A method and system for fault prediction of connectors in intelligent power distribution equipment
By acquiring multi-dimensional feature signals and processing time-series alignment, combined with the probabilistic prediction method of Weibull distribution, the problem of identifying the progressive degradation trend of power distribution connectors was solved, enabling fault prediction and preventive maintenance of connectors, and improving the power supply stability and continuity of the power distribution network.
Patent Information
- Application Number
- CN202611279212.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies cannot effectively identify the progressive degradation trend of power distribution connectors, and cannot achieve early warning and preventive operation and maintenance, resulting in the failure to detect hidden degradation risks, making it difficult to meet the power supply stability and power supply continuity requirements of the power distribution network.
By synchronously acquiring and temporally aligning multidimensional feature signals, a multidimensional dynamic dataset is constructed. Degradation trajectory construction and pattern recognition are performed. Combined with the probability prediction method of Weibull distribution, the remaining lifetime is calculated and graded early warning is given, thereby realizing fault prediction of connectors.
It enables full-process tracking and pattern recognition of the degradation process of power distribution connectors, which can predict the time of failure in advance, improve the reliability of operation and maintenance, and reduce the probability of accidents such as equipment overheating, partial discharge and fire.
Smart Images

Figure CN123116067A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power equipment technology, specifically a fault prediction method and system for intelligent power distribution equipment connectors. Background Technology
[0002] With the development trend of intelligent and refined operation and maintenance of power grids, the connector of intelligent power distribution equipment, as the main connection hub between cables and power distribution equipment, directly determines the stability and continuity of power supply in the power distribution network in the long term. However, connector failure is not an instantaneous failure, but a gradual cumulative degradation process caused by long-term effects of thermal cycling, mechanical vibration, oxidation and corrosion. It is accompanied by the slow evolution of multi-dimensional parameters such as temperature, contact resistance, and partial discharge, and has extremely strong concealment and gradual characteristics.
[0003] Current power distribution connector maintenance and diagnostic technologies generally employ manual periodic inspections, fixed threshold over-limit alarms, and online fault diagnosis modes based on fixed rules. These methods rely on comparing real-time values of a single parameter with preset thresholds to achieve instantaneous monitoring and alarm of connector fault status.
[0004] Among the relevant diagnostic technologies, there is a fundamental technical deficiency in adapting to progressive degradation faults. It completely abandons the temporal evolution characteristics and multi-parameter collaborative change law of the connector degradation process, only focuses on the current instantaneous state of the equipment, and cannot capture the latent degradation trend of parameters slowly rising over a long period of time. It has no ability to identify progressive damage that is below the threshold for a long time but continues to deteriorate. It can only detect obvious faults that have occurred after the fact, and does not have the ability to trace the degradation trajectory, judge the trend, and predict the remaining life. It is easy to miss hidden degradation risks, and it is difficult to meet the intelligent operation and maintenance needs of power distribution connectors for early warning and preventive operation and maintenance.
[0005] Therefore, the present invention provides a fault prediction method and system for intelligent power distribution equipment connectors to solve the technical problems mentioned above. Summary of the Invention
[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.
[0007] The technical solution adopted by this invention to solve its technical problem is as follows: On the one hand, the fault prediction method for intelligent power distribution equipment connectors described in this invention, applied to power distribution equipment connectors, includes the following steps: Step 1: Synchronously collect multi-dimensional operating condition data, including thermal and electrical characteristics, through a sensor network set at the connector monitoring points, and construct a multi-dimensional dynamic dataset reflecting the operating status of the connector; Step 2: Perform a time window-based soft alignment operation on the multidimensional dataset to map the original data with different sampling frequencies to a unified discrete time grid. Use an interpolation algorithm to eliminate the phase deviation introduced by the sampling rate difference, and at the same time perform outlier removal and missing data imputation. Step 3: Normalize the feature signals of each dimension after time alignment, map parameters with different dimensions to a preset value range, determine the weight allocation based on the sensitivity of each parameter to the connector degradation process, and generate a continuous degradation trajectory curve through weighted fusion operation. Step 4: Calculate the slope of the sliding window and determine the stage for the degradation trajectory, identify the changes in degradation rate, divide the degradation process into steady-state degradation stage, accelerated degradation stage and critical failure stage, and at the same time perform correlation matching between the current trajectory features and the pre-stored degradation mode library. Step 5: Select the corresponding lifetime prediction model parameters according to the identified degradation mode type. Based on the probabilistic prediction method, extrapolate the current degradation trajectory to the failure threshold crossing time to calculate the point estimate and confidence interval of the remaining lifetime. Then, dynamically adjust the prediction model parameters according to the duration and rate of degradation.
