematical modeling or algorithms (such as distance metrics, similarity functions), so as to provide a basis for matching decisions (such as correct matching determination or false matching elimination).

[0053] Step S900: Determine the cable defect level according to the feature matching degree calculation result. When the matching degree exceeds the first threshold, it is determined that there is a defect. When the matching degree exceeds the second threshold, it is determined that the defect reaches the critical state.

[0054] The cable defect level is a classification management mechanism based on the threat degree of defects to the safe operation of the power system and the urgency of treatment. Its purpose is to guide the maintenance personnel to give priority to dealing with high-risk defects through grading standards, and ensure the stability and reliability of the cable line.

[0055] The critical state refers to the key stage where the cable defect is in the transformation between safe operation and potential risk. Its characteristics are that although the defect does not reach the judgment standard of a higher level (such as a critical defect), there is a possibility of rapid deterioration or triggering a chain failure in the short term, and it needs to be intervened through dynamic monitoring and predictive maintenance.

[0056] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S100 includes: Step S110: Use an infrared thermal imaging unit to collect the cable surface temperature distribution data, a ultrasonic detection unit to collect the cable internal defect echo signal, and an optical sensing unit to collect the cable appearance image data.

[0057] The core of the infrared thermal imaging unit collecting the cable surface temperature distribution data lies in detecting the heat distribution on the cable surface through infrared radiation to identify potential overheating points or abnormal regions. For example, due to poor contact, the cable joint may cause local temperature rise, and the infrared thermal imaging can capture this abnormality.

[0058] In a possible implementation, a high-resolution infrared thermal imager is used, the temperature measurement range is set from -20°C to 150°C, the scanning frequency is 30 frames per second, and the cable surface temperature distribution is collected.

[0059] The ultrasonic detection unit collects the cable internal defect echo signal. Based on the principle that ultrasonic waves will generate reflections when encountering defects during propagation in the medium, it detects defects such as cracks or bubbles inside the cable.

[0060] Exemplarily, a ultrasonic probe with a frequency of 5MHz is used to scan along the cable axis, emit pulses and receive the cable internal defect echo signal.

[0061] The amplitude and time difference of the echo signals of internal defects in the cable can reflect the size and location of the defects. For example, when a 2-mm air bubble is detected inside a certain section of the cable, the echo signal of the internal defect in the cable shows a high-amplitude spike, and the time delay corresponds to a defect depth of approximately 5 mm.

[0062] The optical sensing unit collects the cable appearance image data for detecting surface physical damages, such as abrasions, cracks, or foreign object attachments. In one embodiment, a high-resolution CCD camera is used, together with an annular light source, to capture the cable surface image with a resolution of 1920×1080 pixels.

[0063] Preferably, image acquisition should be carried out under uniform illumination to avoid the interference of specular reflection.

[0064] Step S120: Through signal processing, denoise and filter the cable surface temperature distribution data, the echo signals of internal cable defects, and the cable appearance image data to obtain the processed cable surface temperature distribution data, the echo signals of internal cable defects, and the cable appearance image data.

[0065] It should be noted that the ambient temperature and wind speed may interfere with the data accuracy. Therefore, the data needs to be collected in a relatively stable environment and adjusted by combining an environmental correction algorithm.

[0066] The processed temperature data can clearly show the temperature gradient on the cable surface. The temperature at the abnormal point may reach 80°C, while the normal area is 40°C, which helps to quickly locate potential faults and improve the inspection efficiency.

[0067] Specifically, signal processing needs to denoise the echo, and wavelet transform is used to filter out the background noise to ensure that the defect signal is clear. This method can accurately locate internal defects and ensure the long-term reliability of cable operation.

[0068] In signal processing, median filtering is used to remove image noise, and then the crack contour is highlighted through an edge detection algorithm. For example, when a 1-cm-long crack is detected on the cable surface, the image shows an obvious black line with a width of approximately 0.5 mm.

[0069] It can be understood that the processed image data can visually present the degree of surface damage. Combining with deep learning algorithms can also automatically classify the damage types and improve the intelligent level of detection.

[0070] In one embodiment, the comprehensive analysis of the above three types of data can achieve a comprehensive evaluation of the cable status. For example, the infrared thermal imaging at a certain cable joint shows an abnormal temperature of 75°C, ultrasonic detection finds a small air bubble defect inside, and the optical image shows slight surface abrasion.

[0071] Comprehensive analysis shows that the joint may heat up due to internal defects and surface damage, and priority maintenance is required. This multi-dimensional detection method significantly improves the diagnostic accuracy, reduces the risk of missed detection, and provides a reliable basis for cable maintenance.

[0072] Furthermore, the cable defect detection method based on artificial intelligence provided in this embodiment, step S200 includes: Step S210: Obtain the original cable detection data from the multi-modal data acquisition device, and use the data cleaning method to perform anomaly detection on the cable surface temperature distribution data, cable internal defect echo signals, and cable appearance image data in the original cable detection data to obtain the first data set.

[0073] The cable detection data obtained by the multi-modal data acquisition device includes cable surface temperature distribution data, cable internal defect echo signals, and cable appearance image data. Data cleaning is a key step to ensure data quality.

[0074] Exemplarily, the cleaning method can detect outliers for the cable surface temperature distribution data. For example, if the temperature of a certain section of the cable suddenly changes to 100°C, far exceeding the normal range of 40°C to 50°C, it may be caused by sensor failure or external interference.

[0075] Use the median filtering method to remove such outliers, and at the same time perform denoising processing on the random spikes caused by environmental noise in the cable internal defect echo signals, such as filtering out the noise with an amplitude exceeding 3 times the standard deviation through a sliding window.

[0076] For the cable appearance image data, it is necessary to check for image blurring or uneven illumination conditions, such as removing the dark spot images caused by light source occlusion.

[0077] The cleaned first data set retains reliable data, laying a foundation for subsequent processing.

[0078] Step S220: According to the first data set, use the format conversion method to convert the data of different modalities into a unified structure. If the timestamp deviation between the cable surface temperature distribution data and the cable internal defect echo signals exceeds the preset threshold, align the timestamps through the interpolation method to obtain the second data set.

[0079] In a possible implementation, the format conversion unifies the data of different modalities into a structured format, such as JSON format, which includes timestamps and data values.

[0080] It should be noted that the timestamp deviation may be caused by device synchronization errors. For example, the timestamp of the cable surface temperature distribution data is 10:00:00, while the timestamp of the cable internal defect echo signal is 10:00:02, and the deviation exceeds the threshold of 1 second.

