A cable termination stress relaxation monitoring system for a high voltage cabinet

Abstract

This invention relates to the field of power equipment monitoring technology, specifically to a stress relaxation monitoring system for cable terminals in high-voltage electrical cabinets. It includes a multi-source data acquisition unit that simultaneously acquires data on the terminal's surface operating temperature, real-time micro-strain, load current, and ambient temperature; a feature extraction unit that performs time-series alignment and normalization preprocessing on the multi-source raw data and calculates core diagnostic features using a sliding window; a stress relaxation diagnosis unit that calculates the terminal's health status index based on a stress relaxation assessment model and generates graded early warning signals; and a predictive analysis unit that uses a multivariate long short-term memory network to obtain predicted values ​​of the health status index and uses linear extrapolation and interpolation algorithms to calculate the terminal's remaining service life. Through multi-source data fusion and core feature extraction, it achieves accurate monitoring of the terminal's condition from the source of mechanical stress relaxation, solving the problem that traditional monitoring methods only focus on electrical parameters and cannot predict the root causes of stress relaxation.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, and more specifically, to a stress relaxation monitoring system for cable terminals in power distribution cabinets. Background Technology

[0002] Cable terminals are critical nodes for current transmission within power distribution cabinets, and their connection reliability directly determines the safe and stable operation of the power system. During long-term operation of power distribution cabinets, the difference in thermal expansion coefficients between the terminals and bolts due to temperature cycles generates mechanical stress. Over time, this leads to bolt loosening, resulting in an exponential increase in contact resistance.

[0003] Existing power distribution cabinet monitoring systems mostly focus on detecting electrical parameters such as temperature and current. Although they can provide alarms for temperature rise at the joints, they cannot capture the root cause of stress relaxation. Therefore, this invention designs a stress relaxation monitoring system for cable terminals in power distribution cabinets. Summary of the Invention

[0004] The purpose of this invention is to provide a stress relaxation monitoring system for cable terminals in power distribution cabinets, in order to solve the problem mentioned in the background art that existing power distribution cabinet monitoring systems mostly focus on detecting electrical parameters such as temperature and current. Although they can realize joint temperature rise alarm, they cannot capture the fundamental cause of stress relaxation from the source.

[0005] To achieve the above objectives, the present invention aims to provide a stress relaxation monitoring system for cable terminals in a power distribution cabinet, comprising a multi-source data acquisition unit. The multi-source data acquisition unit acquires surface operating temperature data, real-time micro-strain data, and load current data of the terminals, and also acquires ambient temperature data inside the cabinet, which together constitute multi-source raw data.

[0006] The feature extraction unit performs time-series alignment and normalization preprocessing on the multi-source raw data to obtain standardized data, and calculates core diagnostic features through a sliding window.

[0007] The stress relaxation diagnostic unit uses standardized data and core diagnostic features as dynamic features, and introduces the material properties and structural geometric parameters of the terminal as static features. It constructs and trains a stress relaxation assessment model through an improved gradient boosting tree regression algorithm, which is used to calculate the health status index of the terminal based on the fused dynamic and static features, and continuously updates the historical health status sequence. At the same time, it generates a graded early warning signal when the health status index exceeds a preset threshold.

[0008] The predictive analysis unit, based on historical health status sequences and dynamic characteristics strongly correlated with stress state, uses a multivariate long short-term memory network to obtain predicted health status index values, and employs linear extrapolation and interpolation algorithms to calculate the remaining service life of the terminals.

[0009] As a further improvement to this technical solution, the multi-source data acquisition unit includes multi-source raw data such as terminal surface operating temperature data, real-time micro-strain data, load current data, and ambient temperature data.

[0010] Among them, the surface operating temperature data is a sequence of instantaneous values ​​of the terminal surface temperature;

[0011] Real-time micro-strain data is a sequence of instantaneous strain values ​​at the terminal fastening points;

[0012] The load current data is a sequence of the effective values ​​of the current in the cable loop;

[0013] The ambient temperature data is a sequence of instantaneous ambient temperature values.

[0014] As a further improvement to this technical solution, the feature extraction unit includes a data preprocessing module and a feature calculation module;

[0015] The data preprocessing module aligns the time series using linear interpolation and standardizes the multi-source raw data using a min-max normalization algorithm to obtain standardized data; the standardized data includes standardized surface temperature data. Standardized strain data Standardized current data and standardized ambient temperature data ; For time indexing;

[0016] Based on the standardized data, the feature calculation module uses a sliding window to extract core diagnostic features representing state changes; these core diagnostic features include the strain baseline sequence. Temperature-strain coefficient sequence Current-temperature correlation sequence and temperature-induced strain separation sequence .

[0017] As a further improvement to this technical solution, the specific steps involved in extracting the core diagnostic features representing state changes using a sliding window in the feature calculation module are as follows:

[0018] Based on standardized data, the window length is... The sliding window moves in preset steps. Swipe forward to extract the standardized surface temperature sequence. Standardized strain sequence Standardized current sequence and standardized ambient temperature series ;in, For the index of sampling points within the window, , This represents the total number of sampling points within the window.

