Data-driven intelligent prediction method, system, and cable for cable life.

Abstract

This invention discloses a data-driven intelligent prediction method, system, and cable for cable life, relating to the field of intelligent operation and maintenance technology for power equipment. The method includes collecting multi-dimensional historical operating sequence data of the cable, such as current, voltage, partial discharge, temperature, and ambient temperature and humidity; generating a comprehensive cable state feature sequence through multi-source heterogeneous data fusion technology; extracting a degradation feature vector from this data, based on aging mechanisms, including accumulated load fatigue, insulation degradation index, and thermal stress damage degree; inputting this vector into a pre-trained model for time-series evolution simulation, outputting the probability distribution of remaining life and key failure time nodes; and generating a preventative maintenance decision-making scheme including maintenance windows, operation priorities, and inventory warnings. This method, through mechanism-driven and probabilistic prediction, achieves more accurate assessment and risk quantification of cable life, providing a direct decision-making basis for proactive operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology for power equipment, specifically a data-driven intelligent prediction method, system, and cable for cable life. Background Technology

[0002] Current cable life prediction methods mainly rely on threshold alarms and trend extrapolation of single or a few parameters. These methods typically monitor parameters such as current and temperature independently or perform simple regression analysis, failing to effectively integrate multidimensional data reflecting the multi-stress coupling effects of cables. Because the characteristics on which they are based have a weak correlation with the physicochemical mechanisms of cable aging, they have low sensitivity to identifying early and gradual degradation processes, resulting in rather coarse predictions.

[0003] At the data processing level, existing technologies typically use raw monitoring data directly or perform simple aggregation, lacking deep integration of multi-source heterogeneous time-series data such as current, voltage, partial discharge, temperature, and ambient humidity. Data from different sources differ in time scale and physical dimension, making it difficult to construct a comprehensive and coordinated state representation through separate processing methods. At the prediction level, conventional models mostly output deterministic lifetime point estimates, failing to quantify the uncertainty of predictions and struggling to identify key turning points in the aging process. This results in a lack of accurate risk time window information for operational decisions, and insufficient targeting and foresight in preventative maintenance.

[0004] The present invention aims to solve the above problems and seek an intelligent prediction method that can deeply integrate multi-source heterogeneous operating data to fully perceive the state and perform probabilistic life extrapolation based on the physical mechanism of cable aging. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art;

[0006] To this end, the present invention proposes a data-driven intelligent prediction method for cable life, comprising:

[0007] Collect a multidimensional historical operation dataset of the target cable, which includes current load time series data, voltage fluctuation time series data, partial discharge time series data, cable sheath temperature time series data, and ambient temperature and humidity time series data.

[0008] The multi-dimensional historical operation dataset is subjected to multi-source heterogeneous data fusion processing to generate a fused comprehensive cable status feature sequence;

[0009] Degradation feature vectors that characterize the aging process of the cable are extracted from the comprehensive state feature sequence of the cable. The degradation feature vectors include the cumulative load fatigue, insulation degradation index and thermal stress damage degree.

[0010] The degradation feature vector is input into a pre-trained cable life prediction model for time-series evolution simulation to generate the remaining life probability distribution and key failure time nodes of the target cable.

[0011] A preventive maintenance decision plan is generated based on the key failure time nodes. The preventive maintenance decision plan includes maintenance time window suggestions, maintenance operation priority ranking, and spare parts inventory early warning information.

[0012] Furthermore, the step of performing multi-source heterogeneous data fusion processing on the multi-dimensional historical operational dataset to generate a fused comprehensive cable status feature sequence includes:

[0013] Electrothermal coupling analysis is performed on the current load time series data and the voltage fluctuation time series data to generate the equivalent temperature rise curve of the cable conductor and the Joule heat accumulation.

[0014] The partial discharge time series data and the cable sheath temperature time series data are correlated and mapped to establish a correspondence matrix between partial discharge intensity and insulation hot spot temperature.

[0015] By integrating the environmental temperature and humidity time series data with the cable sheath temperature time series data, the heat dissipation efficiency coefficient and moisture penetration influence factor of the microenvironment in which the cable is located are calculated.

[0016] Based on the Joule heat accumulation, the corresponding relationship matrix, and the heat dissipation efficiency coefficient, a weighted fusion algorithm is used to generate the comprehensive cable state characteristic value at each sampling time.

[0017] The comprehensive cable status characteristic values ​​at all sampling times are arranged in chronological order to form the comprehensive cable status characteristic sequence.

[0018] Further, the electrothermal coupling analysis of the current load time-series data and the voltage fluctuation time-series data to generate the equivalent temperature rise curve and Joule heat accumulation of the cable conductor includes:

[0019] Calculate the root mean square current value and current harmonic distortion rate in each sampling period based on the current load timing data.

[0020] The number of voltage sags and the duration of overvoltage amplitude are calculated based on the voltage fluctuation time series data.

[0021] By combining the temperature coefficient of resistance and heat dissipation parameters of the cable conductor, a dynamic thermal balance equation is constructed. The dynamic thermal balance equation takes the root mean square current value as the main heat source input and the voltage sag number and overvoltage amplitude duration as additional stress conditions.

[0022] Solving the dynamic thermal balance equation yields the temperature rise of the cable conductor relative to the ambient reference temperature at each sampling moment, forming the equivalent temperature rise curve of the cable conductor.

[0023] The Joule heat accumulation is calculated by integrating the equivalent temperature rise curve of the cable conductor over time and adding the additional losses caused by the current harmonic distortion rate.

[0024] Further, the step of extracting degradation feature vectors characterizing the cable aging process from the overall cable condition feature sequence includes:

[0025] The time-domain and frequency-domain joint analysis of the cable's comprehensive state characteristic sequence was performed to identify periodic overload patterns and sudden abnormal event patterns in the characteristic sequence;

[0026] The cumulative load fatigue of the cable material due to cyclic stress is calculated based on the cyclic overload mode, and the cumulative load fatigue is equivalently accumulated based on the Miller criterion.

[0027] Based on the analysis of the sudden abnormal event patterns, the severity and frequency of electrical shocks to the insulation material are determined, and the insulation degradation index is calculated.

[0028] The duration of operation at high temperatures is separated from the overall cable condition characteristic sequence, and the thermal stress damage degree is calculated using the Arrhenius equation.

[0029] The cumulative load fatigue, the insulation degradation index, and the thermal stress damage degree are combined in a predetermined format to form the degradation feature vector.

[0030] Further, the step of inputting the degradation feature vector into a pre-trained cable life prediction model for time-series evolution extrapolation to generate the remaining life probability distribution and key failure time nodes of the target cable includes:

[0031] The pre-trained cable life prediction model is a hybrid model based on the fusion of a deep recurrent neural network and a Weibull proportional hazards model.

[0032] The degenerate feature vector is input into the deep recurrent neural network according to the time step to predict the evolution trajectory of the degenerate feature vector in the future period of time;

[0033] The predicted evolution trajectory is input into the Weibull proportional hazards model to calculate the probability of functional failure of the cable at different future time points.