[0008] Preferably, the alignment process in step two adopts a piecewise cubic interpolation algorithm, the time window width is set according to the period of the highest sampling frequency, outlier removal is based on the statistical process control principle, and missing data is filled by a weighted moving average method based on adjacent period data.
[0009] Preferably, the normalization process in step three uses a linear mapping method, in which the contact resistance weight coefficient is higher than the temperature and insulation resistance weight coefficients, and the partial discharge weight coefficient is the lowest.
[0010] Preferably, in step four, the length of the analysis window is set to a preset number of time units. Within each analysis window, a linear regression algorithm is used to calculate the trajectory slope. The standard deviation of the slope is used to determine the steady-state degradation stage, and the slope increase is used to determine the accelerated degradation stage.
[0011] Preferably, in step five, the lifetime prediction model adopts the Weibull distribution probability prediction method to calculate the remaining lifetime based on the scale parameters and shape parameters under different degradation modes, and the output includes point estimates, failure probability distribution curves, and recommended maintenance time windows.
[0012] On the other hand, a fault prediction system for a smart power distribution equipment connector includes a processing link, the processing link comprising: Acquire multi-dimensional operating condition data from connector monitoring points, the multi-dimensional operating condition data including thermal and electrical characteristics, and apply logical condition constraints based on sampling frequency differences to the acquired data; Interpolation is performed on a unified discrete-time grid to determine the timing alignment between data with different sampling frequencies and eliminate phase deviations between data. Normalization is performed on the time-aligned multidimensional feature data to map parameters with different dimensions to a preset numerical range. The weight allocation is determined according to the sensitivity of each parameter to the degradation process, and a weighted fusion operation is performed to generate a continuous degradation trajectory curve. Qualitative judgment criteria based on the processing results output and the state of the connector in the industrial field include degradation stage attribution, degradation mode type matching, and remaining lifetime prediction.
[0013] Preferably, when performing normalization processing on the multidimensional feature data, a linear mapping method is used, and the weight allocation is determined based on the sensitivity of contact resistance, temperature, insulation resistance and partial discharge parameters. The fused health indicator sequence is arranged in time to generate a degradation trajectory curve.
[0014] Preferably, when performing degradation pattern recognition, a sliding window slope calculation is performed, and the degradation process is divided into three stages: steady state, acceleration, and critical state based on the slope change. At the same time, the correlation coefficient between the current trajectory and each benchmark pattern in the pre-stored degradation pattern library is calculated to determine the pattern attribution.
[0015] Preferably, when performing remaining lifetime prediction, the Weibull distribution prediction model with corresponding parameters is called according to the identified degradation mode type, the remaining lifetime is calculated based on the cross-pass time extrapolated from the current degradation trajectory to the failure threshold, and the point estimate, failure probability distribution and maintenance time window suggestion are output.
[0016] Preferably, it also includes triggering graded early warnings based on the remaining life prediction value and failure probability, with different early warning levels corresponding to different operation and maintenance response processes, and feeding back the early warning status to the data acquisition stage, automatically adjusting the sampling frequency under high-risk conditions.
[0017] The beneficial effects of this invention include: 1. By synchronously acquiring multi-dimensional feature signals and performing time-series alignment processing, the ability to coordinate multi-parameter monitoring can be improved. Multi-dimensional feature data such as temperature, contact resistance, insulation resistance, and partial discharge are acquired synchronously through heterogeneous sensor networks. Time-series alignment processing is performed based on the piecewise cubic Hermite interpolation algorithm to eliminate phase deviations introduced by sampling rate differences. This solves the technical problem in related technologies where monitoring parameters are isolated from each other and cannot characterize the time-series correlation between parameters, providing a complete and consistent data foundation for the systematic tracking of degradation trajectories.
[0018] 2. By constructing degradation trajectories and identifying degradation patterns, the entire degradation process can be tracked and patterns can be identified. Multidimensional features are fused into a unified health index using a weighted geometric average method. Based on sliding window slope analysis, the degradation process is divided into three stages: steady state, acceleration, and critical state. Correlation matching is performed with a pre-stored degradation pattern library. This solves the technical problem in existing technologies that only focus on whether the current state is normal and cannot identify abnormal degradation trends. It enables systematic tracking and pattern recognition of connector degradation trajectories.