[0081] Using the linear interpolation method, estimate the aligned value based on the data at adjacent time points, such as interpolating the echo signal value at 10:00:00. This method ensures the time alignment of the second data set, facilitating multimodal fusion.

[0082] Step S230: For the second data set, use the z-score normalization method to standardize the cable surface temperature distribution data and the cable internal defect echo signal, and at the same time perform grayscale normalization on the cable appearance image data to obtain the third data set.

[0083] Specifically, the z-score normalization processes the temperature data and the echo signal, converting the data into a standard distribution with a mean of 0 and a standard deviation of 1. For example, the temperature data ranges from 40°C to 80°C, and after normalization, it is mapped to -2 to 2, facilitating cross-modal comparison.

[0084] The cable appearance image data is grayscale-normalized to scale the pixel values to 0 to 1. For example, the original pixel value of 200 is adjusted to 0.78. This normalized third data set eliminates the dimension difference and improves the compatibility of the fusion algorithm.

[0085] Step S240: Fragment the third data set through the batch processing method, and use the multimodal fusion algorithm to splice the fragmented cable surface temperature distribution data, cable internal defect echo signal, and cable appearance image data in the time dimension to obtain the normalized multimodal data set.

[0086] Preferably, the batch processing method fragments the third data set by time period. For example, every 5 minutes of data is a batch, generating a fragmented data set with 1000 records.

[0087] The multimodal fusion algorithm splices the fragmented data in the time dimension. For example, at a certain time point, the temperature data indicates 50°C, the cable internal defect echo signal shows a 2-mm defect, and the image data captures a 0.3-mm crack. After fusion, a complete feature vector is formed. This normalized multimodal data set provides a unified perspective for subsequent analysis.

[0088] It can be understood that each step of the above method is closely linked. Cleaning ensures data quality, format conversion and time alignment guarantee data consistency, normalization eliminates dimension differences, and fragmentation and fusion optimize data organization. This progressive processing method provides high-quality input for cable condition assessment and significantly improves the reliability of subsequent analysis.

[0089] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S300 includes: Step S310: Obtain the cable surface temperature distribution data, the cable internal defect echo signal, and the cable appearance image data from the multimodal dataset. Use the sequence resampling method to convert the cable surface temperature distribution data and the cable internal defect echo signal into two-dimensional matrices that match the resolution of the appearance image, obtaining the first intermediate dataset.

[0090] Exemplarily, when extracting the cable surface temperature distribution data, the cable internal defect echo signal, and the cable appearance image data from the multimodal dataset, it is necessary to ensure that the data format is adapted to subsequent processing.

[0091] The cable surface temperature distribution data is usually recorded in the form of a time series. For example, the temperature value is collected once per second, forming a one-dimensional array. For example, the temperature of a certain cable section is recorded as [45°C, 46°C, 47°C...] within 10 seconds.

[0092] The cable internal defect echo signal is the waveform data generated by ultrasonic detection, which contains the defect position and intensity. For example, the echo amplitude at a defect depth of 2 mm is 0.8 V.

[0093] The cable appearance image data is a high-resolution grayscale image, such as with a resolution of 512×512 pixels, recording the cable surface cracks.

[0094] The sequence resampling method converts the one-dimensional temperature and echo signals into two-dimensional matrices that match the image resolution. For example, resample the 10-second temperature data into a 512×512 matrix, repeat the temperature values in each row and fill them by interpolation along the time axis to ensure alignment with the image pixels.

[0095] Perform a similar process on the echo signal, map the amplitude values to matrix elements, forming the first intermediate dataset.

[0096] Step S320: For the first intermediate dataset, if the timestamp of the cable surface temperature distribution data or the cable internal defect echo signal deviates from the timestamp of the cable appearance image data by more than a preset threshold, align the timestamps by the linear interpolation method, and use the matrix splicing method to merge the aligned multi-channel matrices, obtaining the second intermediate dataset.

[0097] In a possible implementation, timestamp alignment is crucial. Suppose the timestamp of the temperature data is 10:00:00, the echo signal is 10:00:01, and the image data is 10:00:02, with a deviation exceeding the threshold of 0.5 seconds.

[0098] The linear interpolation method can fill in the missing time point data. For example, estimate the echo signal value at 10:00:00, and interpolate it to 0.85 V based on the previous and subsequent amplitudes of 0.8 V and 0.9 V.

[0099] The matrix splicing method combines the aligned temperature matrix, echo matrix, and image matrix by channel to form a three-channel second intermediate data set, similar to the multi-channel structure of an RGB image, which is convenient for unified processing.

[0100] Step S330: According to the second intermediate data set, perform the first convolution operation using a convolutional neural network, extract features for the multi-channel matrix using a preset convolution kernel, and downsample through a max pooling operation to obtain the first feature set.

[0101] Specifically, the first convolution operation of the convolutional neural network extracts features from the second intermediate data set.

[0102] The preset 3×3 convolution kernel scans the temperature, echo, and image channels respectively to capture local patterns, such as temperature gradients or crack edges.

[0103] The max pooling operation downsamples with a 2×2 window, retains significant features. For example, it reduces a 512×512 matrix to 256×256, reduces the computational amount, and generates the first feature set.

[0104] It should be noted that pooling retains high-intensity features, such as pixel values of 0.9 in the crack area, while ignoring low-value noise.

[0105] Step S340: For the first feature set, alternately perform the second and third convolution and max pooling operations, and use the channel weighting method to fuse the results of multiple convolution and pooling operations to obtain a preliminary feature set.

[0106] Preferably, the second and third convolution and pooling operations are alternately performed to further refine the features. For example, the second convolution uses a 5×5 convolution kernel to focus on larger-scale defect patterns, and the matrix is reduced to 128×128 after pooling. The third operation is similar, generating a 64×64 matrix.

[0107] The channel weighting method fuses the results of multiple convolutions. For example, it assigns a weight of 0.4 to the temperature channel, 0.3 to the echo, and 0.3 to the image, and comprehensively generates a preliminary feature set. This hierarchical extraction and fusion method ensures comprehensive features and is suitable for subsequent cable condition analysis.

[0108] It can be understood that the above method gradually converts heterogeneous data into a unified feature representation through resampling, alignment, convolution, and fusion.

[0109] Each step is closely linked. Sequential resampling ensures consistent dimensions, time alignment eliminates biases, convolution and pooling extract key patterns, and channel weighting optimizes feature expression, providing high-quality input for cable defect detection.