[0034] The remaining lifetime probability distribution is formed by accumulating all time points where the probability of risk exceeds a preset failure threshold.

[0035] From the remaining lifetime probability distribution, identify the inflection point where the rate of increase in risk probability changes abruptly, and mark the time point corresponding to the inflection point as the critical failure time node.

[0036] Furthermore, the training process of the pre-trained cable life prediction model includes:

[0037] A large amount of historical data on the entire life cycle of the same type of cable from commissioning to failure was collected to form a training sample set. Each sample contains a time series degradation feature vector and a final failure label.

[0038] The time series data in the training sample set is divided into a training set and a validation set;

[0039] The deep recurrent neural network is trained using the training set to optimize its network weight parameters, enabling it to accurately learn the temporal evolution of degenerative feature vectors.

[0040] The output of the trained deep recurrent neural network is used as a feature, and together with the corresponding failure time data, it is used to train the Weibull proportional risk model to estimate its shape parameters and scale parameters.

[0041] The overall performance of the hybrid model is evaluated using the validation set, and the model hyperparameters are adjusted until the prediction error meets the predetermined requirements.

[0042] Furthermore, the step of generating a preventative maintenance decision plan based on the critical failure time point includes:

[0043] Obtain real-time data on the future maintenance schedule and spare parts inventory of the power grid system;

[0044] The critical failure time points are matched with the future maintenance schedule to find overlapping or adjacent maintenance windows and generate the maintenance time window suggestions.

[0045] Assess the severity level of the failure mode corresponding to the key failure time point and its impact on the power grid operation, and prioritize the maintenance operations of multiple cables to be maintained based on the assessment results.

[0046] Based on the predicted potential replacement demand for the same type of cable within a specific future period according to the key failure time nodes, and combined with the real-time spare parts inventory data, the spare parts inventory early warning information is generated.

[0047] Furthermore, the assessment of the severity level of the failure mode corresponding to the critical failure time point and its impact on power grid operation includes:

[0048] Analyze the evolution trend of the degradation feature vector near the critical failure time node to determine which type of failure mode it belongs to: insulation breakdown, conductor melting, or mechanical damage.

[0049] Based on the determined failure type, a preset failure consequence impact comparison table is queried to obtain the standard consequence data associated with the failure type, including the maximum power outage range, load loss level, and average repair time.

[0050] Based on the location of the target cable in the power grid topology and the importance level of the load it carries, calculate the comprehensive impact index of the failure of the target cable on the power supply reliability of the system.

[0051] The severity level is determined based on the magnitude of the comprehensive impact index.

[0052] Furthermore, the method also includes:

[0053] During the actual operation of the target cable, new multi-dimensional real-time operation data are continuously collected;

[0054] The new multidimensional real-time running data is processed by the multi-source heterogeneous data fusion and degradation feature vector extraction, and then input into the pre-trained cable life prediction model for online update prediction.

[0055] Compare the remaining life probability distribution obtained from the online update prediction with the historical prediction results. When the difference exceeds the set tolerance, trigger the model retraining instruction.

[0056] According to the model retraining instructions, newly added sample data that have run to failure are collected, and the pre-trained cable life prediction model is incrementally learned to update the model parameters to adapt to the current aging rate of the cable.

[0057] Furthermore, the present invention also includes a data-driven intelligent cable life prediction system, the system comprising a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps of the data-driven intelligent cable life prediction method described above.

[0058] Furthermore, the present invention also includes a cable, characterized in that the cable integrates a condition monitoring and lifespan prediction system, the system comprising:

[0059] The multi-source data acquisition module is used to collect real-time operational sequence data of the cable, including current load, voltage fluctuation, partial discharge, sheath temperature, and ambient temperature and humidity.

[0060] The data fusion and feature extraction module is communicatively connected to the multi-source data acquisition module and is configured to perform multi-source heterogeneous fusion processing on the acquired time-series data to generate a comprehensive cable status feature sequence, and extract degradation feature vectors containing load fatigue accumulation, insulation degradation index and thermal stress damage degree from it.

[0061] The life prediction module is communicatively connected to the data fusion and feature extraction module. It has a pre-trained cable life prediction model built in and is configured to receive the degradation feature vector and perform time-series evolution deduction to output the remaining life probability distribution and key failure time nodes of the cable.

[0062] The decision support module is communicatively connected to the life prediction module and is configured to generate a preventive maintenance decision plan based on the critical failure time node. The preventive maintenance decision plan includes maintenance time window suggestions, maintenance operation priority ranking, and spare parts inventory early warning information.

[0063] Compared with the prior art, the beneficial effects of the present invention are:

[0064] By deeply fusing multi-dimensional heterogeneous time-series data, including current load, voltage fluctuations, partial discharge, sheath temperature, and ambient temperature and humidity, a unified comprehensive cable condition characteristic sequence is constructed. This technology achieves effective alignment and intrinsic correlation mining of data from different physical dimensions and time scales, transforming previously isolated data streams into a holistic health status indicator. It overcomes the limitations of single-parameter analysis, simultaneously capturing composite degradation signals caused by the interaction of electrical, thermal, mechanical, and environmental stresses, thereby improving the comprehensiveness of condition monitoring and the ability to detect early, subtle defects.

[0065] Degradation feature vectors, such as cumulative load fatigue, insulation degradation index, and thermal stress damage degree, are extracted from the aforementioned fused features. These vectors are physically meaningful indicators constructed based on the aging mechanism of cable materials, rather than general statistical features. Inputting these mechanistic features into a pre-trained time-series model allows for the output of the probability distribution of remaining life and the identification of critical failure time points. This gives the prediction model stronger physical interpretability and expands the prediction results from a single value to a probabilistic form including confidence intervals. This enables maintenance personnel to quantify failure risks and formulate differentiated maintenance strategies based on the risk level at different time points.

[0066] Based on the critical failure time points identified by probabilistic life prediction, the system can automatically generate decision-making schemes that include specific maintenance time windows, operation priority ranking, and spare parts inventory warnings. This technology directly maps the prediction results into actionable operation and maintenance instructions. It transforms the traditional scheduled maintenance or reactive repair mode into precise preventive maintenance based on the actual health status and risk probability of the equipment, optimizing resource allocation and reducing the risk of unplanned downtime. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of the data-driven intelligent prediction method for cable life described in this invention.

[0068] Figure 2 A flowchart for multi-source heterogeneous data fusion processing;

[0069] Figure 3 A flowchart for extracting degenerate feature vectors;

[0070] Figure 4 A multi-dimensional comparative bar chart for assessing the impact of cable failure;

[0071] Figure 5 A five-dimensional radar chart for cable preventive maintenance decisions. Detailed Implementation

[0072] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0073] See Figure 1 A multi-dimensional historical operational dataset of the target cable was collected, including time-series data on current load, voltage fluctuation, partial discharge, cable sheath temperature, and ambient temperature and humidity. This dataset was then subjected to multi-source heterogeneous data fusion processing to generate a fused comprehensive cable condition feature sequence. Degradation feature vectors characterizing the cable aging process were extracted from this sequence, specifically including accumulated load fatigue, insulation degradation index, and thermal stress damage degree. The extracted degradation feature vectors were input into a pre-trained cable life prediction model for time-series evolution simulation, generating the remaining life probability distribution and critical failure time nodes of the target cable. Based on the obtained critical failure time nodes, a corresponding preventative maintenance decision plan was generated, including maintenance time window suggestions, maintenance operation priority ranking, and spare parts inventory early warning information.