[0019] 3. The remaining lifetime prediction step can enable early prediction of the failure time. Based on the Weibull distribution probability prediction method, the corresponding parameters are selected according to the identified degradation mode type. The point estimate and confidence interval of the remaining lifetime are calculated by trajectory extrapolation. This can solve the technical problem that it can only detect the failure that has occurred but cannot predict the failure that is about to occur. This allows maintenance personnel to obtain a sufficient response time window before the failure occurs, which is conducive to improving the reliability of power distribution equipment operation.
[0020] 4. The early warning feedback module enables closed-loop linkage between operation and maintenance response and data acquisition. Based on the remaining life prediction results and failure probability, it automatically triggers three-level early warnings. Differentiated early warning levels correspond to differentiated operation and maintenance response processes. Under high-risk conditions, the sampling frequency is automatically increased to obtain more refined status data. This can solve the passive operation and maintenance dilemma caused by the lack of predictive capabilities in existing technologies. It can realize the paradigm shift from post-fault handling to pre-fault preventive maintenance, which is conducive to reducing the probability of serious accidents such as equipment overheating, partial discharge and fire caused by connector failure. Attached Figure Description
[0021] The invention will now be further described with reference to the accompanying drawings.
[0022] Figure 1 This is a structural block diagram of a fault prediction system for an intelligent power distribution equipment connector according to the present invention; Figure 2 This is a flowchart of a fault prediction method for a smart power distribution equipment connector according to the present invention. Detailed Implementation
[0023] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0024] Example 1: like Figure 1As shown in the embodiment of this application, a fault prediction system for intelligent power distribution equipment connectors includes a complete processing link consisting of a multi-dimensional feature acquisition module, a time sequence alignment processing module, a degradation trajectory construction module, a degradation pattern recognition module, a remaining lifetime prediction module, and an early warning feedback module.
[0025] The multi-dimensional feature acquisition module is used to synchronously acquire multi-dimensional operating condition data, including thermal and electrical characteristics, through a sensor network set at the connector monitoring point, and to construct a multi-dimensional dynamic dataset that reflects the operating status of the connector. The time-series alignment module is used to perform soft alignment operations based on time windows on the cube, mapping the original data with different sampling frequencies to a unified discrete time grid, using interpolation algorithms to eliminate phase deviations introduced by sampling rate differences, and performing outlier removal and missing data imputation. The degradation trajectory construction module is used to normalize the feature signals of each dimension after time alignment, uniformly map parameters with different dimensions to a preset value range, determine the weight allocation based on the sensitivity of each parameter to the connector degradation process, and generate a continuous degradation trajectory curve through weighted fusion operation. The degradation pattern recognition module is used to calculate the sliding window slope and determine the stage of degradation trajectory, identify the degradation rate change, divide the degradation process into steady-state degradation stage, accelerated degradation stage and critical failure stage, and match the current trajectory features with the pre-stored degradation pattern library. The remaining lifetime prediction module is used to select the corresponding lifetime prediction model parameters according to the identified degradation mode type. Based on the probability prediction method, it extrapolates the current degradation trajectory to the failure threshold crossing time, calculates the point estimate and confidence interval of the remaining lifetime, and dynamically corrects the prediction model parameters according to the duration and rate of degradation. The early warning feedback module is electrically connected to the remaining lifetime prediction module and is used to execute graded early warnings based on the prediction results and feed them back to the multi-dimensional feature acquisition module to adjust the sampling strategy.
[0026] This embodiment is specifically applied to the monitoring scenario of copper busbar connectors in high-voltage switchgear in a power distribution room. In this application scenario, the multi-dimensional feature acquisition module performs data acquisition tasks through a heterogeneous sensor network set on the surface of the connector and key monitoring points around it.
[0027] The contact temperature sensor array contains three temperature probes, which are respectively located on the surface of the upper copper busbar of the connector, the terminal area in the middle of the connector, and the lead-out area at the lower end of the connector, to collect the surface temperature parameters of the connector terminals.
[0028] The temperature sensor uses a platinum resistance sensor with an accuracy of ±0.5 degrees Celsius. The sensor probe is tightly attached to the surface being measured by thermally conductive silicone grease to ensure accurate temperature measurement.