[0019] Within each window, the mean of the normalized strain sequence is calculated as the strain baseline for that window. ;

[0020] Within each window, the temperature-strain coefficient is calculated by fitting the relationship between temperature and strain using linear regression. ;

[0021] Within each window, the Pearson correlation coefficient between the load current and the surface operating temperature is calculated to obtain the current-temperature correlation. ;

[0022] Within each window, the net temperature rise of the terminals is calculated based on the normalized surface temperature sequence and the normalized ambient temperature sequence; this is achieved by analyzing the strain baseline. In, subtracting the temperature-strain coefficient The thermo-induced elastic strain component, characterized by the product of the product with the net temperature rise of the terminal, yields the thermo-induced strain separation value. ;

[0023] Strain baseline calculated based on each window Temperature-strain coefficient Current-temperature correlation Temperature-induced strain separation value The core diagnostic features are obtained by continuously sliding a sliding window at a preset step size consistent with the time interval of the standardized data; among them, the core diagnostic features include the strain baseline sequence. Temperature-strain coefficient sequence Current-temperature correlation sequence Temperature-induced strain separation value sequence .

[0024] As a further improvement to this technical solution, the stress relaxation diagnostic unit includes a health status assessment module and an adaptive early warning module;

[0025] Among them, the health status assessment module constructs and trains a stress relaxation assessment model through an improved gradient boosting tree regression algorithm, which is used to calculate the health status index of the terminal in real time based on the fused dynamic and static features, and continuously update the historical health status sequence.

[0026] The adaptive early warning module uses a multi-level threshold comparison and state machine algorithm to compare the health status index with a preset threshold and generate a graded early warning signal when the health status index exceeds the limit.

[0027] As a further improvement to this technical solution, the specific steps involved in constructing and training the stress relaxation assessment model based on the improved gradient boosting tree regression algorithm in the health status assessment module are as follows:

[0028] S3.1 Collect historical operating data and static parameters of terminals, combine historical standardized data and historical core diagnostic features into a dynamic feature sequence according to time sequence, and combine the static parameters of terminals into a static feature group; wherein, static parameters include material creep coefficient, elastic modulus, and terminal thickness;

[0029] S3.2, for each length of The dynamic feature sequences within the aggregation window are calculated using the arithmetic mean algorithm to calculate the sequence mean, the standard deviation algorithm to calculate the sequence standard deviation, and the first-order linear regression algorithm to calculate the sequence slope.

[0030] S3.3 For each aggregation window, the mean, standard deviation, and slope of the sequences within that window are combined into a dynamic feature group. The dynamic feature group and the static feature group together constitute the complete input feature vector. ;

[0031] S3.4, Based on input feature vector Manually annotate the true value of the terminal health status index at the corresponding time. , constitute the training sample set ;in, For sample index, The number of samples;

[0032] S3.5 Calculate the mean of the health status index of all samples as the initial evaluation value of the stress relaxation assessment model;

[0033] S3.6 Introduce a static feature-first splitting strategy to construct the first decision tree of the stress relaxation evaluation model, and assign a splitting gain bias term to each static feature; wherein, when splitting a node, for candidate static features, the gain with the bias term is superimposed and used for splitting quality evaluation.

[0034] S3.7 Enter the iterative loop. For each sample, calculate the difference between the current stress relaxation assessment model's evaluation value and the true health status index to obtain the pseudo residual. ;in, For the iteration round index, This represents the number of iterations.

[0035] S3.8. Using the current pseudo residual as the target value, train a new decision tree using the feature vectors of the samples, and divide the feature vectors into... Non-overlapping intervals ;in, For range index, The total number of intervals;

[0036] S3.9 For each interval, calculate the average of all sample spurious residuals within that interval, and use it as the error compensation value for that interval by the decision tree;

[0037] S3.10. The error compensation value is superimposed on the previous stress relaxation assessment model according to a certain learning rate to form an updated stress relaxation assessment model.

[0038] S3.11, Repeat steps S3.7~S3.10, after... After multiple iterations, the final stress relaxation evaluation model is obtained. ;

[0039] S3.12. Throughout the model training process, the weighted mean square error is used as the loss function; where the sample weights are negatively correlated with the health status index.

[0040] As a further improvement to this technical solution, the specific steps involved in the real-time calculation of the terminal's health status index in the health status assessment module are as follows:

[0041] Obtain standardized data, core diagnostic features, and static parameters of the terminals at the current moment to construct a real-time input feature vector. ;

[0042] Real-time input feature vector Input into the trained stress relaxation evaluation model The output shows the current health status index. ;

[0043] By time interval Continuously update the health status index to obtain the historical health status sequence. .

[0044] As a further improvement to this technical solution, the specific steps involved in generating a graded early warning signal when the health status index exceeds the limit in the adaptive early warning module are as follows:

[0045] Preset tiered warning thresholds, including a yellow warning threshold. Orange alert threshold and red alert threshold ;

[0046] Current health status index Compare with the tiered early warning threshold;

[0047] when Greater than or equal to When the time is right, it is determined to be a normal state and no warning is triggered;

[0048] when Less than But greater than or equal to When this occurs, a yellow warning is triggered, indicating that the terminal has experienced slight stress relaxation;

[0049] when Less than But greater than or equal to When this occurs, an orange alert is triggered, indicating that moderate stress relaxation has occurred at the terminal.

[0050] when Less than When this occurs, a red alert is triggered, indicating that the terminal is in a state of severe stress relaxation;

[0051] When a yellow alert is issued, the system will issue a warning and suggest that the system should pay special attention to the area during the next inspection.