[0074] In one embodiment of the present invention, see [reference] Figure 2 To generate a fused comprehensive cable status feature sequence, multi-source heterogeneous data fusion processing is performed on multi-dimensional historical operation datasets. For example, for a 35kV cross-linked polyethylene insulated cable operating in an urban power grid, a one-year multi-dimensional historical operation dataset is collected. To achieve continuous acquisition of multi-dimensional operation sequence data such as current, voltage, partial discharge, temperature, and ambient temperature and humidity, the cable integrates a multi-source data acquisition module. Specifically, distributed fiber optic temperature sensors are embedded near the cable conductor for real-time monitoring of conductor operating temperature; a UHF capacitive coupler is integrated between the cable insulation layer and the outer sheath for capturing partial discharge signals; a platinum resistance temperature sensor is laid on the cable sheath to monitor the sheath temperature; in addition, compact ambient temperature and humidity sensors are installed at cable terminal joints or fixed points along the line. These sensors transmit the collected time-series data to the data processing center in real time through a built-in lightweight communication unit. Current load timing data is sampled once per minute to record conductor current value; voltage fluctuation timing data is sampled once per second to record phase voltage value; partial discharge timing data is sampled once every ten minutes to record discharge amplitude; cable sheath temperature timing data is sampled once per minute to record outer sheath surface temperature; and ambient temperature and humidity timing data is sampled once per hour to record the temperature and relative humidity of the installation environment.

[0075] In practical implementation, electrothermal coupling analysis is performed on current load time-series data and voltage fluctuation time-series data to generate the equivalent temperature rise curve of the cable conductor and the Joule heat accumulation. For example, a typical 24-hour data segment of a working day is extracted from the current load time-series data, and the root mean square current value in each sampling period is calculated. The data shows that the average root mean square current during the daytime peak load period is 550 amperes, and the average root mean square current during the nighttime off-peak period is 300 amperes. At the same time, the current harmonic distortion rate is calculated, and the data shows that the current harmonic distortion rate during the peak period is 5%, and the current harmonic distortion rate during the off-peak period is 2%. Based on the voltage fluctuation time-series data, the number of voltage sags in the same time period is calculated. The data shows that there were 2 voltage sag events with a duration of less than 0.1 seconds, and the overvoltage amplitude duration is calculated. The data shows that there was 1 overvoltage event with a duration of 3 seconds and an amplitude of 1.15 times the rated voltage.

[0076] In practice, the time-series data of partial discharge quantity and the time-series data of cable sheath temperature are correlated and mapped to establish a matrix relating partial discharge intensity to hot spot temperature of insulation layer. For example, the time-series data of partial discharge quantity and cable sheath temperature within the same time period are selected. The time-series data of partial discharge quantity shows that the discharge amplitude fluctuates in the range of 5 picoctaves to 50 picoctaves, while the time-series data of cable sheath temperature shows that the temperature varies in the range of 30 degrees Celsius to 75 degrees Celsius. A discrete mapping table is established through statistical analysis. The data shows that when the cable sheath temperature exceeds 70 degrees Celsius, the average partial discharge intensity increases to 40 picoctaves. The correlation matrix stores the correlation weights between temperature range and discharge intensity range in the form of a two-dimensional array. By integrating time-series data on ambient temperature and humidity with time-series data on cable sheath temperature, the heat dissipation efficiency coefficient and moisture penetration impact factor of the cable's microenvironment can be calculated. For example, the ambient temperature and humidity time-series data shows that the average daytime ambient temperature in summer is 35 degrees Celsius and the average relative humidity is 85%, while the cable sheath temperature time-series data shows that the average temperature during the corresponding period is 70 degrees Celsius, resulting in a heat dissipation efficiency coefficient of 0.75. The ambient temperature and humidity time-series data shows that the average nighttime ambient temperature in winter is 0 degrees Celsius and the average relative humidity is 50%, while the cable sheath temperature time-series data shows that the average temperature during the corresponding period is 25 degrees Celsius, resulting in a heat dissipation efficiency coefficient of 1.25. Data comparison shows that the heat dissipation efficiency coefficient increases as the ambient temperature decreases. The moisture penetration impact factor is calculated based on humidity data and temperature gradient. The data shows that the moisture penetration impact factor is 0.9 in the high humidity environment of summer and 0.3 in the low humidity environment of winter.

[0077] Optionally, a weighted fusion algorithm is used to generate the comprehensive cable state feature value for each sampling time based on the Joule heat accumulation, the correspondence matrix, and the heat dissipation efficiency coefficient. In the weighted fusion algorithm, the weight of the Joule heat accumulation is set to 0.5, the weight of the correspondence matrix is ​​set to 0.3, and the weight of the heat dissipation efficiency coefficient is set to 0.2. For a certain sampling time, the normalized value of the Joule heat accumulation is 0.8, the normalized value of the correspondence matrix output is 0.6, and the normalized value of the heat dissipation efficiency coefficient is 0.4. Then, the comprehensive cable state feature value is 0.8 * 0.5 + 0.6 * 0.3 + 0.4 * 0.2 = 0.66. In some embodiments, the comprehensive cable state feature values ​​for all sampling times are arranged in chronological order to form a comprehensive cable state feature sequence. For example, the comprehensive cable state feature values ​​calculated every minute within a year are sorted by timestamp to form a comprehensive cable state feature sequence containing 525,600 data points. It is understandable that the generation of the cable comprehensive condition characteristic sequence depends on the synchronization and alignment of multi-source data. Data comparison shows that the cable comprehensive condition characteristic value sequence shows significant peaks on dates with frequent load fluctuations, while the sequence is relatively flat on dates with stable loads.

