Intelligent low-voltage cable insulation monitoring method and system

Abstract

The invention discloses an intelligent low-voltage cable insulation monitoring method and system. The method comprises the steps of data acquisition, preprocessing, establishment of a low-voltage cable insulation evaluation model and low-voltage cable insulation monitoring. The invention belongs to the field of cable insulation monitoring, and particularly relates to an intelligent low-voltage cable insulation monitoring method and system. According to the scheme, partial discharge characteristics are enhanced based on pulse injection branches; excessive deviation is avoided through a dynamic soft weight distribution mechanism; focusing resource allocation through a fault detection network; based on construction of an insulation stability loss function, contributions of minority cable samples in the overall gradient are strengthened; a window weight factor is introduced, an early window with insufficient information is ignored, a critical section is focused, window weight likelihood loss is constructed, and fault critical section identification is strengthened; introducing a fault severity regression branch to promote convergence of a fault critical section; fault response adjustment is constructed based on a compensation function, and the false positive rate and the omission ratio are reduced; and the cable insulation monitoring effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cable insulation monitoring, and specifically refers to an intelligent low-voltage cable insulation monitoring method and system. Background Art

[0002] The low-voltage cable insulation monitoring method is to collect key diagnostic data in real time or periodically through sensors such as partial discharge, leakage current, temperature, and humidity during the operation or power-off maintenance of the cable, and use data analysis technology to evaluate the insulation health status. However, the general low-voltage cable insulation monitoring method has problems such as being unable to detect early weak discharges, delaying the warning of potential insulation faults, and scarce samples of insulation damage or deterioration, resulting in low monitoring reliability; the general low-voltage cable insulation monitoring method treats all monitoring windows equally, and early weak discharge and leakage current signals are often submerged by noise, resulting in poor monitoring effects. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an intelligent low-voltage cable insulation monitoring method and system. Aiming at the problems of the general low-voltage cable insulation monitoring method, such as being unable to detect early weak discharges, delaying the warning of potential insulation faults, and scarce samples of insulation damage or deterioration, resulting in low monitoring reliability, this solution improves the recognition ability of rare fault modes by constructing four enhancement branches: perturbation injection, time warping, sample synthesis, and pulse injection; enhances the partial discharge characteristics based on the pulse injection branch; avoids over-bias through a dynamic soft weight allocation mechanism; focuses on resource allocation through a fault detection network; strengthens the contribution of minority-class cable samples to the overall gradient based on the constructed insulation stability loss function; and thus improves the reliability of subsequent insulation monitoring. Aiming at the problem that the general low-voltage cable insulation monitoring method treats all monitoring windows equally, and early weak discharge and leakage current signals are often submerged by noise, resulting in poor monitoring effects, this solution introduces a window weight factor, ignores the early windows with insufficient information, focuses on the critical section, and constructs a window weight likelihood loss to strengthen the identification of the fault critical section; introduces a fault severity regression branch to incorporate continuous degradation signals into the penalty to promote the convergence of the fault critical section; and constructs a fault response adjustment based on a compensation function to reduce false positives and missed detections; and thus improves the cable insulation monitoring effect.

[0004] The technical solution adopted by the present invention is as follows: An intelligent low-voltage cable insulation monitoring method provided by the present invention includes the following steps:

[0005] Step S1: Data acquisition; [[ID=2,0]]

[0006] Step S2: Pretreatment;

[0007] Step S3: Establish a low-voltage cable insulation evaluation model;

[0008] Step S4: Low-voltage cable insulation monitoring.

[0009] Further, in step S1, the data acquisition is to collect historical low-voltage cable monitoring data; label the cable status as a data tag; perform standardization processing on the historical low-voltage cable monitoring data, and construct an initial low-voltage cable monitoring data set using a sliding window segmentation.