[0052] When an orange alert is issued, the system will send a maintenance reminder and recommend that maintenance be scheduled in the near future.

[0053] When a red alert is issued, the system will issue an emergency alarm, requiring immediate action to be taken.

[0054] As a further improvement to this technical solution, the specific steps involved in obtaining the predicted value of the health status index using a multivariate long short-term memory network in the predictive analysis unit are as follows:

[0055] Dynamic features strongly correlated with stress state were selected as covariates and compared with historical health state sequences. Together they constitute a multivariate input sequence; among which, covariates include the strain baseline sequence. Temperature-strain coefficient sequence Temperature-induced strain separation sequence Construct a multivariate long short-term memory network, with multivariate input sequences as inputs and health status index prediction sequences as outputs. , To predict the step size; the multivariate long short-term memory network consists of one input layer, two hidden layers, and one fully connected output layer. The hidden layers use the tanh activation function, and the output layer uses the linear activation function.

[0056] As a further improvement to this technical solution, the predictive analysis unit employs linear extrapolation and interpolation algorithms to calculate the remaining service life of the terminals. The specific steps involved are as follows:

[0057] Predicted sequences from health status index Select the most recent one. The predicted values ​​at each time point constitute a recent health status sequence. ,in, , To select the number of predicted values;

[0058] The least squares method was used to perform linear fitting on the recent health status sequence to obtain a trend line of health status changes over time. ;

[0059] Based on the trend line obtained by fitting Extrapolated to the health status index equals the preset failure threshold. The time point is used to obtain the predicted failure time. :

[0060] Calculate the current time To the predicted failure time The time difference is used to determine the remaining service life. .

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] 1. This stress relaxation monitoring system for cable terminals in power distribution cabinets synchronously collects data on terminal surface operating temperature, real-time micro-strain, load current, and ambient temperature. Combined with core diagnostic features such as strain baseline calculation using a sliding window, temperature-strain coefficient, and current-temperature correlation, it achieves precise monitoring of terminal condition from the source of mechanical stress relaxation. This solves the problem that traditional monitoring methods only focus on electrical parameters and cannot predict the root cause of stress relaxation.

[0063] 2. In this stress relaxation monitoring system for cable terminals in a power distribution cabinet, a stress relaxation assessment model is constructed and trained based on an improved gradient boosting tree regression algorithm to calculate the health status index in real time. A multivariate long short-term memory network is used to predict the health status trend, and a linear extrapolation algorithm is combined to calculate the remaining service life, thus realizing full life cycle health management from real-time diagnosis to long-term prediction. Attached Figure Description

[0064] Figure 1 This is an overall flowchart of the present invention;

[0065] The meanings of the labels in the diagram are as follows:

[0066] 1. Multi-source data acquisition unit; 2. Feature extraction unit; 21. Data preprocessing module; 22. Feature calculation module; 3. Stress relaxation diagnosis unit; 31. Health status assessment module; 32. Adaptive early warning module; 4. Predictive analysis unit. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example: Please refer to Figure 1 As shown, a stress relaxation monitoring system for cable terminals in a power cabinet is provided, including a multi-source data acquisition unit 1. The multi-source data acquisition unit 1 acquires surface operating temperature data, real-time micro-strain data and load current data of the terminals, and acquires ambient temperature data inside the cabinet, which together constitute multi-source raw data.

[0069] In this embodiment, the multi-source raw data includes the terminal surface operating temperature data, real-time micro-strain data, and load current data;

[0070] Among them, the surface operating temperature data is a sequence of instantaneous values ​​of the terminal surface temperature, which is collected by thermocouples attached to the terminal surface;

[0071] Real-time micro-strain data is a sequence of instantaneous strain values ​​at the terminal fastening points, acquired by resistance strain gauges attached to the terminal fastening bolts or crimping structures.

[0072] The load current data is a sequence of effective current values ​​in the cable circuit, which is collected by a current transformer connected in series in the cable circuit. It serves as a system load reference and is used for normalization and correlation diagnosis.

[0073] The ambient temperature data is a sequence of instantaneous ambient temperature values, collected by an ambient temperature sensor installed at a specific location within the power cabinet, and used to provide a temperature rise reference for the environment where the terminals are located.

[0074] The cable terminal stress relaxation monitoring system for power cabinets also includes a feature extraction unit 2. The feature extraction unit 2 performs time-series alignment and normalization preprocessing on multi-source raw data to obtain standardized data, and calculates core diagnostic features through a sliding window.

[0075] In this embodiment, the feature extraction unit 2 includes a data preprocessing module 21 and a feature calculation module 22;

[0076] The data preprocessing module 21 aligns the time series using linear interpolation and standardizes the multi-source raw data using a minimum-maximum normalization algorithm to obtain standardized data; the standardized data includes standardized surface temperature data. Standardized strain data Standardized current data and standardized ambient temperature data ; For time indexing;

[0077] Based on the standardized data, feature calculation module 22 uses a sliding window to extract core diagnostic features representing state changes; among which, the core diagnostic features include strain baseline sequences. Temperature-strain coefficient sequence Current-temperature correlation sequence and temperature-induced strain separation sequence ;

[0078] Specifically, the steps involved in extracting core diagnostic features representing state changes using a sliding window are as follows:

[0079] Based on standardized data, the window length is... The sliding window moves in preset steps. Swipe forward to extract the standardized surface temperature sequence. Standardized strain sequence Standardized current sequence and standardized ambient temperature series ;in, For the index of sampling points within the window, , This represents the total number of sampling points within the window.