[0029] The four-wire contact resistance measurement unit is connected to the connector terminal via four independent wires. Two wires are used to inject the measurement current, and the other two wires are used to acquire the voltage signal. This wiring method can effectively eliminate the interference of lead resistance on the measurement results. The instantaneous contact resistance value output by the measurement unit is accurate to 0.01 milliohms.
[0030] Compared to the conventional two-wire measurement method, the four-wire measurement method can eliminate the superimposed influence of the contact resistance and wire resistance of the test leads themselves on the measurement results, ensuring the authenticity of the contact resistance measurement data.
[0031] The insulation resistance monitoring unit is set to a measurement voltage of 500 volts DC. The measuring end is connected to the insulation part between the connector terminal and the equipment housing. The parameters collected include the current value of the insulation resistance and the absorption ratio after a specified time interval, which are used to characterize the insulation performance parameters between the terminal and the housing.
[0032] The ultra-high frequency partial discharge sensor operates in the frequency band from 300 MHz to 1000 MHz. The sensor probe adopts an arc-shaped metal substrate design to achieve omnidirectional reception of partial discharge electromagnetic waves. The sensor is set on the surface of the insulating support near the connector to collect the characteristic spectrum and pulse intensity of the partial discharge signal generated during the operation of the connector.
[0033] When the aforementioned multi-dimensional feature acquisition module performs data acquisition tasks, the clock signals of all sensors are calibrated by a unified clock synchronization unit. This unified clock synchronization unit uses a rubidium atomic clock as the master clock source and distributes high-precision timestamps to each sensor node through a distributed clock synchronization protocol to ensure strict synchronization of multi-channel data acquisition.
[0034] The sampling frequency is configured differently according to the characteristics of each parameter. The physical change of temperature signal is relatively slow, so the sampling frequency is set to once per minute; the contact resistance is affected by load current fluctuations and changes relatively quickly but is still a slowly changing parameter, so the sampling frequency is set to three times per minute; the insulation resistance parameter changes even more slowly, so the sampling frequency is set to once per minute; the generation of partial discharge signal is random and instantaneous, and a high sampling frequency is the key to capturing discharge events, so the sampling frequency is set to 100 times per second.
[0035] All acquired data is appended with a high-precision timestamp, with a timestamp resolution set to 10 milliseconds. The data is transmitted in real time to the back-end data processing platform via an industrial bus interface. The acquisition module also supports remote parameter configuration. Maintenance personnel can adjust the sampling frequency settings of each channel and enable or disable specific acquisition channels through a remote management terminal according to actual monitoring needs.
[0036] The multidimensional dynamic dataset acquired by the multidimensional feature acquisition module is transmitted in real time to the temporal alignment processing module. The temporal alignment processing module integrates a data fusion and quality control operation unit to perform the various processing tasks specified in step two of the independent method claims. The timing alignment processing module performs soft alignment operations based on time windows on the received multidimensional dataset. Since different types of sensors acquire data at their own independent sampling frequencies, the various data points in the acquired multidimensional dataset appear as unaligned discrete point sequences on the time axis. Therefore, timing alignment processing is required to establish a unified time reference. The timing alignment processing module maps the raw data from different sampling frequencies to a unified discrete time grid. The time window width is determined based on the period of the highest sampling frequency. In this embodiment, the highest sampling frequency is 100 times per second for the partial discharge signal, and the corresponding time window width is set to 10 milliseconds. That is, equally spaced discrete time nodes are established on the time axis at 10-millisecond intervals.
[0037] The time-series alignment module employs a piecewise cubic interpolation algorithm to perform interpolation operations. This algorithm constructs a cubic polynomial interpolation function between two adjacent known data points, and uses this interpolation function to calculate the estimated values at each discrete time node. Compared to linear interpolation, the piecewise cubic interpolation algorithm can obtain smoother interpolation results while maintaining computational efficiency, effectively avoiding the Runge phenomenon that may occur in higher-order interpolation. This interpolation algorithm has a significant effect on eliminating phase deviations introduced by sampling rate differences, enabling multidimensional data originally collected according to different time bases to achieve accurate alignment on a unified time grid. During the interpolation process, the timing alignment module also performs outlier removal. Based on the statistical process control principle, the module detects outliers in the data points of each parameter sequence. The specific method is to calculate the mean and standard deviation of each parameter sequence and identify outliers that exceed the adaptive threshold of the sum of the mean and three times the standard deviation.