[0010] Further, in step S2, the preprocessing is to construct 4 enhancement branches for the initial low-voltage cable monitoring data set, each branch corresponding to an enhancement strategy, and adopt a dynamic weighting and insulation stability loss mechanism to automatically allocate learning weights, focusing on strengthening the enhancement effect of minority-class cable fault samples; and then construct a low-voltage cable monitoring data set; specifically including:

[0011] Step S21: Enhancement branch construction unit; the enhancement branches include a perturbation injection branch, a time warping branch, a sample synthesis branch, and a pulse injection branch; each enhancement branch undergoes a single forward training to calculate the loss value of the enhanced cable samples on the main classification network , perform softmax on the loss to obtain the learning weight , expressed as: <o:p>< / o:p> ; Another fault detection network output , the training label is the inverse exponential mapping, expressed as: ; where is the loss value of the enhanced cable samples on the main classification network after a single forward training of the f-th enhancement branch; n and f are both enhancement branch indices; is the training label of the fault detection network; fuse the main classification loss , the enhanced branch weighted loss, and the fault detection network loss, and the initial enhanced loss is expressed as: ; ; where , and are enhanced loss weights; Use cross-entropy loss; is the loss of the enhanced cable samples of the enhancement branch; is the mapping of the main classification network; is the enhanced cable sample of the enhancement branch; y is the true label of the cable sample;

[0012] Step S22: Sample balance compensation; design an insulation stability loss function, let represent the classification error of the i-th enhanced cable sample, then the insulation stability loss is defined as ; The final total enhanced loss is expressed as: ; where, is the category probability value of the enhanced cable sample; and are the sensitivity coefficients; and are the offset coefficients; is the adjustment factor; is the total classification error of the enhanced cable sample; is the control coefficient.

[0013] Furthermore, in step S3, the establishment of the low-voltage cable insulation evaluation model specifically includes the following steps:

[0014] Step S31: Model architecture design; Based on the low-voltage cable monitoring data set, use Multi-ScaleCNN + BiLSTM + attention mechanism as the low-voltage cable insulation evaluation model architecture; Extract local features through multi-scale convolution ; Capture long-term and short-term dependencies through bidirectional LSTM; Aggregate feature information based on adaptive attention; Map through the fully connected classification layer to output the probability distribution of each monitoring category ; Update parameters using the gradient descent algorithm; Verify the model performance based on accuracy and recall;

[0015] Step S32: Loss function design; Specifically include:

[0016] Step S321: Define the window weight factor, expressed as: ; where, is the window weight factor and x is the function variable; and are the window integrity thresholds;

[0017] Step S322: Construct the window weight likelihood loss , and dynamically adjust the penalty according to the ratio of the window length to the full length of the fault, expressed as: ; where, is the time interval length of the i-th cable sample in the j-th window; is the entire fault process length of the i-th cable sample; and are the Gaussian likelihoods of predicting the cable state categories and respectively;

[0018] Step S323: Construct the fault response adjustment; Introduce the fault severity regression branch, and the fault response adjustment is expressed as: ; ; ; where, is the score of the low - voltage cable insulation assessment model for the window; is the compensation function; is the time - interval length of the i - th cable sample in the (j - 1) - th window; q is a constant coefficient used to control the magnitude of the overall impact; Q is a temperature parameter used to adjust the slope and change speed of the function; tanh(·) is the tanh function; and is the predicted severity score of the fault severity regression branch;

[0019] Step S324: The total loss is expressed as: ; where is the total number of cable samples; is the monitoring loss coefficient.

[0020] Furthermore, in step S4, the low - voltage cable insulation monitoring is to collect the low - voltage cable monitoring data in real - time and input it into the low - voltage cable insulation assessment model, and conduct low - voltage cable insulation monitoring based on the cable status category output by the low - voltage cable insulation assessment model; if the cable status category is one of moderate fault, severe fault, and environmental anomaly, warning processing is carried out.

[0021] An intelligent low - voltage cable insulation monitoring system provided by the present invention includes a data acquisition module, a pre - processing module, a low - voltage cable insulation assessment model establishment module, and a low - voltage cable insulation monitoring module;

[0022] The data acquisition module collects historical low - voltage cable monitoring data and constructs an initial low - voltage cable monitoring data set;

[0023] The pre - processing module constructs four enhancement branches of perturbation injection, time warping, sample synthesis, and pulse injection for the initial data set, and constructs a low - voltage cable monitoring data set based on the insulation stability loss mechanism;

[0024] The low - voltage cable insulation assessment model establishment module establishes a low - voltage cable insulation assessment model based on the low - voltage cable monitoring data set, combined with window - weight likelihood loss and fault response adjustment introducing a fault severity regression branch;

[0025] The low - voltage cable insulation monitoring module monitors the real - time low - voltage cable monitoring data based on the low - voltage cable insulation assessment model.