[0110] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S400 includes: Step S410: Obtain the multi-modal feature vector according to the preliminary feature set. Use the matrix decomposition method to split the multi-modal feature vector into independent modal matrices, and convert the independent modal matrices into one-dimensional feature vectors through the vector mapping method to obtain the first feature vector set.

[0111] Exemplarily, in the cable defect detection scenario, the preliminary feature set usually contains the high-dimensional representations of multi-modal data such as temperature, echo signal, and image.

[0112] The process of obtaining the multi-modal feature vector aims to integrate these features into a unified vector representation.

[0113] In a possible implementation, the preliminary feature set is a three-channel matrix of 64×64, containing temperature gradient, echo amplitude, and image texture information.

[0114] Through the flattening operation, each channel matrix is converted into a one-dimensional vector. For example, the temperature channel generates a 4096-dimensional vector to form the initial multi-modal feature vector. This unified representation facilitates subsequent decomposition and processing.

[0115] The matrix decomposition method splits the multi-modal feature vector into independent modal matrices to separate the unique information of each modality.

[0116] Specifically, non-negative matrix factorization can decompose the 4096-dimensional vector into three independent matrices, corresponding to the temperature, echo, and image modalities respectively. For example, the temperature matrix retains the periodic changes of the time series, the echo matrix highlights the signal intensity at the defect location, and the image matrix focuses on the geometric features of the crack.

[0117] After decomposition, the scale of each modal matrix is about 1365 dimensions, maintaining information independence and providing a clear input for subsequent mapping.

[0118] The vector mapping method converts the independent modal matrices into one-dimensional feature vectors to form the first feature vector set.

[0119] In one embodiment, the fully connected layer mapping compresses the 1365-dimensional temperature matrix into a 512-dimensional vector, and the echo and image modalities are processed similarly, finally generating three 512-dimensional vectors. This compression retains key patterns, such as temperature anomaly points or crack edge features, for unified analysis.

[0120] Step S420: For the first feature vector set, use the attention weight calculation module to perform weighted processing on each modal feature vector, and calculate the dynamic weight distribution through the softmax function to obtain the weighted feature vector set.

[0121] The attention weight calculation module performs weighted processing on the first feature vector set to highlight the important modalities.

[0122] Preferably, the attention mechanism calculates dynamic weights through the softmax function. For example, the weight of the temperature vector of a certain cable segment is 0.5, the echo is 0.3, and the image is 0.2, reflecting a greater contribution of temperature anomalies to defect detection.

[0123] Step S430: If the weight value of any modality in the weighted feature vector set is lower than the preset threshold, adjust the weight distribution by the linear interpolation method, and merge the adjusted weighted feature vectors by the vector concatenation method to obtain the second feature vector set.

[0124] If the weight of the image modality is lower than the threshold of 0.25, adjust the weight by the linear interpolation method. For example, increase the image weight to 0.25, and correspondingly reduce the temperature and echo weights to 0.45 and 0.3 to ensure a balanced weight distribution. This dynamic adjustment enhances the robustness of feature expression.

[0125] The vector concatenation method merges the adjusted weighted feature vectors into the second feature vector set. For example, three 512-dimensional vectors are concatenated into a 1536-dimensional vector to integrate multimodal information. This concatenation retains the weighted contributions of each modality and facilitates subsequent dimensionality reduction processing.

[0126] Step S440: According to the second feature vector set, perform dimensionality reduction processing on the high-dimensional feature vectors by the principal component analysis method, and generate an optimized multimodal feature vector through the matrix recombination method to obtain the final feature vector set.

[0127] The principal component analysis method reduces the dimension of the second feature vector set to reduce redundancy.

[0128] It should be noted that the 1536-dimensional vector retains 90% of the variance through the principal component analysis and is reduced to 256 dimensions to form an optimized multimodal feature vector.

[0129] It can be understood that the key information such as temperature anomalies, echo intensity, and crack features is still retained in the vector after dimensionality reduction.

[0130] The matrix recombination method further adjusts the vector structure. For example, the 256-dimensional vector is recombined into three sub-vectors according to the modality ratio to ensure that the final feature vector set is applicable to the cable state classification or defect location task. This hierarchical processing method generates an efficient feature representation through decomposition, mapping, weighting, and dimensionality reduction, providing high-quality input for subsequent analysis.

[0131] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S500 includes: Step S510: Obtain an initial feature matrix from the multimodal feature vectors, split the initial feature matrix into independent modality sub-matrices by the matrix decomposition method, and convert the independent modality sub-matrices into one-dimensional vectors through the vector mapping method to obtain the first vector set.

[0132] Exemplarily, in the cable defect detection scenario, the acquisition of the initial feature matrix starts from multi-modal feature vectors.

[0133] Suppose the multi-modal feature vectors integrate the cable surface temperature distribution data, the internal defect echo signals of the cable, and the cable appearance image data to form a high-dimensional representation.

[0134] Specifically, the cable surface temperature distribution data reflects the surface heat distribution of the cable, the internal defect echo signals of the cable capture the echo characteristics of internal defects, and the cable appearance image data reveals the thermal anomaly areas.

[0135] In one embodiment, the initial feature matrix is a three-dimensional matrix of 64×64×3, and each channel corresponds to one modality. The matrix decomposition method splits this matrix into three independent modality sub-matrices to separate the unique information of each modality.

[0136] Preferably, the singular value decomposition technique can separate the temperature sub-matrix into a two-dimensional matrix of 64×64, retaining the thermal gradient features; the ultrasonic sub-matrix highlights the echo peaks; the infrared image sub-matrix focuses on the hot spot areas. After decomposition, each sub-matrix independently expresses the single-modality information, facilitating subsequent processing.

[0137] It should be noted that the vector mapping method converts the independent modality sub-matrices into one-dimensional vectors to form the first vector set.

[0138] In one possible implementation, the temperature sub-matrix generates a 4096-dimensional vector through a flattening operation, and then compresses it into a 512-dimensional vector through a fully connected layer, retaining the key thermal anomaly points.

[0139] The ultrasonic and infrared modalities are processed similarly, and finally three 512-dimensional vectors are generated to form the first vector set. This mapping method simplifies the processing complexity of high-dimensional data.

[0140] Step S520: For the first vector set, use the attention mechanism to calculate the dynamic weights of each modality vector, and normalize the dynamic weights through the softmax function to obtain the dynamic weight distribution.

[0141] The attention mechanism calculates the dynamic weights for the first vector set to highlight the key modalities. For example, the softmax function assigns weights according to the modality contributions. The temperature vector may obtain a weight of 0.6, the ultrasonic 0.25, and the infrared 0.15.