[0080] Among them, window length seconds; sliding step size The time interval between seconds and standardized data Consistent;

[0081] Within each window, the mean of the normalized strain sequence is calculated as the strain baseline for that window. ;

[0082] Calculate the strain baseline :

[0083] ;

[0084] in, The arithmetic mean of the standardized strain within a single window reflects its static tightness and is used to determine whether there is a persistent loosening trend. For the first Standardized strain values ​​at each sampling point;

[0085] Within each window, the temperature-strain coefficient is calculated by fitting the relationship between temperature and strain using linear regression. ;

[0086] Calculate the temperature-strain coefficient :

[0087] ;

[0088] in, It represents the slope of the temperature-strain linear regression, i.e., the amount of strain change caused by a unit temperature change, and is used to diagnose material expansion or loosening of connections caused by temperature rise. An increase in absolute value usually indicates that the terminal is more sensitive to temperature, suggesting that stress relaxation is intensifying; For the first Standardized surface operating temperature values ​​at each sampling point;

[0089] Within each window, the Pearson correlation coefficient between the load current and the surface operating temperature is calculated to obtain the current-temperature correlation. ;

[0090] Calculate the current-temperature correlation :

[0091] ;

[0092] in, This represents the average current within the window. This represents the average temperature within the window. This reflects the degree of influence of load current changes on terminal temperature rise, with a value range of [−1, 1]. During the operation of the power distribution cabinet... An abnormally low value may indicate an increase in contact resistance, leading to a disconnect between heating and current, which is an indirect characteristic of stress relaxation. For the first in the window Standardized current values ​​at each sampling point;

[0093] Within each window, the net temperature rise of the terminals is calculated based on the normalized surface temperature sequence and the normalized ambient temperature sequence; this is achieved by analyzing the strain baseline monitored in real time. In, subtracting the temperature-strain coefficient The thermo-induced elastic strain component, characterized by the product of the product with the net temperature rise of the terminal, yields the thermo-induced strain separation value. ;

[0094] ;

[0095] in, For the net temperature rise of the terminals, For thermo-elastic strain components; This is the temperature-induced strain separation value, designed to deduct the elastic strain caused by the current temperature rise. Specifically, a decrease in clamping force leads to a reduction in terminal contact pressure, resulting in plastic deformation under temperature cycling, manifested as... Increase; It is the window mean of the surface temperature sequence. It is the window mean of the ambient temperature series;

[0096] Strain baseline calculated based on each window Temperature-strain coefficient Current-temperature correlation Temperature-induced strain separation value The core diagnostic features are obtained by continuously sliding a sliding window at a preset step size consistent with the time interval of the standardized data; among them, the core diagnostic features include the strain baseline sequence. Temperature-strain coefficient sequence Current-temperature correlation sequence Temperature-induced strain separation value sequence.

[0097] The stress relaxation monitoring system for cable terminals in a power cabinet also includes a stress relaxation diagnosis unit 3. The stress relaxation diagnosis unit 3 constructs and trains a stress relaxation assessment model based on an improved gradient boosting tree regression algorithm. It is used to calculate the health status index of the terminal based on standardized data and core diagnostic features, and continuously update the historical health status sequence. At the same time, it generates a graded early warning signal when the health status index exceeds a preset threshold.

[0098] In this embodiment, the stress relaxation diagnosis unit 3 includes a health status assessment module 31 and an adaptive early warning module 32;

[0099] Among them, the health status assessment module 31 constructs and trains a stress relaxation assessment model through an improved gradient boosting tree regression algorithm, which is used to calculate the health status index of the terminal in real time based on the fused dynamic and static features, and continuously update the historical health status sequence.

[0100] Specifically, the steps involved in constructing and training the stress relaxation assessment model based on the improved gradient boosting tree regression algorithm are as follows:

[0101] S3.1 Collect historical operating data and static parameters of terminals, and combine the historical standardized data and historical core diagnostic features into a dynamic feature sequence according to time sequence. The static parameters of the terminals are combined into a static feature group; among which, the static parameters include the material creep coefficient (i.e., the core parameter of the material property), the elastic modulus (i.e., the stiffness parameter of the material property), and the terminal thickness (i.e., a typical representative of the structural geometric parameters).

[0102] Dynamic feature sequences:

[0103] ;

[0104] in, It is a dynamic feature sequence;

[0105] Static feature group:

[0106] ;

[0107] in, It is a static feature group; The creep coefficient (unit: 1 / Pa / s) determines the inherent tendency of a material to undergo plastic deformation (i.e., relaxation) over time under temperature and stress, and is a core parameter. The elastic modulus (in GPa) determines the stiffness of a material against elastic deformation. Terminals with a high elastic modulus have less strain under the same clamping force, but are more sensitive to relaxation. The thickness of the terminal (in mm) directly affects the bending stiffness and heat capacity of the structure. Thicker terminals have greater stiffness, greater thermal inertia, and slower temperature rise and cooling.