[0026] The beneficial effects achieved by the present invention using the above - mentioned scheme are as follows:

[0027] (1) Aiming at the problems existing in the general low-voltage cable insulation monitoring method, such as the inability to detect early weak discharges, delaying the warning of potential insulation faults, and the scarcity of samples of insulation damage or deterioration, which lead to low monitoring reliability. This solution enhances the recognition ability of rare fault modes by constructing four enhancement branches: perturbation injection, time distortion, sample synthesis, and pulse injection; enhances the partial discharge characteristics based on the pulse injection branch; avoids over-bias through a dynamic soft weight allocation mechanism; focuses resource allocation through a fault detection network; strengthens the contribution of minority-class cable samples to the overall gradient based on constructing an insulation stability loss function; and thus improves the reliability of subsequent insulation monitoring.

[0028] (2) Aiming at the problem that the general low-voltage cable insulation monitoring method treats all monitoring windows equally, and early weak discharge and leakage current signals are often submerged by noise, resulting in poor monitoring effects. This solution introduces a window weight factor, ignores early windows with insufficient information, focuses on the critical section, and constructs a window weight likelihood loss to strengthen the identification of the fault critical section; introduces a fault severity regression branch to incorporate continuous degradation signals into the penalty to promote the convergence of the fault critical section; and constructs a fault response adjustment based on a compensation function to reduce the false positive and missed detection rates; and thus improves the cable insulation monitoring effect. Description of the Drawings

[0029] Figure 1 It is a schematic flow chart of an intelligent low-voltage cable insulation monitoring method provided by the present invention;

[0030] Figure 2 It is a schematic diagram of an intelligent low-voltage cable insulation monitoring system provided by the present invention.

[0031] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. Detailed Embodiments

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0034] Embodiment 1, refer to Figure 1 , an intelligent low-voltage cable insulation monitoring method provided by the present invention, the method comprising the following steps:

[0035] Step S1: Data acquisition; collect historical low-voltage cable monitoring data and construct an initial low-voltage cable monitoring data set;

[0036] Step S2: Preprocessing; construct four enhancement branches of perturbation injection, time warping, sample synthesis, and pulse injection for the initial data set, and construct a low-voltage cable monitoring data set based on the insulation stability loss mechanism;

[0037] Step S3: Establish a low-voltage cable insulation evaluation model; establish a low-voltage cable insulation evaluation model based on the low-voltage cable monitoring data set in combination with the window weight likelihood loss and the fault response adjustment introducing the fault severity regression branch;

[0038] Step S4: Low-voltage cable insulation monitoring; monitor the real-time low-voltage cable monitoring data based on the low-voltage cable insulation evaluation model.

[0039] Embodiment 2, refer to Figure 1 , this embodiment is based on the above embodiment. In step S1, the data acquisition is to collect historical low-voltage cable monitoring data; the historical low-voltage cable monitoring data includes partial discharge count, average discharge amplitude, leakage current, and ambient temperature and humidity; label the cable status as a data label; the cable status includes healthy, early insulation deterioration, moderate fault, severe fault, and environmental anomaly; perform standardization processing on the historical low-voltage cable monitoring data and construct an initial low-voltage cable monitoring data set using a sliding window segmentation.

[0040] Embodiment 3, refer to Figure 1 , this embodiment is based on the above embodiment. In step S2, the preprocessing is to construct 4 enhancement branches for the initial low-voltage cable monitoring data set, each branch corresponding to an enhancement strategy, and adopt a dynamic weighting and insulation stability loss mechanism to automatically allocate learning weights, focusing on strengthening the enhancement effect of minority-class cable fault samples; it is of great significance for accurately identifying low-voltage cable insulation faults and ensuring the stable operation of the power system; and then construct a low-voltage cable monitoring data set; specifically including:

[0041] Step S21: Enhance the branch construction unit; the enhanced branches include a perturbation injection branch, a time warping branch, a sample synthesis branch, and a pulse injection branch. The perturbation injection branch simulates random interferences such as acquisition device noise and electromagnetic interference, and adds Gaussian white noise to the original cable sample data. The time warping branch simulates the fault development speed or sampling rate fluctuation, introduces non-linear deformation on the time axis, and resamples the original signal to a new time axis using piecewise linear interpolation. The sample synthesis branch expands the data by interpolation between a small number of fault samples. Since partial discharge is the most typical physical phenomenon of low-voltage cable insulation degradation, but its amplitude is low and duration is short in the original monitoring data and is extremely easy to be ignored, controllable PD pulse injection is introduced. The pulse injection branch is expressed as: ; where and are the enhanced cable sample data and the original cable sample data of the pulse injection branch respectively; t is the cable sample sampling time; U is the number of partial discharge pulses, and u is the pulse index; is the amplitude of the u-th pulse; is the absolute time position at which the u-th pulse occurs; is the equivalent half-width of the pulse; is a gating function, which outputs 1 only in the effective interval of the pulse and 0 otherwise; Since different fault types show different characteristics in the monitoring data and some fault features are relatively weak and difficult to enhance; Therefore, the more difficult the enhancement effect is, the higher learning weight will be obtained, which can prevent simple enhancement strategies from dominating the training and ensure that the model can comprehensively learn various fault features; After one forward training for each enhanced branch, calculate the loss value of the enhanced cable sample on the main classification network, perform softmax on these losses to obtain the learning weight , which is expressed as: ; Another fault detection network outputs , and the training label is the inverse exponential mapping, which is expressed as: ; where is the loss value of the enhanced cable sample on the main classification network after one forward training for the f-th enhanced branch; both n and f are enhanced branch indices; is the training label of the fault detection network; Through the fault detection network, ensure the alignment of the two distributions of high loss → large label and high loss → large weight, so that the enhanced branches can allocate resources more reasonably during the training process; Integrate the main classification loss , the weighted loss of the enhanced branches, and the loss of the fault detection network. The initial enhanced loss is expressed as: ; ; where , and are the enhanced loss weights; Use cross - entropy loss; is the loss of the cable samples after enhancement by the enhancement branch; is the mapping of the main classification network; is the cable sample after enhancement by the enhancement branch; y is the true label of the cable sample; automatically balance the contributions among multiple enhancement strategies, pay more attention to the few - class fault cable samples that are difficult to enhance, so as to improve the recognition ability of rare fault patterns and more accurately detect potential insulation fault hidden dangers in low - voltage cable insulation monitoring;

[0042] Step S22: Sample balance compensation; In low - voltage cable insulation monitoring data, the number of cable samples in the healthy state is usually much larger than that of fault cable samples; By controlling the overall loss contribution of the majority class in the enhanced cable samples, prevent the model from over - fitting a large number of healthy cable samples; And during the operation of low - voltage cables, the occurrence of insulation deterioration and faults is relatively rare, and the features contained in these few - class fault cable samples are crucial for accurate fault monitoring; Therefore, strengthen the contribution of the few - class enhanced cable samples to the gradient and improve the learning strength of the fault pattern; Design an insulation stability loss function, let represents the classification error of the i - th enhanced cable sample, then the insulation stability loss is defined as ; The final enhanced total loss is expressed as: ; Among them, is the class probability value of the enhanced cable sample; and are sensitivity coefficients, used to control the weights and influence degrees of different parts in the enhancement loss function; and are offset coefficients, used to adjust the offset of the enhancement loss function; is a regulation factor; is the total classification error of the enhanced cable samples; is a control coefficient; Pay attention to rare but critical insulation deterioration features; The features in the initial stage of insulation deterioration are often not obvious and account for a small proportion in the dataset, but these features are the key to early fault detection; By increasing the attention to these features, improve the subsequent fault detection rate and timely discover potential insulation faults to ensure the safe operation of low - voltage cables.

[0043] By performing the above operations, for the general low-voltage cable insulation monitoring method, there are problems such as the inability to detect early weak discharges, delaying the warning of potential insulation faults, and scarce samples of insulation damage or deterioration, which lead to low monitoring reliability. This solution improves the recognition ability of rare fault modes by constructing four enhancement branches: perturbation injection, time warping, sample synthesis, and pulse injection; enhances the partial discharge characteristics based on the pulse injection branch; avoids over-bias through the dynamic soft weight allocation mechanism; focuses on resource allocation through the fault detection network; strengthens the contribution of minority-class cable samples to the overall gradient based on the constructed insulation stability loss function; and thus improves the reliability of subsequent insulation monitoring.