[0078] In one embodiment of the present invention, see [reference] Figure 3 Degradation feature vectors that characterize the aging process of cables are extracted from the comprehensive condition feature sequence of cables. For example, to accurately calculate the cumulative load fatigue, specific alloy conductor materials with known fatigue characteristic curves are selected for the cables; to correlate partial discharge with insulation hotspot temperatures to calculate the insulation degradation index, cross-linked polyethylene insulation material with clear temperature-discharge correlation characteristics is used in the cables, and the mapping relationship between insulation hotspots and surface temperatures is determined during the design phase; to calculate the degree of thermal stress damage using the Arrhenius equation, key parameters such as the thermal aging activation energy of the cable insulation material are clearly defined in the product design specifications. These designs based on the cable's own materials and structure are the premise that the degradation feature vectors have clear physical meaning. For example, the comprehensive condition feature sequence of a medium-voltage power cable operating in a coastal chemical industrial park is analyzed. The cable comprehensive condition feature sequence is three months long, with a sampling interval of ten minutes. The cable comprehensive condition feature sequence data shows that there are two significant load peak periods every day, and the cable comprehensive condition feature sequence values ​​fluctuate between 0.7 and 0.9. At night, the cable comprehensive condition feature sequence values ​​are stable between 0.3 and 0.5. Five peaks exceeding the threshold of 1.2 are recorded in the cable comprehensive condition feature sequence. In practice, the cable comprehensive condition characteristic sequence was analyzed in both the time and frequency domains to identify periodic overload patterns and sudden abnormal event patterns. The time domain analysis employed a sliding window statistical method, identifying stable periodic overload pattern windows from 9:00 AM to 11:00 AM and from 2:00 PM to 4:00 PM daily. Data showed that the mean of the cable comprehensive condition characteristic sequence exceeded 0.75 within these two windows. The frequency domain analysis used a Fast Fourier Transform, revealing a significant peak in the spectrum at a 24-hour period, confirming the daily cyclical pattern. For sudden abnormal event patterns, a threshold trigger mechanism was used for identification. Five spikes exceeding a threshold of 1.2 in the cable comprehensive condition characteristic sequence were marked as sudden abnormal events, with data showing that the duration of these spikes ranged from 2 to 15 minutes.

[0079] In some embodiments, the cumulative load fatigue caused by cyclic stress in the cable material is calculated based on identified periodic overload patterns. This cumulative load fatigue is equivalently accumulated based on the Miller criterion. The calculation process considers the stress cycles formed by the cable's overall state characteristic sequence within the periodic overload pattern window, mapping the cable's overall state characteristic sequence values ​​to equivalent mechanical stress levels. For each identified stress cycle, the resulting damage is calculated. (Cumulative Load Fatigue) The calculation formula is:

[0080]

[0081] in: This indicates the cumulative amount of load fatigue. Indicates the number of different stress level grades; Indicates the first The actual number of cycles experienced at each stress level; Indicates the cable material in the first The number of cycles required to reach failure at a given stress level is determined by examining the material fatigue characteristic curve. Data shows that, over a three-month monitoring period, the cumulative load fatigue contribution from cyclic overload modes was calculated to be 0.15. It can be understood that the insulation degradation index is calculated by analyzing the severity and frequency of electrical shocks to the insulation material based on identified sudden abnormal event patterns. For example, analyzing five marked sudden abnormal events, the peak amplitude and duration of the cable's comprehensive state characteristic sequence for each event are extracted. The peak amplitude data are 1.25, 1.31, 1.28, 1.19, and 1.40, respectively, and the duration data are 2 minutes, 5 minutes, 3 minutes, 15 minutes, and 2 minutes, respectively. Insulation Degradation Index The impact intensity of an event is calculated as follows: the product of the peak amplitude and duration of each event is taken as the impact intensity of that event. The impact intensities of all events are then weighted and summed, with the weights positively correlated with the frequency of the events. Data comparison shows that an event lasting 15 minutes with an amplitude of 1.19 contributes more to the insulation degradation index than an event lasting 2 minutes with an amplitude of 1.40. The final calculated insulation degradation index for the three-month monitoring period is 0.42.

[0082] Optionally, durations of prolonged high-temperature operation can be extracted from the cable's overall condition characteristic sequence to calculate thermal stress damage using the Arrhenius equation. For example, a threshold of 0.7 can be set to represent high-temperature operation. All periods exceeding this threshold can be extracted from the sequence. Data shows that daily peak hours constitute high-temperature operation periods lasting 2-3 hours, with a cumulative high-temperature operation time of 450 hours over three months. Thermal stress damage degree The calculation is based on the Arrhenius equation, which is expressed as:

[0083]

[0084] in: Indicates the degree of thermal stress damage. Indicates the pre-exponential factor; This represents the activation energy of the thermal aging reaction of cable insulation materials. Represents the ideal gas constant; Indicates time The absolute temperature, which is obtained by converting the cable's comprehensive state characteristic sequence value into the insulation hot spot temperature through a linear mapping relationship; This indicates the total monitoring time.

[0085] During calculation, the comprehensive state characteristic sequence of the cable in each high-temperature operating section is converted into the corresponding temperature value and substituted into the integral. Data shows that in the operating section with an average mapped temperature of 355 Kelvin, its contribution to the thermal stress damage degree is much higher than that in the operating section with an average mapped temperature of 340 Kelvin, and the final calculated thermal stress damage degree is 0.08. In some embodiments, the calculated load fatigue accumulation, insulation degradation index, and thermal stress damage degree are combined according to a predetermined format to form a degradation feature vector. The predetermined format adopts a triplet ordered data structure, for example, combining the load fatigue accumulation of 0.15, the insulation degradation index of 0.42, and the thermal stress damage degree of 0.08 into a degradation feature vector [0.15, 0.42, 0.08]. It can be understood that data comparison shows that another cable operating in a commercial area with a stable load has a degradation feature vector of [0.05, 0.11, 0.02] generated during a three-month monitoring period. The degradation feature vector of the cable in the industrial area has significantly higher values ​​for each component, indicating a more severe aging process.

[0086] In one embodiment of the invention, degradation feature vectors are input into a pre-trained cable life prediction model for time-series evolution simulation to generate the remaining life probability distribution and critical failure time nodes of the target cable. For example, for a medium-voltage cable operating in a wind farm collector line, its historical degradation feature vector sequence includes degradation feature vectors extracted monthly over the past 36 months. Each degradation feature vector contains three components: cumulative load fatigue, insulation degradation index, and thermal stress damage degree. Data shows that the degradation feature vector sequence exhibits a slow upward trend, with the insulation degradation index component showing a faster growth rate over the past 12 months. In specific implementation, the pre-trained cable life prediction model is a hybrid model based on the fusion of a deep recurrent neural network and a Weibull proportional hazards model. The deep recurrent neural network adopts a long short-term memory network structure with three hidden layers, each containing 128 neurons. The Weibull proportional hazards model defines the relationship between cable failure risk and covariates. To ensure the long-term, consistent, and high-quality data input required for the pre-trained model, this intelligent prediction cable emphasizes the durability and stability of the condition monitoring module. The sensor modules integrated within the cable are designed with a wide temperature range and high electromagnetic interference resistance to ensure continuous data acquisition throughout the cable's entire lifespan. Sensor data is transmitted via shielded twisted-pair cables or fiber optic communication buses integrated within the cable, reducing signal attenuation and external interference. This integrated embedded monitoring design avoids errors caused by external sensor installations, providing a reliable data source for the lifespan prediction model. In practical implementation, degradation feature vectors are input into a deep recurrent neural network at time steps to predict the evolution trajectory of degradation feature vectors over a future period. The time step is set to 30 months, using the degradation feature vector sequence of the first 30 months as input to predict the degradation feature vector sequence of the following 12 months. The predicted data output by the deep recurrent neural network shows that in the next 12 months, the predicted value of cumulative load fatigue will reach 0.85, the predicted value of insulation degradation index will reach 1.15, and the predicted value of thermal stress damage will reach 0.45.