[0108] S3.2, for each length of (Aggregation window length) It can be set according to the monitoring granularity requirements, for example, it can be set as the sliding window length in the feature extraction unit. Dynamic feature sequences within an aggregation window (integer multiples of) The arithmetic mean algorithm is used to calculate the mean of the sequence. The standard deviation of a sequence is calculated using the standard deviation algorithm. The slope of the sequence is calculated using a first-order linear regression algorithm. ;

[0109] Calculate the series mean:

[0110] ;

[0111] in, The length of the aggregation window; The dynamic feature sequence within the aggregation window; The mean of the sequence is used to characterize the average state level of the feature within the window;

[0112] Calculate the standard deviation of the sequence:

[0113] ;

[0114] in, The standard deviation of the sequence is used to quantify the degree of fluctuation of this feature within the window;

[0115] Calculating the sequence slope: A first-order linear regression algorithm is used, indexing the time points... As the independent variable, the sequence value As the dependent variable, the regression coefficients are obtained by fitting a straight line using the least squares method. That is, the slope of the sequence;

[0116] ;

[0117] in, The slope of the sequence is used to characterize the trend and rate of change of this feature within the window; For bias terms;

[0118] S3.3 For each aggregation window, calculate the mean of the sequences within that window. , sequence standard deviation sequence slope The features are combined into a dynamic feature set, and the dynamic feature set and the static feature set together constitute a complete input feature vector. ;

[0119] ;

[0120] in, This is the input feature vector for the stress relaxation assessment model;

[0121] S3.4, Based on input feature vector Manually annotate the true value of the terminal health status index at the corresponding time. , constitute the training sample set ;in, For sample index, The number of samples;

[0122] in, For the first Feature vectors of each sample; For the first The health status index of each sample is determined by manual inspection, with a value range of [0,1], where 1 indicates complete health and 0 indicates complete failure due to stress relaxation.

[0123] S3.5 Calculate the mean of the health status index of all samples as the initial evaluation value of the stress relaxation assessment model;

[0124] Initial assessment value:

[0125] ;

[0126] in, The loss function is the mean squared error loss, i.e. ; The initial evaluation value obtained can minimize the mean squared error loss between all sample evaluation values ​​and the true value; These are the initial evaluation parameters;

[0127] S3.6 To accurately distinguish the degradation characteristics of different terminal types and improve the generalization ability of the stress relaxation assessment model, a static feature-priority splitting strategy is introduced (i.e., for each static feature...). Assign a split gain bias term Construct the first decision tree for the stress relaxation assessment model; where, during node splitting, a bias term is superimposed on the candidate static features. The gain is used for split quality assessment;

[0128] This bias term satisfies:

[0129] ;

[0130] in, This is the initial estimate of the average splitting gain of the dynamic feature set. A bias coefficient greater than 1 (e.g.) =2); This operation aims to prioritize the construction of the evaluation backbone framework based on physical properties in the early stages of model building; The first in the static feature group The split gain bias term of a static feature ;

[0131] ;

[0132] in, The first in the static feature group One static feature; The original information gain is obtained by traversing all possible split points of the static features and calculating the information gain that each split point can bring based on the current node sample set. This is the gain after biasing;

[0133] S3.7 Enter the iterative loop, and for each sample, calculate the evaluation value of the current stress relaxation assessment model. Compared with the true health status index The difference is used to obtain the pseudo residual. ;in, For the iteration round index, This represents the number of iterations.

[0134] Calculate pseudo residuals:

[0135] ;

[0136] in, For the first In the first iteration The pseudo residuals of each sample For the first After rounds of iterations, the model evaluates the samples. The evaluation value;

[0137] S3.8. Using the current pseudo residual as the target value, train a new decision tree using the feature vectors of the samples, and divide the feature vectors into... Non-overlapping intervals ;in, For range index, The total number of intervals;

[0138] For the first In the first iteration of the regression tree, the first partition... There are 1 feature intervals, each interval corresponding to a set of feature value ranges (e.g., ... and The pseudo residuals of samples within the same interval are similar, which facilitates targeted optimization.

[0139] S3.9 For each interval, calculate the average of all sample spurious residuals within that interval, and use it as the error compensation value for that interval by the decision tree;

[0140] Calculate the error compensation value for each interval:

[0141] ;

[0142] in, For the first In the first iteration Error compensation values ​​for each interval; To find the mean function;

[0143] S3.10. The error compensation value is superimposed on the previous stress relaxation assessment model according to a certain learning rate to form an updated stress relaxation assessment model.

[0144] Update the model:

[0145] ;

[0146] in, The learning rate (usually a value) To control the contribution intensity of each decision tree and prevent a single decision tree from having too large an influence ( The smaller the value, the more robust the model, but the more iterations are required. This is an indicator function used to identify which region a sample belongs to; For the first Model evaluation value after rounds of iteration;

[0147] S3.11, Repeat steps S3.7~S3.10, after... After multiple iterations, the final stress relaxation evaluation model is obtained. ;

[0148] ;

[0149] in, This serves as the final stress relaxation assessment model;

[0150] S3.12. To address the issue that healthy samples far outnumber faulty samples in actual operational data, a weighted mean squared error is used as the loss function throughout the model training process; where the sample weights... With sample Health Status Index Negative correlation;

[0151] ;

[0152] ;

[0153] in, For sample weights; This is an imbalance adjustment factor (greater than 0), which makes samples with poor health status contribute more to model training, thereby improving the model's sensitivity to early failures. The number of samples; This is the loss function for the stress relaxation assessment model;

[0154] Finally, the stress relaxation assessment model was adjusted using cross-validation. The parameters are set to minimize the evaluation error on the validation set; where the stress relaxation evaluation model... The parameters include the maximum depth of the decision tree and the learning rate. and number of iterations Bias coefficient and imbalance regulators ;