[0044] Example 4, refer to Figure 1 , based on the above example, in step S3, establishing the low-voltage cable insulation evaluation model specifically includes the following steps:

[0045] Step S31: Model architecture design; based on the low-voltage cable monitoring dataset, using Multi-ScaleCNN + BiLSTM + attention mechanism as the low-voltage cable insulation evaluation model architecture; extracting local features through multi-scale convolution , expressed as: ; capturing long-term and short-term dependencies through bidirectional LSTM, expressed as: ; aggregating feature information based on adaptive attention, expressed as: ; mapping through the fully connected classification layer to output the probability distribution of each monitoring category , expressed as: ; where, is the jth cable sample of the low-voltage cable monitoring dataset; and are the convolutional kernel weights and biases; ReLU is the ReLU activation function; and are the forward and backward hidden state variables at the current time step respectively; is the bidirectional long short-term memory network; is the feature sequence of the previous time step; is the attention weight; and are the forward and backward hidden state variables at the kth time step respectively; is the weight vector of the attention mechanism; z is the feature vector aggregated through the attention mechanism; softmax is the softmax function; and are the weight matrix and bias vector of the fully connected layer respectively; the gradient descent algorithm is used to update the parameters; the model performance is verified based on accuracy and recall;

[0046] Step S32: Loss function design; specifically includes:

[0047] Step S321: Define the window weight factor; for a short window, insufficient information does not contribute to the loss. For a complete window closer to the fault end point, the weight increases, which is expressed as: ; In the low-voltage cable insulation monitoring, the fragments of early events may contain more noise and incomplete information; through the window weight factor, those fragments with insufficient information are ignored, and efforts are concentrated on processing more complete fragments closer to the fault end point; where is the window weight factor and x is the function variable; and are the window integrity thresholds;

[0048] Step S322: Construct the window weight likelihood loss , and dynamically adjust the penalty according to the ratio of the window length to the full length of the fault, which is expressed as: ; where is the time interval length of the i-th cable sample in the j-th window; is the full length of the fault process of the i-th cable sample; and are the Gaussian likelihoods of predicting the cable state categories and respectively; In the low-voltage cable insulation monitoring, the fragments closer to the critical section of insulation failure with high partial discharge intensity and increasing leakage current can contain more key information about the fault; by constructing the window weight likelihood loss, the loss weights of these fragments are larger, and the evaluation model will pay more attention to these fragments during the training process, thereby improving the sensitivity of early warning;

[0049] Step S323: Construct the fault response adjustment; for the three common errors in the monitoring process, including: early alarm, the model judges a fault but has not reached the true starting point; delayed alarm, the window crosses the true end point but the score is insufficient; score reversal, the current window is closer to the fault than the previous window but the score is lower instead; construct the fault response adjustment; when the score of the current window is higher than that of the previous window and meets the abnormal conditions of temperature, humidity, and PD characteristics, increase the penalty; and introduce the fault severity regression branch to quantify the continuous degree of insulation degradation, accelerating the convergence of the evaluation model to the fault stage, the fault response adjustment is expressed as: ; ; ; where is the score of the low-voltage cable insulation evaluation model for the window; is the compensation function; is the time interval length of the i-th cable sample in the (j - 1)-th window; q is a constant coefficient used to control the magnitude of the overall influence; Q is the temperature parameter used to adjust the slope and change speed of the function; tanh(·) is the tanh function; and is the predicted severity score of the fault severity regression branch, and the predicted severity score ∈ [0, 1], where 0 represents insulation health and 1 represents critical failure; is the severity weight coefficient; in the insulation monitoring of low-voltage cables, environmental interference such as sudden temperature rise can lead to false positive monitoring results; by constructing a fault response adjustment, these false monitoring results caused by environmental interference are suppressed to ensure the stability of continuous monitoring; when early warning or score reversal occurs, increase the penalty to make the evaluation model judge more cautiously;

[0050] Step S324: Total loss is expressed as: ; where is the total number of cable samples; is the monitoring loss coefficient.