[0087] In some embodiments, the predicted evolution trajectory is input into the Weibull proportional hazards model to calculate the probability of functional failure of the cable at different future time points. The risk function of the Weibull proportional hazards model is in the form of:

[0088]

[0089] in: This indicates that given the covariates at time point t Risk probability density at that time; The shape parameter of the Weibull distribution controls the trend of the risk function over time. The scale parameter of the Weibull distribution represents the characteristic lifetime. This represents the covariate function at time t, whose value is a linear combination of the degenerate feature vectors predicted by the deep recurrent neural network. This represents the regression coefficient vector of the covariates. Data shows that when the degradation feature vector for the predicted 12th month is input, the calculated risk probability for that month is 0.78. It can be understood that the cumulative risk probability exceeds a preset failure threshold at all time points to form the remaining life probability distribution. The preset failure threshold is set to 0.8. The risk probability is calculated for each future month starting from the current month. Data shows that the risk probability for the 13th month is 0.83, and the risk probability for the 14th month is 0.87. Therefore, the remaining life probability distribution shows that the cable's failure probability exceeds the threshold between the 13th and 14th months. The inflection point of the rate of change in risk probability growth was identified from the remaining lifetime probability distribution, and the time point corresponding to the inflection point was marked as the critical failure time node. The data showed that the risk probability increased from 0.65 in the 11th month to 0.78 in the 12th month, with a growth rate of 0.13; from 0.78 in the 12th month to 0.83 in the 13th month, with a growth rate of 0.05; and from 0.83 in the 13th month to 0.87 in the 14th month, with a growth rate of 0.04. The rate of change in risk probability growth was the largest from the 11th month to the 12th month. Therefore, the 12th month was marked as the critical failure time node.

[0090] The training process of the pre-trained cable life prediction model is implemented as follows: A large amount of historical data on the entire lifecycle of cables of the same model, from commissioning to failure, is collected to form a training sample set. For example, historical data for 200 cables of the same specification are collected. Each data point contains a monthly degradation feature vector sequence from installation to failure or replacement, along with the corresponding failure time label. The failure time label records the total operating life in months. The data shows that the lifespan of the sample cables ranges from 60 to 180 months. The time-series data in the training sample set is divided into a training set and a validation set. The training set contains data for 160 cables, and the validation set contains data for 40 cables. The training set is used to train a deep recurrent neural network to optimize its network weight parameters so that it can accurately learn the temporal evolution of degradation feature vectors. During training, the historical sequence is used as input, and the degradation feature vector of the next time step is used as the prediction target. The prediction error is minimized through backpropagation. After 200 rounds of training, the mean square error of the deep recurrent neural network's prediction on the training set is reduced to 0.02. In some embodiments, the output of the trained deep recurrent neural network is used as a feature along with the corresponding failure time data to train a Weibull proportional hazards model to estimate its shape and scale parameters. The training set sequence is processed using the deep recurrent neural network to obtain the final hidden state vector, which is then input as a covariate into the Weibull proportional hazards model. The maximum likelihood estimation method is used to fit the shape parameter μ and the scale parameter ν. The fitting results show that the estimated value of the shape parameter μ is 2.5, and the estimated value of the scale parameter ν is 120 months. Optionally, the overall performance of the hybrid model is evaluated using a validation set to adjust the model hyperparameters until the prediction error meets the predetermined requirements. The historical degradation feature vector of the validation set cable is input into the hybrid model to predict its remaining lifetime distribution and compare it with the actual failure time. The deviation between the predicted lifetime quantile and the actual lifetime is calculated. Data shows that the prediction deviation of 80% of the samples is within ±10%. After adjusting the number of hidden layer nodes, learning rate, and other hyperparameters of the deep recurrent neural network, the deviation is further reduced to within ±8%. Understandably, data comparison shows that for a cable that failed due to insulation breakdown after 45 months of actual operation, the hybrid model predicted the critical failure time point to be the 46th month when it was in operation for the 36th month, and the prediction result is basically consistent with the actual failure time; while for another cable with a milder operating environment, the model predicted the critical failure time point to be the 132nd month, and it was actually taken out of operation for routine replacement at 140 months, which is consistent with the predicted trend.

[0091] In one embodiment of the present invention, a preventative maintenance decision-making scheme is generated based on critical failure time nodes. For example, for three different circuit cables in a city subway power supply network, the cable life prediction model outputs the critical failure time nodes for the three cables: November 15, 2025 for cable A, March 22, 2026 for cable B, and January 10, 2026 for cable C. To achieve accurate maintenance decision support, each intelligent prediction cable is assigned a unique identification code at the factory. This identification code is bound to information such as the cable's model specifications, design parameters, installation location, and load level, and is entered into the power grid asset management system. When the life prediction module outputs the critical failure time nodes, the decision support module can accurately assess its location importance and load level in the power grid topology by calling the cable background information associated with this identification code, thereby generating targeted maintenance time window suggestions and operation priority rankings. In practice, the system obtains real-time data on the future maintenance schedule and spare parts inventory of the power grid system. The future maintenance schedule shows that there is a planned maintenance window for this power supply section from November 10 to November 20, 2025, and another planned maintenance window for a different section from January 5 to January 15, 2026. The real-time spare parts inventory data shows that the current inventory of spare parts for the same type of cable is 2 pieces. By matching critical failure time points with future maintenance schedules to identify overlapping or adjacent maintenance windows, maintenance time window recommendations are generated. For cable A, the critical failure time point of November 15, 2025, falls entirely within the planned maintenance window of November 10 to November 20, 2025. Therefore, the recommended maintenance time window is "November 10 to 20, 2025, synchronized with planned maintenance." For cable C, the critical failure time point of January 10, 2026, falls within the planned maintenance window of January 5 to 15, 2026. The recommended maintenance time window is "January 5 to 15, 2026, synchronized with planned maintenance." For cable B, the critical failure time point of March 22, 2026, does not overlap with any planned maintenance window. The recommended maintenance time window is "March 15 to 30, 2026, special maintenance is recommended."

[0092] In some embodiments, the severity level of the failure mode corresponding to the critical failure time point and its impact on power grid operation are assessed to prioritize maintenance operations for multiple cables to be maintained based on the assessment results. The assessment process first analyzes the evolution trend of the degradation feature vector near the critical failure time point to determine whether the failure mode belongs to insulation breakdown, conductor melting, or mechanical damage. For cable A, the degradation feature vector shows a sharp increase in the insulation degradation index during the prediction period, while the load fatigue accumulation increases slowly, indicating insulation breakdown as the failure mode. For cable B, the degradation feature vector shows a rapid increase in both load fatigue accumulation and thermal stress damage, indicating conductor melting as the failure mode. For cable C, the degradation feature vector shows a slow growth trend in all three components, without exhibiting a single mode of rapid degradation, indicating a combination of mechanical damage and insulation aging as the failure mode. Based on the determined failure type, a preset failure consequence impact comparison table is consulted to obtain the standard consequence data associated with the failure type. See Table 1 for the contents of the failure consequence impact comparison table.