[0142] Step S530: If the weight value of any modality in the dynamic weight distribution is lower than the preset threshold, then adjust the dynamic weight distribution through the linear interpolation method, and perform weighted processing on the first vector set using the vector weighting method to obtain the weighted vector set.

[0143] If the infrared weight is lower than the threshold of 0.2, the linear interpolation method is used to adjust the weight to 0.2, and the temperature and ultrasonic weights are correspondingly reduced to 0.55 and 0.25. This dynamic adjustment ensures the balanced contribution of each modality.

[0144] It can be understood that the vector weighting method weights the first vector set according to the dynamic weights to generate a weighted vector set.

[0145] Specifically, the 512-dimensional temperature vector is multiplied by the weight of 0.55, and the ultrasonic and infrared vectors are weighted similarly to generate a weighted vector set, highlighting the important features.

[0146] Step S540: According to the weighted vector set, the weighted vectors of each modality are combined by using the vector splicing method, and the combined vectors are dimensionally reduced by using the principal component analysis method to obtain a fused feature set.

[0147] The vector splicing method combines the three weighted vectors into a 1536-dimensional vector, integrating multi-modal information.

[0148] The principal component analysis method reduces the dimension of the combined vectors, retains 95% of the variance, and reduces it to 256 dimensions to form a fused feature set. For example, the vectors after dimension reduction still retain features such as temperature anomalies, echo intensity, and hot spots, which are suitable for cable defect classification tasks. This hierarchical processing method generates an efficient feature representation through decomposition, mapping, weighting, and dimension reduction, providing high-quality input for subsequent analysis.

[0149] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S600 includes: Step S610: Obtain a working condition parameter feature vector containing working condition parameters from the fused feature set, use the matrix decomposition method to split the working condition parameter feature vector into a time dimension sub-vector and a parameter dimension sub-vector, and perform alignment processing on the time dimension sub-vector and the parameter dimension sub-vector through the linear interpolation method to obtain a first feature set.

[0150] Exemplarily, in the cable condition monitoring scenario, obtaining a working condition parameter feature vector containing working condition parameters from the fused feature set aims to capture the dynamic characteristics during the operation of the cable.

[0151] The fused feature set integrates multi-modal data, such as temperature, current load, and vibration signals, to form a high-dimensional feature representation.

[0152] Assume that the fused feature set is a 256-dimensional vector, containing time series information and working condition parameters, such as temperature change over time, current fluctuations, and vibration frequency.

[0153] The matrix decomposition method splits this vector into a time dimension sub-vector and a parameter dimension sub-vector.

[0154] Preferably, using non - negative matrix factorization technology, the 256 - dimensional vector is split into a 128 - dimensional time - dimension sub - vector, which reflects the temporal changes of temperature and vibration, and a 128 - dimensional parameter - dimension sub - vector, which highlights the static operating condition characteristics such as current load. This decomposition method facilitates the separate processing of dynamic and static information.

[0155] In a possible implementation, the linear interpolation method is used to align the time - dimension sub - vector and the parameter - dimension sub - vector.

[0156] Since the time - dimension sub - vectors may have different sampling frequencies, such as temperature data sampled per second and current data sampled every 5 seconds, linear interpolation interpolates the low - frequency data to a unified time step to form the first aligned feature set. For example, after interpolation, the time - dimension sub - vector contains the temperature and vibration values per second, and the parameter - dimension sub - vector synchronously matches the corresponding current values to ensure time consistency.

[0157] Step S620: If the standard deviation of the time - dimension sub - vector in the first feature set is lower than the preset threshold, the sliding window method is used to smooth the time - dimension sub - vector, and the smoothed sub - vector is standardized by the vector normalization method to obtain the second feature set.

[0158] It should be noted that if the standard deviation of the time - dimension sub - vector in the first feature set is lower than the threshold, such as 0.1, it indicates that the data fluctuation is small, which may be caused by noise or insufficient sensor accuracy.

[0159] The sliding window method is used to smooth the time - dimension sub - vector. Specifically, a mean filter with a window size of 5 is used to eliminate instantaneous noise and make the temperature change curve smoother.

[0160] The vector normalization method is used to standardize the smoothed sub - vector. For example, the temperature values are normalized to the interval from 0 to 1 to form the second feature set. This processing enhances the stability of the data and facilitates subsequent modeling.

[0161] Step S630: According to the second feature set, a long short - term memory network is used to perform time - series modeling on the time - dimension sub - vector, and a gated recurrent unit is used to perform weighted processing on the parameter - dimension sub - vector to obtain the third feature set.

[0162] Specifically, the long short - term memory network performs time - series modeling on the time - dimension sub - vector in the second feature set. For example, the long short - term memory network contains 128 hidden units to capture the long - term dependencies of temperature and vibration, such as the change trend of vibration frequency after the temperature rises.

[0163] The gated recurrent unit weights the parameter dimension sub-vectors. Preferably, 32 units are used to assign dynamic weights to the current load, highlighting the influence of key operating conditions and forming the third feature set. This modeling method effectively extracts the internal correlation between time series and parameters.

[0164] Step S640: Perform dimensionality reduction on the third feature set by the principal component analysis method, and use the vector concatenation method to merge the time dimension features and parameter dimension features after dimensionality reduction to obtain the fourth feature set.

[0165] In one embodiment, the principal component analysis method performs dimensionality reduction on the third feature set, retains 90% of the variance, compresses the high-dimensional features to 64 dimensions, and reduces the computational complexity.

[0166] The vector concatenation method merges the time dimension features and parameter dimension features after dimensionality reduction to form the fourth feature set. For example, the merged 64-dimensional vector integrates the temperature change trend, vibration mode, and current load characteristics, providing a comprehensive feature representation for cable condition assessment. This hierarchical processing method generates an efficient feature set through decomposition, alignment, smoothing, modeling, and dimensionality reduction, providing a reliable input for subsequent analysis.

[0167] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S700 includes: Step S710: Obtain the original data set containing defect features and operating condition parameters from the historical detection data, perform alignment processing on the time dimension data by the linear interpolation method, and standardize the operating condition parameter data by the vector normalization method to obtain the first data set.

[0168] In the field of cable condition monitoring, obtaining the original data set containing defect features and operating condition parameters from the historical detection data is the basis for constructing a prediction model. Exemplarily, the original data set may include the size data of cable surface defects, such as crack length and depth, and operating condition parameters, such as ambient temperature, current load, and mechanical vibration frequency.