[0051] By performing the above operations, for the general low-voltage cable insulation monitoring method, all monitoring windows are treated equally. Early weak discharge and leakage current signals are often submerged by noise, resulting in poor monitoring effects. In this solution, by introducing a window weight factor, early windows with insufficient information are ignored, the critical section is focused on, and a window weight likelihood loss is constructed to strengthen the identification of the fault critical section; a fault severity regression branch is introduced to incorporate continuous degradation signals into the penalty to promote the convergence of the fault critical section; and a fault response adjustment is constructed based on the compensation function to reduce false positives and missed detection rates; thereby improving the cable insulation monitoring effect.

[0052] Example Five, refer to Figure 1 , based on the above example, in step S4, the low-voltage cable insulation monitoring is to collect the low-voltage cable monitoring data in real time and input it into the low-voltage cable insulation evaluation model, and perform low-voltage cable insulation monitoring based on the cable status category output by the low-voltage cable insulation evaluation model; if the cable status category is one of moderate fault, severe fault, and environmental anomaly, warning processing is performed.

[0053] Example Six, refer to Figure 2 , based on the above example, an intelligent low-voltage cable insulation monitoring system provided by the present invention includes a data acquisition module, a preprocessing module, a low-voltage cable insulation evaluation model establishment module, and a low-voltage cable insulation monitoring module;

[0054] The data acquisition module collects historical low-voltage cable monitoring data and constructs an initial low-voltage cable monitoring data set;

[0055] The preprocessing module constructs four enhancement branches of perturbation injection, time warping, sample synthesis, and pulse injection for the initial data set, and constructs a low-voltage cable monitoring data set based on the insulation stability loss mechanism;

[0056] The low-voltage cable insulation evaluation model establishment module establishes a low-voltage cable insulation evaluation model based on the low-voltage cable monitoring data set, combined with window weight likelihood loss and introducing fault severity regression branches for fault response adjustment;

[0057] The low-voltage cable insulation monitoring module monitors the real-time low-voltage cable monitoring data based on the low-voltage cable insulation evaluation model.

[0058] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0059] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made in these embodiments without departing from the principles and spirit of the present invention.

[0060] The above description of the present invention and its embodiments is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and design similar structural modes and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent low-voltage cable insulation monitoring method, characterized in that: The method includes the following steps: Step S1: Data collection; collect historical low-voltage cable monitoring data and construct an initial low-voltage cable monitoring data set; Step S2: Preprocessing; construct four enhancement branches of perturbation injection, time warping, sample synthesis, and pulse injection for the initial data set, and construct a low-voltage cable monitoring data set based on the insulation stability loss mechanism; Step S3: Establish a low-voltage cable insulation evaluation model; establish a low-voltage cable insulation evaluation model based on the low-voltage cable monitoring data set in combination with window weight likelihood loss and fault response adjustment introducing a fault severity regression branch; Step S4: Low-voltage cable insulation monitoring; monitor the real-time low-voltage cable monitoring data based on the low-voltage cable insulation evaluation model.

2. The intelligent low-voltage cable insulation monitoring method according to claim 1, wherein: In step S2, the preprocessing is to construct 4 enhancement branches for the initial low-voltage cable monitoring data set, each branch corresponding to an enhancement strategy, and adopt a dynamic weighting and insulation stability loss mechanism to automatically allocate learning weights, focusing on strengthening the enhancement effect of minority-class cable fault samples; Furthermore, construct a low-voltage cable monitoring data set; specifically including: Step S21: Enhance the branch construction unit; the enhanced branches include a perturbation injection branch, a time warping branch, a sample synthesis branch, and a pulse injection branch; each enhanced branch undergoes one forward training to calculate the loss value of the enhanced cable sample on the main classification network , perform softmax on the loss to obtain the learning weight , which is expressed as: ; Additionally, set a fault detection network output , and the training label is the inverse exponential mapping, which is expressed as: ; where is the loss value of the enhanced cable sample on the main classification network after one forward training of the f-th enhanced branch; both n and f are enhanced branch indices; is the fault detection network training label; fuse the main classification loss , the enhanced branch weighted loss, and the fault detection network loss, and the initial enhanced loss is expressed as: ; ; where , and are the enhanced loss weights; Use cross-entropy loss; is the loss of the enhanced cable sample of the enhanced branch; is the main classification network mapping; is the enhanced cable sample of the enhanced branch; y is the true label of the cable sample; Step S22: Sample balance compensation.