[0038] The time-series alignment processing module fills in the detected missing data segments using a weighted moving average method based on adjacent period data; The specific data filling process is as follows: When a parameter is found to have missing data at a certain time point, the historical data of the parameter for three periods before and after the missing time point are extracted and assigned weight coefficients of 0.3, 0.2, 0.1, 0.1, 01, and 0.3 respectively. The weighted historical data are summed to obtain the filling value, which is used to fill the missing data at that time point. The design concept of this weighted moving average method is that the closer the historical data is to the missing point, the greater the contribution to the filling value, and the farther away the data is, the smaller the contribution. In this way, the fluctuation of the filling value is effectively smoothed while utilizing historical data information, ensuring the integrity and continuity of time series data. The time-series alignment processing module simultaneously performs data quality assessment, calculating the data integrity rate and confidence index for each parameter sequence. The data integrity rate is defined as the ratio of the number of valid data points to the number of theoretical data points. The confidence index comprehensively considers factors such as data integrity rate, data volatility, and outlier removal ratio. The results of this quality assessment will serve as input references for subsequent analysis modules.
[0039] After the temporal alignment process is completed, the complete and temporally aligned multidimensional feature dataset is sent to the degradation trajectory construction module. The degradation trajectory construction module has a built-in normalization processing unit and a weighted geometric mean calculation unit, which are used to perform the various processing tasks specified in step three of the independent method claim. The degradation trajectory construction module performs normalization processing on the feature signals of each dimension, and maps parameters with different dimensions such as temperature, contact resistance, insulation resistance and partial discharge intensity to a numerical range of 0 to 1. The normalization process uses a linear mapping method, and the specific calculation process includes: To obtain the minimum and maximum values of each parameter within a certain statistical period, taking the contact resistance parameter as an example, the statistical period is set to all contact resistance measurements within the most recent 720 hours. Find the minimum and maximum contact resistance values within this period, and calculate the normalized result of the current contact resistance value according to the mapping formula. The mapping formula is the current value minus the minimum value and then divided by the difference between the maximum and minimum values. This calculation process ensures that the normalized value is within the range of 0 to 1, and the normalized result is 0 when the contact resistance value is equal to the minimum value, and 1 when the contact resistance value is equal to the maximum value.
[0040] After normalization, the degradation trajectory construction module determines the weight allocation based on the sensitivity of each parameter to the connector degradation process. A continuous degradation trajectory curve is generated through weighted fusion calculation. The weight allocation is determined based on the sensitivity analysis results of each parameter to the connector degradation process. The contact resistance parameter directly reflects the conductivity of the connector contact interface, and its value change has a high correlation with the degree of connector degradation; therefore, it is assigned the highest weight coefficient of 0.4. The temperature parameter reflects the thermal characteristics of the connector during operation; increased contact resistance leads to increased heat generation, and abnormal temperature rise is often an accompanying phenomenon of accelerated degradation; therefore, it is assigned a weight coefficient of 0.25. The insulation resistance parameter reflects the insulation performance of the connector; when the insulation material ages or becomes damp, the insulation resistance decreases, which is also an important indicator of connector degradation; therefore, it is assigned a weight coefficient of 0.25. The partial discharge parameter reflects the partial discharge activity inside the connector's insulation medium. In the early stages of degradation, the partial discharge signal may be weak, and its indicative role is relatively less sensitive than the previous three; therefore, it is assigned a weight coefficient of 0.1.
[0041] The degradation trajectory construction module employs a weighted geometric average method to perform multi-dimensional feature fusion calculations. The specific calculation process includes: adding one to the normalized values of each parameter after normalization, taking the logarithm, multiplying each logarithmic value by its corresponding weight coefficient, summing the results, and finally performing an exponential operation on the sum to obtain the fused health index value. This weighted geometric average method ensures that the fusion calculation comprehensively considers the contribution of each parameter to the degradation process, with parameters having higher weight coefficients having a greater impact in the fusion result. The degradation trajectory construction module calculates the health index values at each time point according to the time series, and connects these health index values sequentially in a coordinate system with time as the horizontal axis and health index as the vertical axis to generate a continuous degradation trajectory curve for the connector. This degradation trajectory curve visually presents the gradual degradation process of the connector from its initial healthy state to its failure state, with the health index value gradually decreasing over time, and the change in the rate of decrease reflecting the transition of the degradation stage. The degradation trajectory construction module simultaneously outputs three types of data: the original feature sequence, the normalized feature value sequence, and the fused health index sequence, for use by subsequent analysis modules.