[0169] Assume that the data set covers 1000 time points, the defect feature is the change of crack length over time, and the operating condition parameters include the temperature and current values recorded every minute.

[0170] The data set may have inconsistent time steps due to different sensor sampling frequencies. For example, the crack length data is recorded every minute, while the temperature data is recorded every 5 minutes.

[0171] The linear interpolation method is used to align the data in the time dimension. Specifically, for the low-frequency sampling of temperature data, it is extended to one data point per minute through the linear interpolation method. For example, if the temperature changes from 20°C to 25°C during a certain period, the temperature values at the intermediate time points are generated after interpolation, forming a time step consistent with the crack length data. This alignment method ensures the time consistency of the data for subsequent analysis.

[0172] The vector normalization method standardizes the operating condition parameter data to generate the first data set. Preferably, the temperature and current values are normalized to the range of 0 to 1. For example, if the temperature range is from 10°C to 50°C, 20°C is mapped to 0.25 after normalization, and the current load from 100 A to 500 A is mapped to the corresponding proportional values. This standardization facilitates the comparability of data with different dimensions in the model.

[0173] Step S720: Determine whether the standard deviation of the time dimension data in the first data set is lower than the preset threshold. If it is lower than the preset threshold, the sliding window method is used to smooth the time dimension data, and the principal component analysis method is used to reduce the dimension of the smoothed data to obtain the second data set.

[0174] Determining whether the standard deviation of the time dimension data in the first data set is lower than the preset threshold, such as 0.1, is to evaluate the data volatility.

[0175] It should be noted that if the standard deviation of the crack length data is lower than 0.1, it indicates that the data changes gently and may be affected by noise interference.

[0176] In a possible implementation, the sliding window method is used to smooth the time dimension data. For example, using a mean filter with a window size of 3, for the crack length data of 1.2 mm, 1.3 mm, and 1.1 mm at a certain time point, it is smoothed to 1.2 mm after smoothing, reducing the impact of instantaneous fluctuations.

[0177] The principal component analysis method reduces the dimension of the smoothed data to generate the second data set.

[0178] It can be understood that the principal component analysis compresses the original 128-dimensional features to 32 dimensions by retaining 80% of the variance, retaining the crack length change trend and key operating condition parameter information. This dimension reduction method reduces the computational complexity of subsequent modeling.

[0179] Step S730: According to the second data set, the gated recurrent unit is used to weight the operating condition parameter data, and the weighted operating condition parameter data and the time dimension data are merged through the feature splicing method to obtain the third data set.

[0180] The gated recurrent unit weights the operating condition parameter data in the second data set.

[0181] Specifically, the gated recurrent unit assigns dynamic weights to the current load. For example, 16 units are used to highlight the current impact under high temperature and high load.

[0182] The feature splicing method combines the weighted operating condition parameter data with the time dimension data to generate a third data set. For example, after splicing, a 64-dimensional vector is formed, integrating the crack length trend and the weighted temperature and current features.

[0183] Step S740: Perform time series modeling on the third data set through a long short-term memory network, output a prediction sequence of the defect expansion trend for the next three detection cycles, and obtain a feature change amount sequence.

[0184] The long short-term memory network performs time series modeling on the third data set and outputs a prediction sequence of the defect expansion trend for the next three detection cycles.

[0185] In one embodiment, the long short-term memory network includes 64 hidden units to capture the long-term dependence of the crack length on temperature and current. For example, the prediction results show that the crack length may expand from 1.2 mm to 1.5 mm in the next three cycles. Such a prediction sequence provides a key reference for cable maintenance.

[0186] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S800 includes: Step S810: Obtain a fifth feature set containing multi-dimensional data from the fusion feature set and the defect expansion trend prediction sequence, perform dimensionality reduction processing on the fifth feature set using the principal component analysis method, and standardize the data after dimensionality reduction using the vector normalization method to obtain a sixth feature set.

[0187] In the field of cable condition monitoring, the fusion feature set and the prediction sequence provide a basis for constructing a multi-dimensional fifth feature set. Exemplarily, the fusion feature set may include the cable crack width and the surface corrosion degree, and the prediction sequence includes the crack expansion trend in the next three cycles, such as the width increasing from 0.8 mm to 1.0 mm.

[0188] Operating condition parameters such as humidity and voltage fluctuations are also incorporated to form a fifth feature set, which may have a dimension as high as 100.

[0189] The principal component analysis method performs dimensionality reduction processing on the first feature set and retains the main information.

[0190] Preferably, 90% of the variance is retained through the principal component analysis, compressing from 100 dimensions to 20 dimensions, retaining the crack width trend and key operating condition parameters, and reducing the complexity of subsequent processing.

[0191] The vector normalization method standardizes the data after dimensionality reduction to generate a sixth feature set.

[0192] Specifically, the crack width ranging from 0.5 mm to 2.0 mm is normalized to the range of 0 to 1. For example, 1.0 mm is mapped to 0.33; the humidity ranging from 30% to 80% is mapped to the corresponding proportional values. This normalization ensures the comparability of features with different dimensions. It should be noted that the normalized data is convenient for subsequent threshold comparison and clustering processing.

[0193] Step S820: According to the sixth feature set, use a preset threshold group to compare each feature dimension in the sixth feature set. If the feature value of a feature dimension exceeds the upper limit of the corresponding dimension in the threshold group, it is marked as an abnormal feature, and an abnormal feature set is obtained.

[0194] According to the sixth feature set, a preset threshold group is used for feature comparison. If the feature value of a certain feature dimension exceeds the upper limit, it is marked as abnormal.

[0195] In one embodiment, the upper limit of the normalized value of the crack width is 0.8. If a data point is 0.9, it is marked as abnormal. The upper limit of the humidity is 0.7. If the value is 0.75, it is also marked as abnormal. These abnormal features form an abnormal feature set. The threshold setting is based on historical data analysis to ensure the sensitivity of anomaly detection.

[0196] Step S830: Cluster the abnormal feature set through a classification module, and use the K-means algorithm to group the abnormal feature set to obtain multiple feature subsets.

[0197] Cluster the abnormal feature set through a classification module and group it using the K-means algorithm.

[0198] It can be understood that the K-means algorithm divides the abnormal features into three groups, such as the high crack width group, the high humidity group, and the mixed abnormal group.

[0199] The grouping is based on the Euclidean distance between features, and the value of K is determined to be 3 by the elbow method. For example, the high crack width group contains crack data with normalized values from 0.85 to 0.95, reflecting potential serious defects.