3. The intelligent low-voltage cable insulation monitoring method according to claim 2, wherein: In step S2, the sample balance compensation is to design an insulation stability loss function, and let represent the classification error of the i-th enhanced cable sample, then the insulation stability loss is defined as ; The final total enhanced loss is expressed as: ; where is the class probability value of the enhanced cable sample; and are sensitivity coefficients; and are offset coefficients; is a regulation factor; is the total classification error of the enhanced cable sample; is a control coefficient.

4. An intelligent low-voltage cable insulation monitoring method according to claim 3, characterized in that: In step S3, the establishment of the low-voltage cable insulation evaluation model specifically includes the following steps: Step S31: Model architecture design; Based on the low-voltage cable monitoring dataset, use Multi-ScaleCNN + BiLSTM + attention mechanism as the low-voltage cable insulation evaluation model architecture; Extract local features through multi-scale convolution ; Capture long-term and short-term dependencies through bidirectional LSTM; Aggregate feature information based on adaptive attention; Map through a fully connected classification layer to output the probability distribution of each monitoring category ; Update parameters using the gradient descent algorithm; Verify the model performance based on accuracy and recall Step S32: Loss function design; specifically including: Step S321: Define the window weight factor, expressed as: ; where is the window weight factor and x is the function variable; and are the window integrity thresholds; Step S322: Construct the window weight likelihood loss ; Step S323: Construct a fault response adjustment ; Step S324: Total loss It is expressed as: ; where is the total number of cable samples; is the monitoring loss coefficient.

5. The intelligent low-voltage cable insulation monitoring method according to claim 4, characterized in that: In step S3, the construction of the window weight likelihood loss is dynamically adjusted according to the ratio of the window length to the entire fault length, expressed as: ; where is the time interval length of the i-th cable sample in the j-th window; is the entire fault process length of the i-th cable sample; and are the Gaussian likelihoods of predicting the cable state categories and respectively.

6. The intelligent low-voltage cable insulation monitoring method according to claim 5, characterized in that: In step S3, the construction of the fault response adjustment is to introduce a fault severity regression branch, and the fault response adjustment is expressed as: ; ; ; where is the score of the low-voltage cable insulation evaluation model for the window; is the compensation function; is the time interval length of the i-th cable sample in the (j - 1)-th window; q is a constant coefficient used to control the magnitude of the overall influence; Q is a temperature parameter used to adjust the slope and change speed of the function; tanh(·) is the tanh function; and are the predicted severity scores of the fault severity regression branch.

7. The intelligent low-voltage cable insulation monitoring method according to claim 6, wherein: In step S1, the data collection is to collect historical low-voltage cable monitoring data; label the cable status as a data label; perform normalization processing on the historical low-voltage cable monitoring data, and construct an initial low-voltage cable monitoring data set using sliding window segmentation.

8. The intelligent low-voltage cable insulation monitoring method according to claim 7, wherein: In step S4, the low-voltage cable insulation monitoring is to collect real-time low-voltage cable monitoring data and input it into the low-voltage cable insulation evaluation model, and perform low-voltage cable insulation monitoring based on the cable status category output by the low-voltage cable insulation evaluation model; If the cable status category is one of moderate fault, severe fault, and environmental anomaly, warning processing is performed.

9. An intelligent low-voltage cable insulation monitoring system for implementing an intelligent low-voltage cable insulation monitoring method as described in any one of claims 1-8, characterized in that: It includes a data collection module, a preprocessing module, a low-voltage cable insulation evaluation model establishment module, and a low-voltage cable insulation monitoring module; The data collection module collects historical low-voltage cable monitoring data and constructs an initial low-voltage cable monitoring data set; The preprocessing module constructs four enhancement branches of perturbation injection, time warping, sample synthesis, and pulse injection for the initial data set, and constructs a low-voltage cable monitoring data set based on the insulation stability loss mechanism; The low-voltage cable insulation evaluation model establishment module establishes a low-voltage cable insulation evaluation model based on the low-voltage cable monitoring data set in combination with window weight likelihood loss and fault response adjustment introducing a fault severity regression branch; The low-voltage cable insulation monitoring module monitors the real-time low-voltage cable monitoring data based on the low-voltage cable insulation evaluation model.

Images