[0093] Table 1: Comparison of Failure Consequences

[0094] Failure type Maximum power outage area Load loss level Average repair time Insulation breakdown 2 substations Level 1 (>80% load) 72 hours Conductor meltdown 1 substation Level 2 (50%-80% load) 48 hours Mechanical damage Local lines Level 3 (<50% load) 24 hours

[0095] The comprehensive impact index of the failure of the target cable on the power supply reliability of the system is calculated by combining the location of the target cable in the power grid topology and the importance level of the load it carries. Cable A is located at the connection node of the double ring network and carries the load of the core substation of the subway. Its location importance level is "critical" and its load importance level is "level one". Cable B is located at the end of a single radial line and carries general commercial load. Its location importance level is "minor" and its load importance level is "level two". Cable C is located in the middle section of the main line and carries residential and public facility load. Its location importance level is "important" and its load importance level is "level two".

[0096] Optionally, based on the predicted critical failure time points, the potential replacement demand for the same type of cable within a specific future period is generated in conjunction with real-time spare parts inventory data to produce a spare parts inventory warning. The specific future period is set to the next 6 months, and the critical failure time points for both cables A and C fall within this period. The predicted potential replacement demand is 2 cables; the real-time spare parts inventory data shows 2 cables. Data comparison shows that the predicted demand equals the current inventory, generating a spare parts inventory warning: "Predicted replacement demand in the next 6 months: 2 cables; current inventory: 2 cables; it is recommended to maintain inventory or replenish stock in advance according to the procurement cycle." In some embodiments, if data comparison shows that the predicted demand is 3 cables while the inventory is 2 cables, then the spare parts inventory warning is generated: "Predicted replacement demand in the next 6 months: 3 cables; current inventory: 2 cables; inventory shortfall: 1 cable; it is recommended to immediately initiate the procurement process." It can be understood that the maintenance time window recommendation, maintenance operation priority ranking, and spare parts inventory warning information together constitute a preventative maintenance decision-making scheme for this subway power supply network.

[0097] See Figure 4 This is a multi-dimensional comparative bar chart used for cable failure impact assessment. It clearly quantifies the different levels of impact of three typical cable failure modes on the power grid system. Insulation breakdown has the most severe impact, with a load loss rate exceeding 80%, a repair time of approximately 72 hours, and affecting two substations. It is the failure mode that poses the greatest threat to power grid operation. Mechanical damage has the lowest impact, with a load loss rate of approximately 35%, a repair time of approximately 24 hours, and only affects local lines. It is usually caused by external force damage or construction damage. This chart provides a quantitative basis for prioritizing maintenance; for example, "insulation breakdown" should be listed as the highest priority for prevention and repair. High-impact failure modes (such as insulation breakdown) require more backup cables and emergency repair resources to shorten recovery time. It can guide the operation and maintenance team to strengthen the monitoring of insulation performance (such as partial discharge detection) to reduce the probability of high-risk failures.

[0098] In one embodiment of the invention, new multi-dimensional real-time operational data is continuously collected during the actual operation of the target cable. For example, an online monitoring system is deployed for a 35kV underground cable supplying power to an industrial park. The online monitoring system continuously collects current load time-series data, voltage fluctuation time-series data, partial discharge time-series data, cable sheath temperature time-series data, and ambient temperature and humidity time-series data at a frequency of once per minute. The new multi-dimensional real-time operational data is streamed to a data center. To enable online updates and predictions of the model, the data acquisition module integrated into the cable has edge computing capabilities. The microprocessor built into the module can perform preliminary filtering, calibration, and caching of the raw monitoring data, and maintain low-latency communication with the cloud data center through an Industrial Internet of Things (IIoT) protocol. The cable body takes into account the power supply of the sensors, and can use inductive energy harvesting or micro-energy collection technology to provide continuous power to the monitoring module. This integrated design of self-sensing, self-powering, and self-communication ensures stable and continuous streaming transmission of multi-dimensional real-time operational data, providing a data foundation for online updates of lifespan prediction. New multidimensional real-time operating data is processed through multi-source heterogeneous data fusion and degradation feature vector extraction, and then input into a pre-trained cable life prediction model for online update prediction. Multi-source heterogeneous data fusion is performed according to the method described in the embodiment to generate the comprehensive cable status feature value at the current moment. Degradation feature vector extraction is performed according to the method described in the embodiment to calculate a degradation feature vector containing the latest load fatigue accumulation, insulation degradation index, and thermal stress damage degree. The latest degradation feature vector is combined with the historical sequence and then input into the pre-trained cable life prediction model. The model outputs the online update prediction of the remaining life probability distribution based on the latest data. The data shows that the online update prediction indicates that the probability of cable failure exceeds the threshold after 18 months.

[0099] The difference between the remaining lifetime probability distribution obtained from the online update prediction and the historical prediction results is compared to determine whether a set tolerance has been exceeded. The historical prediction results were generated three months ago based on data at that time, and the predicted probability of cable failure after 24 months exceeded a threshold. The set tolerance is defined as the absolute value of the relative deviation between the median lifetime of the historical prediction and the median lifetime of the online update prediction exceeding 15%. The median lifetime of the historical prediction is 24 months, and the median lifetime of the online update prediction is 18 months, so the absolute value of the relative deviation is |(24-18) / 24|=25%. A difference greater than 15% triggers a model retraining instruction. In some embodiments, the set tolerance can also be defined based on the overall shape difference of the remaining lifetime probability distribution, for example, by calculating the Jason-Shannon divergence between the two probability distributions. A model retraining instruction is triggered when the divergence value exceeds 0.1. Understandably, the model retraining instruction collects newly added sample data from operation to failure to incrementally learn the pre-trained cable life prediction model in order to update the model parameters to adapt to the current aging rate of the cable. After the model retraining instruction is triggered, it retrieves recent failure cases of the same type of cable from the asset management database and collects the full life cycle data of a cable that was put into operation six months ago and recently decommissioned due to overheating failure as a new sample. The new sample contains a complete temporal degradation feature vector and an accurate failure time label.

[0100] In practice, incremental learning is performed on the pre-trained cable life prediction model. This process retains the original model parameters while merging new sample data with some existing training sample data to form an incremental training set. The deep recurrent neural network is then incrementally trained using this training set to optimize its network weight parameters. A small learning rate is used during incremental training to prevent significant overwriting of existing knowledge. The output features of the incrementally trained deep recurrent neural network, along with the corresponding failure time data, are used to update the shape and scale parameters of the Weibull proportional hazards model. The update process employs a Bayesian update method, treating the original model parameters as prior information and deriving posterior parameters based on the likelihood function of the new samples. The model parameters after incremental learning reflect the accelerated aging rate of the cable due to recent changes in the operating environment. Optionally, data comparison shows that before incremental learning, the model's prediction of the remaining life of an in-service cable with similar operating conditions to a recently failed cable was overly optimistic. After incremental learning, based on the same input data, the model's prediction of the remaining life of this in-service cable was shortened by approximately 20%, and the prediction trend is more consistent with the recently observed accelerated aging phenomenon. In some embodiments, the threshold setting tolerance for triggering the model retraining instruction can be dynamically adjusted based on the prediction confidence interval. When the confidence interval of the online updated prediction is wide, the tolerance is appropriately relaxed to avoid unnecessary frequent retraining. It can be understood that the triggering of the model retraining instruction and the execution of incremental learning form a closed-loop process, ensuring that the pre-trained cable life prediction model can adapt to the time-varying characteristics of cable aging behavior. The formula is used to quantify the difference between the online updated prediction and the historical prediction results, and the degree of difference... The calculation formula is:

[0101]

[0102] in: Indicates the degree of difference in predictions; This represents the median lifespan based on historical forecasts. This represents the median lifetime predicted through online updates. Greater than the preset threshold When this happens, the model retraining instruction is triggered.