[0200] Step S840: For the feature subsets, calculate the matching degree between the feature subsets and the defect threshold group, and comprehensively evaluate the matching degree through a weighted average method to obtain a classification result.

[0201] For the feature subsets, calculate the matching degree with the defect threshold group. Exemplarily, the defect threshold group defines that the serious crack width is above the normalized value of 0.9. 80% of the data in the high crack width group exceeds this value, indicating a high matching degree.

[0202] The weighted average method comprehensively evaluates the matching degree, and the weights are assigned according to the feature importance. For example, the weight of the crack width is 0.6, and the weight of the humidity is 0.4. Subsets with a high matching degree indicate that priority maintenance is required.

[0203] In a possible implementation, the clustering result of the abnormal feature set can be further refined. For example, by adjusting the K value to 4, the situation of high crack width accompanied by high humidity can be distinguished, and the composite defects can be accurately located.

[0204] Preferably, the matching degree evaluation combines the historical maintenance records to verify the reliability of the classification result. This multi-dimensional analysis improves the accuracy of defect identification and provides a reliable basis for cable maintenance.

[0205] Furthermore, for the cable defect detection method based on artificial intelligence provided in this embodiment, step S900 includes: Step S910: Obtain a multi-dimensional information feature vector containing multi-dimensional information from the cable status data, perform dimensionality reduction processing on the multi-dimensional information feature vector by using principal component analysis, and process the data after dimensionality reduction by using the vector normalization method to obtain a standardized feature set.

[0206] In the field of cable status monitoring, obtaining the multi-dimensional information feature vector is the core of analyzing the cable health status.

[0207] Exemplarily, the multi-dimensional information feature vector may include dimensions such as cable insulation resistance, partial discharge intensity, temperature distribution, and operating load.

[0208] The insulation resistance reflects the degree of cable aging, the partial discharge intensity indicates internal defects, and the temperature and load are related to the operating environment. These data are collected in real time from sensors to form an initial high-dimensional feature vector, and the dimension may reach 50 dimensions.

[0209] It should be noted that the high-dimensional data increases the analysis complexity, so further processing is required.

[0210] Principal component analysis is a commonly used method for dimensionality reduction processing and is used to retain key information. Specifically, principal component analysis projects the original features onto a new coordinate system through a linear transformation, and preferentially retains the dimensions with larger variances. For example, the 50-dimensional feature vector is compressed to 15 dimensions through principal component analysis, and the main information such as insulation resistance and partial discharge intensity is retained. This dimensionality reduction reduces the redundant dimensions and facilitates subsequent processing.

[0211] The vector normalization method standardizes the data after dimensionality reduction to ensure the comparability of features with different dimensions. In one embodiment, the insulation resistance is normalized from 10 MΩ to 100 MΩ to the interval of 0 to 1, for example, 50 MΩ is mapped to 0.4; the partial discharge intensity is mapped from 100 pC to 500 pC to 0 to 1, for example, 300 pC is mapped to 0.5.

[0212] Preferably, the standardized feature set after normalization provides a consistent basis for threshold comparison.

[0213] Step S920: According to the standardized feature set, compare each feature dimension with a preset first threshold and a second threshold. If the feature value of a feature dimension exceeds the first threshold, it is marked as an abnormal feature; if the feature value of a feature dimension exceeds the second threshold, it is marked as a critical feature, obtaining an abnormal feature set and a critical feature set.

[0214] According to the standardized feature set, perform feature comparison using the first threshold and the second threshold. Exemplarily, the first threshold is set to 0.7 and the second threshold is 0.9. If the normalized value of the insulation resistance reaches 0.75, it is marked as abnormal; if it reaches 0.95, it is marked as critical. If the normalized value of the partial discharge intensity is 0.8, it is marked as abnormal; if it is 0.92, it is marked as critical. These markings form an abnormal feature set and a critical feature set. It can be understood that critical features indicate a higher risk and need to be prioritized for attention.

[0215] Step S930: For the abnormal feature set and the critical feature set, use the K-means algorithm to cluster and group the features, obtaining multiple feature subsets.

[0216] For the abnormal feature set and the critical feature set, the K-means algorithm is used for clustering and grouping. For example, the algorithm divides the abnormal features into three groups: the high insulation resistance abnormal group, the high partial discharge abnormal group, and the mixed abnormal group. The grouping is based on the distance between features, and the value of K is determined to be 3 through analysis. In a possible implementation, the high partial discharge group contains data with normalized values from 0.85 to 0.95, reflecting the risk of serious defects.

[0217] Step S940: Calculate the matching degree value between the feature subsets and the defect levels through a weighted evaluation method, determine the defect levels and the critical status, and obtain the determination result.

[0218] Calculate the matching degree between the feature subsets and the defect levels through a weighted evaluation method. Specifically, the defect levels are divided into minor, medium, and severe, corresponding to different threshold ranges respectively. For example, a severe defect is defined as a normalized partial discharge value above 0.9. The weights are assigned according to the importance of the features. For example, the weight of the partial discharge is 0.5, the insulation resistance is 0.3, and the temperature is 0.2. 90% of the data in the high partial discharge group exceeds 0.9, with a high matching degree, and it is determined as a severe defect. The critical status indicates a potential deterioration trend.

[0219] In one embodiment, the clustering result is further refined by combining the operating load data. For example, adjust the value of K to 4 to distinguish the situation of high partial discharge accompanied by high temperature and accurately locate the composite defect.

[0220] Preferably, the matching degree evaluation refers to historical data to enhance the reliability of the determination. This multi-dimensional analysis provides an accurate basis for cable maintenance.

[0221] The present invention relates to an artificial intelligence-based cable defect detection system for implementing the above-mentioned artificial intelligence-based cable defect detection method. The artificial intelligence-based cable defect detection system includes a first acquisition module, a second acquisition module, a first generation module, a second generation module, a third acquisition module, a judgment module, an output module, a calculation module, and a determination module. Among them, the first acquisition module is used to obtain cable detection data from a multimodal data acquisition device, and the multimodal data acquisition device includes an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit; the second acquisition module is used to perform standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multimodal data set; the first generation module is used to extract features from the multimodal data set using a convolutional neural network, and generate a preliminary feature set by alternately performing three convolutional operations and pooling operations; the second generation module is used to obtain multimodal feature vectors according to the preliminary feature set, and generate a dynamic weight distribution of each modal feature through an attention weight calculation module; the third acquisition module is used to perform weighted fusion on the multimodal feature vectors based on the dynamic weight distribution to obtain a fused feature set; the judgment module is used to judge whether the fused feature set contains a working condition parameter feature vector. If the fused feature set contains a working condition parameter feature vector, the fused feature set is input into a long short-term memory network for time series modeling; the output module is used to output a defect expansion trend prediction sequence through the long short-term memory network, and the defect expansion trend prediction sequence includes the feature change amounts in the next three detection cycles; the calculation module is used to input the fused feature set and the defect expansion trend prediction sequence into a classification module, and calculate the feature matching degree using a preset defect threshold group; the determination module is used to determine the cable defect level according to the feature matching degree calculation result. When the matching degree exceeds the first threshold, it is determined that there is a defect. When the matching degree exceeds the second threshold, it is determined that the defect reaches a critical state.