[0042] The degradation trajectory data generated by the degradation trajectory construction module is transmitted to the degradation pattern recognition module. The degradation pattern recognition module integrates a trajectory analysis engine and a pattern matching unit to perform the processing tasks specified in step four of the independent method claim. The degradation pattern recognition module performs in-depth analysis and pattern matching on the received degradation trajectory data. The main processing logic includes two dimensions: degradation stage determination and degradation pattern matching.
[0043] In terms of degradation stage determination, the degradation pattern recognition module identifies the degradation rate change by calculating the trajectory slope; the module divides the degradation trajectory curve into several analysis windows according to the time series, and the window length is set according to the preset number of time units. In this embodiment, the window length is set to 120 time units, each time unit corresponds to 10 minutes, so the time span covered by each analysis window is 20 hours. The degradation pattern recognition module uses a linear regression algorithm to calculate the slope of the degradation trajectory within each analysis window. The specific calculation process includes: forming a data point set by combining all health indicator values and their corresponding time values within the window; fitting an optimal straight line using the least squares method; and the slope of this straight line is the average degradation rate within the window. A positive slope indicates an increase in health indicators, while a negative slope indicates a decrease. Since the degradation process is manifested as a gradual decline in health indicators, the slope value is usually negative, and its absolute value reflects the rate of degradation. The calculation results of the slope value are stored in a slope sequence for use in subsequent stages.
[0044] The degradation pattern recognition module determines the degradation process based on the slope sequence of continuous analysis windows, dividing it into three states: steady-state degradation, accelerated degradation, and critical failure. The steady-state degradation stage is defined as the standard deviation of the slopes of three consecutive analysis windows being less than a preset threshold. During this stage, the degradation rate is relatively stable and slow, and health indicators show a steady downward trend. The accelerated degradation stage is defined as the increase in slope between two consecutive analysis windows exceeding twice the average slope of the steady-state stage. When the degradation process transitions from the steady-state to the accelerated degradation stage, the degradation rate begins to accelerate, and the rate of decline in health indicators gradually increases. The critical failure stage is defined as health indicator values dropping to a preset failure threshold while the degradation rate continues to increase. When the degradation process enters the critical failure stage, the connector is nearing functional failure, and the system needs to issue an early warning to alert maintenance personnel.
[0045] In terms of degradation pattern matching, the degradation pattern library stored inside the degradation pattern recognition module contains feature parameters of three benchmark degradation patterns; The first baseline mode is the poor contact type. The typical characteristics of this mode are that the contact resistance value increases and the temperature parameter increases simultaneously. The degradation trajectory shows a shape that is first flat and then steep. It corresponds to the contact performance degradation caused by oxidation and corrosion of the contact interface or mechanical loosening. The second benchmark mode is the insulation aging type. The typical characteristics of this mode are that the insulation resistance value continues to decrease and the intensity of the partial discharge signal gradually increases. The degradation trajectory shows a progressively accelerating form, which corresponds to the insulation performance deterioration caused by the aging and degradation of the insulation material. The third reference mode is the terminal loosening type. The typical characteristics of this mode are that the contact resistance value fluctuates drastically and is accompanied by obvious temperature fluctuations. The degradation trajectory presents an irregular sawtooth shape, which corresponds to the scenario of unstable contact state caused by loose fasteners.
[0046] The degradation pattern recognition module determines the pattern attribution by calculating the correlation coefficient between the current degradation trajectory and three benchmark patterns. The specific calculation process includes: extracting the slope sequence and fluctuation feature sequence of the current degradation trajectory within the sliding window as the current trajectory feature vector, and calculating the correlation coefficient with the feature vectors of the three benchmark patterns respectively. The correlation coefficient ranges from 0 to 1, with the value closer to 1 indicating a higher degree of matching. The degradation pattern recognition module determines the benchmark pattern with the highest correlation coefficient as the degradation pattern type to which the current degradation trajectory belongs, and outputs the recognition result to the remaining life prediction module. The degradation pattern recognition module also supports remote updates and incremental learning functions for the degradation pattern library. When the system identifies a new degradation trajectory pattern that cannot be matched with an existing benchmark pattern, the trajectory feature is stored in a pending review queue. After the maintenance personnel confirm the fault type, the new trajectory pattern is included in the degradation pattern library as a new benchmark pattern.