[0155] Furthermore, the specific steps involved in real-time calculation of the terminal's health status index are as follows:

[0156] Obtain standardized data, core diagnostic features, and static parameters of the terminals at the current moment to construct a real-time input feature vector. ;

[0157] Real-time input feature vector Input into the trained stress relaxation evaluation model The output shows the current health status index. ;

[0158] ;

[0159] in, The health status index at the current moment is a real number between [0,1]. 1 indicates that the terminal is in a completely healthy state, and 0 indicates that it has completely failed due to stress relaxation. The health status index comprehensively reflects the terminal's fastening condition, temperature rise characteristics, and current-temperature coupling relationship, thereby realizing a quantitative assessment of the terminal's health status.

[0160] By time interval (Set to 1 second) Continuously update the health status index to obtain the historical health status sequence. ;

[0161] The adaptive early warning module 32 uses a multi-level threshold comparison and state machine algorithm to compare the health status index with a preset threshold and generate a graded early warning signal when the health status index exceeds the limit.

[0162] Specifically, the health status index is compared with a preset threshold through multi-level threshold comparison and state machine algorithm. The specific steps involved are as follows:

[0163] Preset tiered warning thresholds, including a yellow warning threshold. Orange alert threshold and red alert threshold ;

[0164] in, The yellow warning threshold is typically set to 0.8, indicating that the health status has begun to deteriorate. The orange alert threshold is typically set to 0.6, indicating a significant decline in health status. The red warning threshold is typically set to 0.4, indicating a severe deterioration in the health status. These thresholds are determined based on the safety operation standards of the power cabinet terminals and historical fault data.

[0165] Current health status index Compare with the tiered early warning threshold;

[0166] when Greater than or equal to When the time is right, it is determined to be a normal state and no warning is triggered;

[0167] when Less than But greater than or equal to When this occurs, a yellow warning is triggered, indicating that the terminal has experienced slight stress relaxation;

[0168] when Less than But greater than or equal to When this occurs, an orange alert is triggered, indicating that moderate stress relaxation has occurred at the terminal.

[0169] when Less than When this occurs, a red alert is triggered, indicating that the terminal is in a state of severe stress relaxation;

[0170] To avoid frequent alert switching due to data fluctuations near the threshold, the system adopts a state maintenance mechanism; it will only upgrade to a higher alert level when the health status index is continuously below a certain threshold for 5 consecutive sampling periods; and it will only downgrade to a lower alert level or return to normal state when the health status index is continuously above a certain threshold for 10 consecutive sampling periods.

[0171] When a yellow alert is issued, the system will issue a warning and suggest that the system should pay special attention to the area during the next inspection.

[0172] When an orange alert is issued, the system will send a maintenance reminder and recommend that maintenance be scheduled in the near future.

[0173] When a red alert is issued, the system will issue an emergency alarm, requiring immediate action to be taken.

[0174] The stress relaxation monitoring system for cable terminals in a power cabinet also includes a predictive analysis unit 4. The predictive analysis unit 4 uses a multivariate long short-term memory network to obtain the predicted value of the health status index based on the historical health status sequence and dynamic characteristics strongly correlated with the stress status. It then uses a linear extrapolation and interpolation algorithm to calculate the remaining service life of the terminal.

[0175] In this embodiment, the specific steps involved in obtaining the predicted value of the health status index using a multivariate long short-term memory network are as follows:

[0176] Dynamic features strongly correlated with stress state were selected as covariates and compared with historical health state sequences. Together they constitute a multivariate input sequence; among which, covariates include the strain baseline sequence. Temperature-strain coefficient sequence Temperature-induced strain separation sequence ;

[0177] Among them, the strain baseline, temperature-strain coefficient, and temperature-induced strain separation value directly reflect the stress state of the terminal and are strongly correlated with the health status index, and are therefore used as covariates.

[0178] The constructed multivariate input sequence is Z-score standardized so that the mean of each dimension is 0 and the standard deviation is 1. Then, the sequence is divided into training set, validation set and test set in chronological order (e.g., in a 7:2:1 ratio).

[0179] Construct a multivariate long short-term memory network to take multivariate input sequences as input and output predicted values ​​of health status index;

[0180] The multivariate long short-term memory network consists of one input layer (the number of neurons in the input layer is the same as the dimension of the multivariate input sequence), two hidden layers (64 neurons in each layer) and one fully connected output layer. The hidden layers use the tanh activation function, and the output layer uses the linear activation function.

[0181] The input layer receives a multivariate input sequence as input, and the output layer outputs a predicted sequence of health status index for a future period. , To predict the step size;

[0182] Set the hyperparameters of the multivariable long short-term memory network, including the time window length. The number of hidden layer neurons, learning rate, batch size, and number of training rounds;

[0183] The network is trained using the training set, and the prediction error is minimized through backpropagation and the Adam optimizer. The loss function is the mean squared error.

[0184] ;

[0185] in, The number of training samples, A true health status index. The predicted value output by the LSTM network; This is the mean square error loss;

[0186] The predictive performance of the multivariate long short-term memory network was evaluated on the validation set, and the hyperparameters were tuned to optimize the prediction accuracy and prevent overfitting.