[0222] Compared with the prior art, the artificial intelligence-based cable defect detection method and system provided in this embodiment obtain cable detection data through a multimodal data acquisition device, perform standardization processing and time alignment on the data, extract features using a convolutional neural network, and perform dynamic weighted fusion on multimodal features through an attention mechanism. If working condition parameters are included, a long short-term memory network is used for time series modeling to predict the defect expansion trend in the next three detection cycles. Finally, the fused features and the prediction sequence are input into a classification module, and the defect level is determined based on a preset threshold. This embodiment realizes the accurate detection of cable defects and the prediction of future development trends, effectively improves the safety and reliability of the power system, provides a scientific basis for cable maintenance decision-making, and has important engineering application value.

[0223] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An artificial intelligence-based cable defect detection method, characterized in that, Including the following steps: Obtain cable detection data from a multi-modal data acquisition device, where the multi-modal data acquisition device includes an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit; Perform standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multi-modal data set; Use a convolutional neural network to extract features from the multi-modal data set, and generate a preliminary feature set by alternately performing three convolution operations and pooling operations; Obtain a multi-modal feature vector according to the preliminary feature set, and generate a dynamic weight distribution of each modal feature through an attention weight calculation module; Perform weighted fusion on the multi-modal feature vector based on the dynamic weight distribution to obtain a fusion feature set; Determine whether the fusion feature set contains a working condition parameter feature vector. If the fusion feature set contains a working condition parameter feature vector, input the fusion feature set into a long short-term memory network for time series modeling; Output a defect expansion trend prediction sequence through the long short-term memory network, where the defect expansion trend prediction sequence includes the feature change amounts in the next three detection cycles; Input the fusion feature set and the defect expansion trend prediction sequence into a classification module, and calculate the feature matching degree using a preset defect threshold group; Determine the cable defect level according to the feature matching degree calculation result. When the matching degree exceeds the first threshold, it is determined that there is a defect. When the matching degree exceeds the second threshold, it is determined that the defect reaches a critical state.

2. The method for detecting cable defects based on artificial intelligence according to claim 1, wherein, The steps of performing standardization processing and time dimension alignment operations on the cable detection data to obtain a normalized multi-modal data set include: Obtain the original cable detection data from a multi-modal data acquisition device, and perform anomaly detection on the cable surface temperature distribution data, the cable internal defect echo signal, and the cable appearance image data in the original cable detection data using a data cleaning method to obtain a first data set; According to the first data set, use a format conversion method to convert data of different modalities into a unified structure. If the time stamp deviation between the cable surface temperature distribution data and the cable internal defect echo signal exceeds a preset threshold, align the time stamps by interpolation to obtain a second data set; For the second data set, use the z-score normalization method to standardize the cable surface temperature distribution data and the cable internal defect echo signal, and perform gray normalization on the cable appearance image data at the same time to obtain a third data set; Slice the third data set through a batch processing method, and splice the sliced cable surface temperature distribution data, the cable internal defect echo signal, and the cable appearance image data in the time dimension using a multi-modal fusion algorithm to obtain a normalized multi-modal data set.

3. The cable defect detection method based on artificial intelligence according to claim 1, wherein The steps of using a convolutional neural network to extract features from the multi-modal data set and generating a preliminary feature set by alternately performing three convolution operations and pooling operations include: Obtain the cable surface temperature distribution data, the echo signals of internal cable defects, and the cable appearance image data from the multimodal dataset. Use the sequence resampling method to convert the cable surface temperature distribution data and the echo signals of internal cable defects into two-dimensional matrices that match the resolution of the appearance images, obtaining a first intermediate dataset; For the first intermediate dataset, if the timestamp of the cable surface temperature distribution data or the echo signals of internal cable defects deviates from the cable appearance image data by more than a preset threshold, align the timestamps by the linear interpolation method, and use the matrix splicing method to merge the aligned multi-channel matrices, obtaining a second intermediate dataset; According to the second intermediate dataset, perform the first convolution operation using a convolutional neural network, extract features using a preset convolutional kernel for the multi-channel matrix, and downsample through the max pooling operation to obtain a first feature set; For the first feature set, alternately perform the second and third convolution and max pooling operations, and use the channel weighting method to fuse the results of multiple convolution and pooling operations to obtain a preliminary feature set.

4. The method for detecting cable defects based on artificial intelligence according to claim 1, wherein, The steps of obtaining the multimodal feature vector according to the preliminary feature set and generating the dynamic weight distribution of each modal feature through the attention weight calculation module include: Obtain the multimodal feature vector according to the preliminary feature set, use the matrix decomposition method to split the multimodal feature vector into independent modal matrices, and convert the independent modal matrices into one-dimensional feature vectors through the vector mapping method to obtain a first feature vector set; For the first feature vector set, use the attention weight calculation module to weight each modal feature vector, and calculate the dynamic weight distribution through the softmax function to obtain a weighted feature vector set; If the weight value of any modality in the weighted feature vector set is lower than the preset threshold, adjust the weight distribution by the linear interpolation method, and use the vector splicing method to merge the adjusted weighted feature vectors to obtain a second feature vector set; According to the second feature vector set, use the principal component analysis method to reduce the dimension of the high-dimensional feature vectors, and generate an optimized multimodal feature vector through the matrix recombination method to obtain a final feature vector set.

5. The method for detecting cable defects based on artificial intelligence according to claim 1, wherein The steps of weighted fusion of the multimodal feature vector based on the dynamic weight distribution to obtain a fusion feature set include: Obtain the initial feature matrix from the multimodal feature vector, use the matrix decomposition method to split the initial feature matrix into independent modal sub-matrices, and convert the independent modal sub-matrices into one-dimensional vectors through the vector mapping method to obtain a first vector set; For the first vector set, use the attention mechanism to calculate the dynamic weights of each modal vector, and normalize the dynamic weights through the softmax function to obtain the dynamic weight distribution; If the weight value of any modality in the dynamic weight distribution is lower than the preset threshold, adjust the dynamic weight distribution by the linear interpolation method, and weight the first vector set through the vector weighting method to obtain a weighted vector set; According to the weighted vector set, the weighted vectors of each modality are merged using a vector concatenation method, and the merged vectors are dimensionally reduced by a principal component analysis method to obtain a fused feature set.