[0047] The degradation pattern recognition module outputs the recognition results, which are then transmitted to the remaining lifetime prediction module. The remaining lifetime prediction module performs the processing tasks specified in step five of the independent method claim. Based on the identified degradation pattern type, the remaining lifetime prediction module selects the corresponding lifetime prediction model parameter set. This module has a built-in model parameter storage unit that stores the scale parameters and shape parameters corresponding to different degradation patterns. The scale parameters and shape parameters are two key parameters of the Weibull distribution, representing the average time of failure occurrence and the trend of failure rate changes, respectively.
[0048] When the degradation mode identification module determines that the current degradation mode is poor contact type, the remaining life prediction module calls the parameter set of poor contact type degradation mode, where the scale parameter is set to 800 hours and the shape parameter is set to 2.5; When the degradation mode identification module determines that the current degradation mode is insulation aging, the remaining lifetime prediction module calls the parameter set of the insulation aging degradation mode, where the scale parameter is set to 1200 hours and the shape parameter is set to 3.0. When the degradation mode identification module determines that the current degradation mode is terminal loosening, the remaining life prediction module calls the parameter set of the terminal loosening degradation mode, where the scale parameter is set to 600 hours and the shape parameter is set to 2.0.
[0049] The remaining lifetime prediction module uses the Weibull distribution probability prediction method to perform remaining lifetime prediction calculations. The specific prediction process consists of the following steps: The first step is to extrapolate the current degradation trajectory to the failure threshold and calculate the predicted crossing time by using a linear extrapolation method that extrapolates the current health index value to the preset failure threshold. The second step is to obtain the probability density function expression of the Weibull distribution and calculate the probability density value corresponding to the predicted travel time based on the probability density function of the Weibull distribution. The third step is to calculate the cumulative failure probability corresponding to the predicted travel time based on the cumulative distribution function of the Weibull distribution. The fourth step is to incorporate the current observation data as evidence into the probability model based on the Bayesian update principle, and update the probability distribution of the remaining lifetime. The fifth step is to calculate the point estimate of the remaining lifetime based on the updated probability distribution, and output the remaining lifetime prediction, failure probability distribution curve, and recommended maintenance time window.
[0050] The remaining lifetime prediction module dynamically adjusts the prediction model parameters based on the duration and rate of the degradation phase. When the degradation pattern recognition module determines that the current degradation process is in an accelerated degradation phase, the remaining lifetime prediction module dynamically adjusts the scale parameters based on the duration and rate of the accelerated degradation phase. The specific adjustment strategy is as follows: When the duration of the accelerated degradation phase exceeds 24 hours and the degradation rate exceeds 1.5 times the average degradation rate of the steady-state phase, the scale parameter is multiplied by a correction factor of 0.8; when the duration of the accelerated degradation phase exceeds 48 hours and the degradation rate exceeds 2 times the average degradation rate of the steady-state phase, the scale parameter is multiplied by a correction factor of 0.6. The introduction of a dynamic correction mechanism enables the prediction model to adaptively adjust the prediction results according to the actual degradation trend, effectively improving prediction accuracy. The remaining lifetime prediction module supports a rolling update function for prediction results, automatically refreshing the prediction results after each data collection cycle and sending the updated results to the early warning feedback module.
[0051] The early warning feedback module is electrically connected to the remaining life prediction module. It is used to perform tiered early warning and feedback control based on the prediction results. The early warning feedback module presets three early warning thresholds: Attention Level, Warning Level, and Emergency Level. The trigger condition for a Attention Level early warning is a remaining life prediction value greater than 720 hours and a failure probability less than 10%. At this point, the connector is in good condition, but monitoring for degradation is recommended. The trigger condition for a Warning Level early warning is a remaining life prediction value greater than 168 hours and less than or equal to 720 hours, and a failure probability greater than or equal to 10% and less than 50%. At this point, the connector shows obvious signs of degradation and a maintenance plan needs to be arranged. The trigger condition for an Emergency Level early warning is a remaining life prediction value less than or equal to 168 hours or a failure probability greater than or equal to 50%. At this point, the connector is nearing failure and immediate maintenance measures are required.