[0103] See Figure 5This is a five-dimensional radar chart for cable preventative maintenance decisions. It quantifies the comprehensive evaluation results of maintenance decisions across five core dimensions, visually demonstrating the priority and importance of each dimension. The high-scoring combination of "Maintenance Time Window Recommendation" and "Maintenance Operation Priority" directly supports the decision to "immediately initiate maintenance actions synchronized with planned overhauls." Differences in scores across dimensions can guide the maintenance team in allocating resources such as time, manpower, and spare parts, for example, prioritizing resource needs for high-scoring dimensions. The overall shape of the radar chart reflects the balance between risk, cost, and efficiency in the decision-making process. The current chart shows no obvious weaknesses, indicating a relatively mature maintenance decision-making scheme.

[0104] In one embodiment of the present invention, the multi-source data acquisition module comprises multiple physical sensing units and a data aggregation unit. For an operating 35kV cross-linked polyethylene insulated cable, the module includes: a Hall effect current sensor installed near the cable conductor for monitoring load current; a voltage transformer installed at the cable terminal for monitoring voltage fluctuations; an ultra-high frequency partial discharge sensor integrated at the cable joint; a distributed fiber optic temperature sensing system laid on the cable sheath; and temperature and humidity sensors deployed in the cable trench or shaft. Each sensor independently acquires raw signals at a preset sampling frequency and uploads standardized packaged runtime sequence data in real time to the data aggregation unit located in the substation via a fieldbus or wireless network integrated within the cable. This aggregation unit is responsible for preliminary data verification, timestamp synchronization, and buffering, constituting the front end of the data acquisition.

[0105] The data fusion and feature extraction module is deployed on an edge server or cloud data center at the substation. This module establishes a communication connection with the data aggregation unit via industrial Ethernet or a dedicated power communication network, receiving time-series data streams from the multi-source data acquisition module. Internally, the module executes multi-level processing logic: first, it cleans and aligns heterogeneous data to ensure that data from different sampling rates can be correlated and analyzed under a unified time reference; then, it executes an electrothermal coupling analysis subroutine to convert current and voltage data into conductor equivalent temperature rise curves and Joule heat accumulation; next, it runs an association mapping subroutine to analyze the correlation between partial discharge activity and cable sheath temperature; and finally, it executes an environmental fusion subroutine to assess the impact of temperature and humidity on cable heat dissipation and insulation status. Finally, through a weighted fusion algorithm, the intermediate physical quantities generated from the above analysis are combined into a scalar at each sampling time, generating a long-term comprehensive cable state characteristic sequence. Furthermore, the module has a built-in feature extraction engine that performs time-domain and frequency-domain pattern recognition on the feature sequence to identify typical aging modes such as periodic overload, sudden anomalies, and continuous high temperature. Based on Miller's criterion, electrical shock damage model, and Arrhenius thermal aging model, it calculates the cumulative load fatigue, insulation degradation index, and thermal stress damage degree, and combines them into a degradation feature vector.

[0106] The cable lifetime prediction module is also deployed on an edge server or in the cloud. This module communicates with the data fusion and feature extraction module via an internal data bus or API interface, receiving the degradation feature vectors in real-time or periodically. The core of the module is a pre-trained cable lifetime prediction model, a hybrid structure of a deep recurrent neural network and a Weibull proportional hazards model, trained and solidified using a large amount of historical full-lifecycle data of similar cables. When a new degradation feature vector is received, the model first uses its deep recurrent neural network portion to learn the state evolution trend and predict the future evolution trajectory; subsequently, the Weibull proportional hazards model portion calculates the failure probability of the cable at various future time points based on this predicted trajectory. The module's output is not a single lifetime value, but rather a remaining lifetime probability distribution presented as a probability distribution function or curve, along with one or more key failure time points determined by analyzing the inflection points of this distribution curve.

[0107] The decision support module is deployed on the power grid company's operation and maintenance management platform. This module communicates with the lifespan prediction module via an enterprise service bus or dedicated data interface, subscribing to and obtaining predicted critical failure time points. The module integrates linkage logic with external systems: it proactively queries the future maintenance plan database of the power grid production management system, intelligently matching and optimizing predicted critical failure time points with planned outage windows to generate specific maintenance time window suggestions. Simultaneously, the module accesses cable topology and load importance information from the asset management system, combining this with predicted failure modes to assess the severity of failure consequences, thereby generating a maintenance operation priority ranking for multiple cables awaiting maintenance. Furthermore, the module connects to the real-time inventory database of the materials management system, statistically analyzing future demand based on the predicted critical failure time points of all cables, comparing it with current inventory levels, and automatically generating different levels of spare parts inventory warning information. After integration, the aforementioned maintenance time window suggestions, operation priority ranking, and spare parts inventory warning information are output in the form of structured reports or visual dashboards, constituting a complete preventative maintenance decision-making scheme for operation and maintenance personnel to implement.

[0108] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A data-driven intelligent prediction method for cable life, characterized in that, include: Collect a multidimensional historical operation dataset of the target cable, which includes current load time series data, voltage fluctuation time series data, partial discharge time series data, cable sheath temperature time series data, and ambient temperature and humidity time series data. The multi-dimensional historical operation dataset is subjected to multi-source heterogeneous data fusion processing to generate a fused comprehensive cable status feature sequence; Extracting degradation feature vectors that characterize the cable aging process from the overall cable condition feature sequence includes: The time-domain and frequency-domain joint analysis of the cable's comprehensive state characteristic sequence was performed to identify periodic overload patterns and sudden abnormal event patterns in the characteristic sequence; The cumulative load fatigue of the cable material due to cyclic stress is calculated based on the periodic overload mode, and the cumulative load fatigue is equivalently accumulated based on Miller's criterion. Based on the analysis of the sudden abnormal event patterns, the severity and frequency of electrical shocks to the insulation material are determined, and the insulation degradation index is calculated. The duration of operation at high temperatures is separated from the overall cable condition characteristic sequence, and the thermal stress damage degree is calculated by combining the Arrhenius equation. The cumulative load fatigue, the insulation degradation index, and the thermal stress damage degree are combined in a predetermined format to form the degradation feature vector. The degradation feature vector is input into a pre-trained cable life prediction model for time-series evolution simulation to generate the remaining life probability distribution and key failure time nodes of the target cable, including: The pre-trained cable life prediction model is a hybrid model based on the fusion of a deep recurrent neural network and a Weibull proportional hazards model. The degenerate feature vector is input into the deep recurrent neural network according to the time step to predict the evolution trajectory of the degenerate feature vector in the future period of time; The predicted evolution trajectory is input into the Weibull proportional hazards model to calculate the probability of functional failure of the cable at different future time points. The remaining lifetime probability distribution is formed by accumulating all time points where the probability of risk exceeds a preset failure threshold. From the remaining lifetime probability distribution, identify the inflection point where the risk probability growth rate changes abruptly, and mark the time point corresponding to the inflection point as the critical failure time node; A preventive maintenance decision plan is generated based on the key failure time nodes. The preventive maintenance decision plan includes maintenance time window suggestions, maintenance operation priority ranking, and spare parts inventory early warning information.