6. The method for detecting cable defects based on artificial intelligence according to claim 1, wherein Determine whether the fused feature set contains a working condition parameter feature vector. If the fused feature set contains a working condition parameter feature vector, the steps of inputting the fused feature set into a long short-term memory network for time series modeling include: Obtain a working condition parameter feature vector containing working condition parameters from the fused feature set, split the working condition parameter feature vector into a time dimension sub-vector and a parameter dimension sub-vector using a matrix decomposition method, and perform alignment processing on the time dimension sub-vector and the parameter dimension sub-vector by a linear interpolation method to obtain a first feature set; If the standard deviation of the time dimension sub-vector in the first feature set is lower than a preset threshold, smooth the time dimension sub-vector using a sliding window method, and standardize the smoothed sub-vector by a vector normalization method to obtain a second feature set; According to the second feature set, perform time series modeling on the time dimension sub-vector using a long short-term memory network, and perform weighted processing on the parameter dimension sub-vector by a gated recurrent unit to obtain a third feature set; Perform dimensional reduction processing on the third feature set by a principal component analysis method, and merge the dimensionally reduced time dimension features and parameter dimension features using a vector concatenation method to obtain a fourth feature set.

7. The method for detecting cable defects based on artificial intelligence according to claim 1, wherein Output a defect expansion trend prediction sequence through the long short-term memory network. The steps for the defect expansion trend prediction sequence to include the feature change amounts in the next three detection cycles include: Obtain an original data set containing defect features and working condition parameters from historical detection data, perform alignment processing on the time dimension data by a linear interpolation method, and standardize the working condition parameter data by a vector normalization method to obtain a first data set; Determine whether the standard deviation of the time dimension data in the first data set is lower than a preset threshold. If it is lower than the preset threshold, smooth the time dimension data using a sliding window method, and perform dimensional reduction processing on the smoothed data by a principal component analysis method to obtain a second data set; According to the second data set, perform weighted processing on the working condition parameter data by a gated recurrent unit, and merge the weighted working condition parameter data and the time dimension data by a feature concatenation method to obtain a third data set; Perform time series modeling on the third data set through the long short-term memory network, output a defect expansion trend prediction sequence containing the next three detection cycles, and obtain a feature change amount sequence.

8. The method for detecting cable defects based on artificial intelligence according to claim 1, characterized in that Input the fused feature set and the defect expansion trend prediction sequence into a classification module, and the steps for calculating the feature matching degree using a preset defect threshold group include: Obtain a fifth feature set containing multi-dimensional data from the fused feature set and the defect expansion trend prediction sequence, perform dimensional reduction processing on the fifth feature set by a principal component analysis method, and standardize the dimensionally reduced data by a vector normalization method to obtain a sixth feature set; According to the sixth feature set, each feature dimension in the sixth feature set is compared using a preset threshold group. If the feature value of a feature dimension exceeds the upper limit of the corresponding dimension in the threshold group, it is marked as an abnormal feature, and an abnormal feature set is obtained; The abnormal feature set is subjected to clustering processing by a classification module, and the K-means algorithm is used to group the abnormal feature set to obtain multiple feature subsets; For the feature subsets, the matching degree between the feature subsets and the defect threshold group is calculated, and the matching degree is comprehensively evaluated by a weighted average method to obtain a classification result.

9. The method for detecting cable defects based on artificial intelligence according to claim 1, characterized in that, The steps of determining the cable defect level according to the calculation result of the feature matching degree, where it is determined that there is a defect when the matching degree exceeds the first threshold, and it is determined that the defect reaches a critical state when the matching degree exceeds the second threshold, include: A multi-dimensional information feature vector containing multi-dimensional information is obtained from the cable state data, and principal component analysis is used to perform dimensionality reduction processing on the multi-dimensional information feature vector, and the dimensionality-reduced data is processed by a vector normalization method to obtain a standardized feature set; According to the standardized feature set, each feature dimension is compared using a preset first threshold and second threshold. If the feature value of a feature dimension exceeds the first threshold, it is marked as an abnormal feature; if the feature value of a feature dimension exceeds the second threshold, it is marked as a critical feature, and an abnormal feature set and a critical feature set are obtained; For the abnormal feature set and the critical feature set, the K-means algorithm is used to cluster and group the features to obtain multiple feature subsets; The matching degree value between the feature subsets and the defect level is calculated by a weighted evaluation method to determine the defect level and the critical state, and a determination result is obtained.

10. An artificial intelligence-based cable defect detection system for implementing the artificial intelligence-based cable defect detection method according to any one of claims 1 to 9, characterized in that, The cable defect detection system based on artificial intelligence includes: A first acquisition module for acquiring cable detection data from a multi-modal data acquisition device, where the multi-modal data acquisition device includes an infrared thermal imaging unit, an ultrasonic detection unit, and an optical sensing unit; A second acquisition module for performing standardization processing and time dimension alignment operation on the cable detection data to obtain a normalized multi-modal data set; A first generation module for extracting features from the multi-modal data set using a convolutional neural network, and generating a preliminary feature set by alternately performing three convolutional operations and pooling operations; A second generation module for obtaining a multi-modal feature vector according to the preliminary feature set, and generating a dynamic weight distribution of each modal feature by an attention weight calculation module; A third acquisition module for weighted fusion of the multi-modal feature vector based on the dynamic weight distribution to obtain a fusion feature set; A judgment module for judging whether the fusion feature set contains a working condition parameter feature vector. If the fusion feature set contains a working condition parameter feature vector, the fusion feature set is input into a long short-term memory network for time series modeling; An output module for outputting a defect expansion trend prediction sequence through the long short-term memory network, where the defect expansion trend prediction sequence includes the feature change amounts in the next three detection cycles; A calculation module, configured to input the fusion feature set and the defect expansion trend prediction sequence into a classification module, and calculate a feature matching degree by using a preset defect threshold group; A determination module, configured to determine a cable defect level according to a result of the feature matching degree calculation, determine that there is a defect when the matching degree exceeds a first threshold, and determine that the defect reaches a critical state when the matching degree exceeds a second threshold.

Images