[0187] A trained multivariate long short-term memory network is used to perform rolling predictions of future health status indices, resulting in a health status index prediction sequence. , To predict the step size;

[0188] Furthermore, the specific steps involved in calculating the remaining service life of the terminals using linear extrapolation and interpolation algorithms are as follows:

[0189] Predicted sequences from health status index Select the most recent one. The predicted values ​​at each time point constitute a recent health status sequence. ,in, , To select the number of predicted values;

[0190] The least squares method was used to perform linear fitting on the recent health status sequence to obtain a trend line of health status changes over time:

[0191] ;

[0192] in, The slope of the fitted graph reflects the rate of decline in health status. The intercept of the fit;

[0193] Based on the trend line obtained by fitting Extrapolated to the health status index equals the preset failure threshold. The time point is used to obtain the predicted failure time. :

[0194] ;

[0195] in, The failure threshold for the health status is typically set at 0.2, determined based on the safety operation standards of the power cabinet terminals and historical fault data. When the health status index falls below the failure threshold... If the value is not met, the terminal is considered to be faulty and needs to be replaced or repaired. To predict the failure time;

[0196] Calculate the current time To the predicted failure time The time difference is used to determine the remaining service life. ;

[0197] ;

[0198] in, The remaining service life is expressed in seconds, which can be converted to minutes, hours, or days as needed. The current moment;

[0199] like This indicates that the health status has not shown a downward trend, and the remaining lifespan is set to a maximum value.

[0200] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A stress relaxation monitoring system for cable terminals in a power distribution cabinet, characterized in that: include Multi-source data acquisition unit (1) acquires surface operating temperature data, real-time micro-strain data and load current data of the terminal, and acquires ambient temperature data inside the cabinet, which together constitute multi-source raw data; Feature extraction unit (2) performs time-series alignment and normalization preprocessing on multi-source raw data to obtain standardized data, and calculates core diagnostic features through a sliding window; The stress relaxation diagnostic unit (3) uses standardized data and core diagnostic features as dynamic features, and introduces the material properties and structural geometric parameters of the terminals as static features. An improved gradient boosting tree regression algorithm is used to construct and train a stress relaxation assessment model, which is used to calculate the health status index of the terminal based on the fused dynamic and static features, and continuously update the historical health status sequence. At the same time, a graded early warning signal is generated when the health status index exceeds a preset threshold. The prediction analysis unit (4) uses a multivariate long short-term memory network to obtain the predicted value of the health status index based on the historical health status sequence and dynamic characteristics strongly correlated with the stress state, and uses linear extrapolation and interpolation algorithms to calculate the remaining service life of the terminal. In the multi-source data acquisition unit (1), the multi-source raw data includes the surface operating temperature data of the terminal, real-time micro-strain data, load current data, and ambient temperature data; Among them, the surface operating temperature data is a sequence of instantaneous values ​​of the terminal surface temperature; Real-time micro-strain data is a sequence of instantaneous strain values ​​at the terminal fastening points; The load current data is a sequence of the effective values ​​of the current in the cable loop; The ambient temperature data is a sequence of instantaneous ambient temperature values; The feature extraction unit (2) includes a data preprocessing module (21) and a feature calculation module (22). The data preprocessing module (21) aligns the time series using linear interpolation and standardizes the multi-source raw data using a minimum-maximum normalization algorithm to obtain standardized data; the standardized data includes standardized surface temperature data. Standardized strain data Standardized current data and standardized ambient temperature data ; For time indexing; The feature calculation module (22) extracts core diagnostic features representing state changes using a sliding window based on the standardized data; wherein, the core diagnostic features include the strain baseline sequence. Temperature-strain coefficient sequence Current-temperature correlation sequence and temperature-induced strain separation sequence ; In the feature calculation module (22), the specific steps involved in extracting the core diagnostic features representing state changes using a sliding window are as follows: Based on standardized data, the window length is... The sliding window moves in preset steps. Swipe forward to extract the standardized surface temperature sequence. Standardized strain sequence Standardized current sequence and standardized ambient temperature series ;in, For the index of sampling points within the window, , This represents the total number of sampling points within the window. Within each window, the mean of the normalized strain sequence is calculated as the strain baseline for that window. ; Within each window, the temperature-strain coefficient is calculated by fitting the relationship between temperature and strain using linear regression. ; Within each window, the Pearson correlation coefficient between the load current and the surface operating temperature is calculated to obtain the current-temperature correlation. ; Within each window, the net temperature rise of the terminals is calculated based on the normalized surface temperature sequence and the normalized ambient temperature sequence; this is achieved by analyzing the strain baseline. In, subtracting the temperature-strain coefficient The thermo-induced elastic strain component, characterized by the product of the product with the net temperature rise of the terminal, yields the thermo-induced strain separation value. ; Strain baseline calculated based on each window Temperature-strain coefficient Current-temperature correlation Temperature-induced strain separation value The core diagnostic features are obtained by continuously sliding a sliding window at a preset step size consistent with the time interval of the standardized data; among them, the core diagnostic features include the strain baseline sequence. Temperature-strain coefficient sequence Current-temperature correlation sequence Temperature-induced strain separation value sequence ; The stress relaxation diagnostic unit (3) includes a health status assessment module (31) and an adaptive early warning module (32). Among them, the health status assessment module (31) constructs and trains the stress relaxation assessment model through the improved gradient boosting tree regression algorithm, which is used to calculate the health status index of the terminal in real time based on the fused dynamic and static features, and continuously update the historical health status sequence. The adaptive early warning module (32) uses a multi-level threshold comparison and state machine algorithm to compare the health status index with a preset threshold and generate a graded early warning signal when the health status index exceeds the limit. In the health status assessment module (31), the specific steps involved in constructing and training the stress relaxation assessment model based on the improved gradient boosting tree regression algorithm are as follows: S3.1 Collect historical operating data and static parameters of terminals, combine historical standardized data and historical core diagnostic features into a dynamic feature sequence according to time sequence, and combine the static parameters of terminals into a static feature group; wherein, static parameters include material creep coefficient, elastic modulus, and terminal thickness; S3.2, for each length of The dynamic feature sequences within the aggregation window are calculated using the arithmetic mean algorithm to calculate the sequence mean, the standard deviation algorithm to calculate the sequence standard deviation, and the first-order linear regression algorithm to calculate the sequence slope. S3.3 For each aggregation window, the mean, standard deviation, and slope of the sequences within that window are combined into a dynamic feature group. The dynamic feature group and the static feature group together constitute the complete input feature vector. ; S3.4, Based on input feature vector Manually annotate the true value of the terminal health status index at the corresponding time. , constitute the training sample set ;in, For sample index, The number of samples; S3.5 Calculate the mean of the health status index of all samples as the initial evaluation value of the stress relaxation assessment model; S3.6 Introduce a static feature-first splitting strategy to construct the first decision tree of the stress relaxation evaluation model, and assign a splitting gain bias term to each static feature; wherein, when splitting a node, for candidate static features, the gain with the bias term is superimposed and used for splitting quality evaluation. S3.7 Enter the iterative loop. For each sample, calculate the difference between the current stress relaxation assessment model's evaluation value and the true health status index to obtain the pseudo residual. ;in, For the iteration round index, This represents the number of iterations. S3.