2. The data-driven intelligent prediction method for cable life as described in claim 1, characterized in that, The step of performing multi-source heterogeneous data fusion processing on the multi-dimensional historical operation dataset to generate a fused comprehensive cable status feature sequence includes: Electrothermal coupling analysis is performed on the current load time series data and the voltage fluctuation time series data to generate the equivalent temperature rise curve of the cable conductor and the Joule heat accumulation. The partial discharge time series data and the cable sheath temperature time series data are correlated and mapped to establish a correspondence matrix between partial discharge intensity and insulation hot spot temperature. By integrating the environmental temperature and humidity time series data with the cable sheath temperature time series data, the heat dissipation efficiency coefficient and moisture penetration influence factor of the microenvironment in which the cable is located are calculated. Based on the Joule heat accumulation, the corresponding relationship matrix, and the heat dissipation efficiency coefficient, a weighted fusion algorithm is used to generate the comprehensive cable state characteristic value at each sampling time. The comprehensive cable status characteristic values ​​at all sampling times are arranged in chronological order to form the comprehensive cable status characteristic sequence.

3. The data-driven intelligent prediction method for cable life according to claim 2, characterized in that, The electrothermal coupling analysis of the current load time series data and the voltage fluctuation time series data to generate the equivalent temperature rise curve and Joule heat accumulation of the cable conductor includes: Calculate the root mean square current value and current harmonic distortion rate in each sampling period based on the current load timing data. The number of voltage sags and the duration of overvoltage amplitude are calculated based on the voltage fluctuation time series data. By combining the temperature coefficient of resistance and heat dissipation parameters of the cable conductor, a dynamic thermal balance equation is constructed. The dynamic thermal balance equation takes the root mean square current value as the main heat source input and the voltage sag number and overvoltage amplitude duration as additional stress conditions. Solving the dynamic thermal balance equation yields the temperature rise of the cable conductor relative to the ambient reference temperature at each sampling moment, forming the equivalent temperature rise curve of the cable conductor. The Joule heat accumulation is calculated by integrating the equivalent temperature rise curve of the cable conductor over time and adding the additional losses caused by the current harmonic distortion rate.

4. The data-driven intelligent prediction method for cable life according to claim 3, characterized in that, The training process of the pre-trained cable life prediction model includes: A large amount of historical data on the entire life cycle of the same type of cable from commissioning to failure was collected to form a training sample set. Each sample contains a time series degradation feature vector and a final failure label. The time series data in the training sample set is divided into a training set and a validation set; The deep recurrent neural network is trained using the training set to optimize its network weight parameters, enabling it to accurately learn the temporal evolution of degenerative feature vectors. The output of the trained deep recurrent neural network is used as a feature, and together with the corresponding failure time data, it is used to train the Weibull proportional risk model to estimate its shape parameters and scale parameters. The overall performance of the hybrid model is evaluated using the validation set, and the model hyperparameters are adjusted until the prediction error meets the predetermined requirements.

5. The data-driven intelligent prediction method for cable life according to claim 1, characterized in that, The step of generating a preventative maintenance decision plan based on the critical failure time points includes: Obtain real-time data on the future maintenance schedule and spare parts inventory of the power grid system; The critical failure time points are matched with the future maintenance schedule to find overlapping or adjacent maintenance windows and generate the maintenance time window suggestions. Assess the severity level of the failure mode corresponding to the key failure time point and its impact on the power grid operation, and prioritize the maintenance operations of multiple cables to be maintained based on the assessment results. Based on the predicted number of potential replacements for the same type of cable within a specific future period according to the key failure time nodes, and combined with the real-time spare parts inventory data, the spare parts inventory early warning information is generated. The assessment of the severity level of the failure mode corresponding to the critical failure time point and its impact on power grid operation includes: Analyze the evolution trend of the degradation feature vector near the critical failure time node to determine which type of failure mode it belongs to: insulation breakdown, conductor melting, or mechanical damage. Based on the determined failure type, a preset failure consequence impact comparison table is queried to obtain the standard consequence data associated with the failure type, including the maximum power outage range, load loss level, and average repair time. Based on the location of the target cable in the power grid topology and the importance level of the load it carries, calculate the comprehensive impact index of the failure of the target cable on the power supply reliability of the system. The severity level is determined based on the magnitude of the comprehensive impact index.

6. The data-driven intelligent prediction method for cable life according to claim 1, characterized in that, The method further includes: During the actual operation of the target cable, new multi-dimensional real-time operation data are continuously collected; The new multidimensional real-time running data is processed by the multi-source heterogeneous data fusion and degradation feature vector extraction, and then input into the pre-trained cable life prediction model for online update prediction. Compare the remaining life probability distribution obtained from the online update prediction with the historical prediction results. When the difference exceeds the set tolerance, trigger the model retraining instruction. According to the model retraining instructions, newly added sample data that have run to failure are collected, and the pre-trained cable life prediction model is incrementally learned to update the model parameters to adapt to the current aging rate of the cable.

7. A data-driven intelligent cable life prediction system, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps of the data-driven intelligent prediction method for cable life as described in any one of claims 1 to 6.

8. A cable, characterized in that, The cable integrates a condition monitoring and life prediction system, the system comprising: The multi-source data acquisition module is used to collect real-time operational sequence data of the cable, including current load, voltage fluctuation, partial discharge, sheath temperature, and ambient temperature and humidity. The data fusion and feature extraction module is communicatively connected to the multi-source data acquisition module and is configured to perform multi-source heterogeneous fusion processing on the acquired time-series data to generate a comprehensive cable status feature sequence, and extract degradation feature vectors containing load fatigue accumulation, insulation degradation index and thermal stress damage degree from it. The life prediction module is communicatively connected to the data fusion and feature extraction module. It has a pre-trained cable life prediction model built in and is configured to receive the degradation feature vector and perform time-series evolution deduction to output the remaining life probability distribution and key failure time nodes of the cable. The decision support module is communicatively connected to the life prediction module and is configured to generate a preventive maintenance decision plan based on the critical failure time node. The preventive maintenance decision plan includes maintenance time window suggestions, maintenance operation priority ranking, and spare parts inventory early warning information.

Images