8. Using the current pseudo residual as the target value, train a new decision tree using the feature vectors of the samples, and divide the feature vectors into... Non-overlapping intervals ;in, For range index, The total number of intervals; S3.9 For each interval, calculate the average of all sample spurious residuals within that interval, and use it as the error compensation value for that interval by the decision tree; S3.

10. The error compensation value is superimposed on the previous stress relaxation assessment model according to a certain learning rate to form an updated stress relaxation assessment model. S3.11, Repeat steps S3.7~S3.10, after... After multiple iterations, the final stress relaxation evaluation model is obtained. ; S3.

12. Throughout the model training process, the weighted mean squared error is used as the loss function; where the sample weights are negatively correlated with the health status index. In the health status assessment module (31), the specific steps involved in the real-time calculation of the terminal's health status index are as follows: Obtain standardized data, core diagnostic features, and static parameters of the terminals at the current moment to construct a real-time input feature vector. ; Real-time input feature vector Input into the trained stress relaxation evaluation model The output shows the current health status index. ; By time interval Continuously update the health status index to obtain the historical health status sequence. ; In the predictive analysis unit (4), the specific steps involved in obtaining the predicted value of the health status index using a multivariate long short-term memory network are as follows: Dynamic features strongly correlated with stress state were selected as covariates and compared with historical health state sequences. Together they constitute a multivariate input sequence; among which, covariates include the strain baseline sequence. Temperature-strain coefficient sequence Temperature-induced strain separation sequence ; Construct a multivariate long short-term memory network, with multivariate input sequences as input and health status index prediction sequences as output. , To predict the step size; the multivariate long short-term memory network includes one input layer, two hidden layers, and one fully connected output layer. The hidden layers use the tanh activation function, and the output layer uses the linear activation function. In the predictive analysis unit (4), the specific steps involved in calculating the remaining service life of the terminal using linear extrapolation and interpolation algorithms are as follows: Predicted sequences from health status index Select the most recent one. The predicted values ​​at each time point constitute a recent health status sequence. ,in, , To select the number of predicted values; The least squares method was used to perform linear fitting on the recent health status sequence to obtain a trend line of health status changes over time. ; Based on the trend line obtained by fitting Extrapolated to the health status index equals the preset failure threshold. The time point is used to obtain the predicted failure time. : Calculate the current time To the predicted failure time The time difference is used to determine the remaining service life. .

2. The stress relaxation monitoring system for cable terminals in a power distribution cabinet according to claim 1, characterized in that: The specific steps involved in generating a graded early warning signal when the health status index exceeds the limit in the adaptive early warning module (32) are as follows: Preset tiered warning thresholds, including a yellow warning threshold. Orange alert threshold and red alert threshold ; Current health status index Compare with the tiered early warning threshold; when Greater than or equal to When the time is right, it is determined to be a normal state and no warning is triggered; when Less than But greater than or equal to When this occurs, a yellow warning is triggered, indicating that the terminal has experienced slight stress relaxation; when Less than But greater than or equal to When this occurs, an orange alert is triggered, indicating that moderate stress relaxation has occurred at the terminal. when Less than When this occurs, a red alert is triggered, indicating that the terminal is in a state of severe stress relaxation; When a yellow alert is issued, the system will issue a warning and suggest that the system should pay special attention to the area during the next inspection. When an orange alert is issued, the system will send a maintenance reminder and recommend that maintenance be scheduled in the near future. When a red alert is issued, the system will issue an emergency alarm, requiring immediate action to be taken.

Images