[0052] The early warning feedback module pushes differentiated operation and maintenance response process suggestions to operation and maintenance personnel when different early warning levels are triggered. A "Note" level early warning suggests that operation and maintenance personnel monitor the connector's operational status during routine inspections; a "Warning" level early warning suggests that operation and maintenance personnel complete on-site inspections and develop a maintenance plan within 72 hours. The emergency warning recommends that maintenance personnel immediately shut down the system and perform emergency maintenance. The warning information is displayed to maintenance personnel through a visual interface, which simultaneously displays the remaining life prediction curve, failure probability distribution curve, and historical degradation trajectory data, allowing maintenance personnel to fully understand the degradation status of the connector. The warning feedback module also feeds back the warning status to the multi-dimensional feature acquisition module. When the system determines that the connector is in a high-risk state, it automatically sends a sampling frequency adjustment command to the multi-dimensional feature acquisition module, increasing the sampling frequency of the corresponding monitoring point of the connector to twice the original value to obtain more refined status data.
[0053] like Figure 2 As shown, based on the above system, the fault prediction method for intelligent power distribution equipment connectors of the present invention includes the following steps: Step 1: Synchronously collect multi-dimensional operating condition data, including thermal and electrical characteristics, through a sensor network set at the connector monitoring points, and construct a multi-dimensional dynamic dataset reflecting the operating status of the connector.
[0054] In this step, the multi-dimensional feature acquisition module performs data acquisition tasks through a sensor network set at the connector monitoring points. This sensor network consists of various types of heterogeneous sensors used to acquire multi-dimensional operating condition data reflecting the connector's operating status. For thermal characteristic acquisition, a contact temperature sensor array is set in the key heat distribution area of the connector, with the sensor probes in close contact with the measured surface through a heat-conducting medium to acquire surface temperature data of the connector terminals. For electrical characteristic acquisition, a four-wire precision resistance measuring device is connected to the lead wires of the connector terminals to acquire the instantaneous value and change in contact resistance. An insulation resistance monitoring unit is connected to the insulation part between the terminals and the shell to acquire insulation performance parameters between the terminals and the shell. A UHF sensor is set on the surface of the insulating support near the connector to acquire the characteristic spectrum and pulse intensity data of partial discharge signals. The acquisition process of the above multi-dimensional feature data is performed using multi-channel parallel sampling under a unified clock reference. The sampling frequency is configured differently according to the characteristics of each parameter. All acquired data is accompanied by a high-precision timestamp and transmitted to the back-end processing platform in real time to construct a multi-dimensional dynamic dataset reflecting the connector's operating status.
[0055] Step 2: Perform a time window-based soft alignment operation on the multidimensional dataset to map the original data with different sampling frequencies to a unified discrete time grid. Use an interpolation algorithm to eliminate the phase deviation introduced by the sampling rate difference, and at the same time perform outlier removal and missing data imputation.
[0056] In this step, the temporal alignment processing module performs temporal alignment processing on the multidimensional dataset output by the multidimensional feature acquisition module. This processing maps the original data with different sampling frequencies to a unified discrete time grid, and the time window width is set to 10 milliseconds according to the period of the highest sampling frequency. A piecewise cubic interpolation algorithm is used to construct a cubic polynomial interpolation function between two adjacent known data points to calculate the estimated value at each discrete time node, thereby eliminating the phase deviation introduced by the sampling rate difference. During the interpolation operation, this module simultaneously performs outlier removal and missing data imputation. Outlier removal is based on the principle of statistical process control. By calculating the mean and standard deviation of each parameter sequence, numerical points exceeding the adaptive threshold are identified as outliers and removed. Missing data imputation adopts the weighted moving average method based on adjacent periodic data. The imputation value is calculated according to the historical data around the missing point and its weight coefficient, ensuring the integrity and continuity of the time series data and outputting a complete and time-aligned multidimensional feature dataset.
[0057] Step 3: Normalize the feature signals of each dimension after time alignment, map parameters with different dimensions to a preset value range, determine the weight allocation based on the sensitivity of each parameter to the connector degradation process, and generate a continuous degradation trajectory curve through weighted fusion operation.
[0058] In this step, the degradation trajectory construction module performs normalization and feature fusion operations on the multidimensional feature data output by the time-series alignment processing module. The normalization process uses a linear mapping method to uniformly map parameters with different dimensions, such as temperature, contact resistance, insulation resistance, and partial discharge intensity, to a numerical range of 0 to 1. Specifically, the calculation process involves subtracting the minimum value from the current value and then dividing by the difference between the maximum and minimum values. Regarding weight allocation, the degradation trajectory construction module determines the weight allocation based on the sensitivity of each parameter to the connector degradation process, with contact resistance having the highest weight coefficient, followed by temperature and insulation resistance, and partial discharge having the lowest. The weighted geometric average method is used to perform the multidimensional feature fusion operation. The fused health index value is obtained by calculating the weighted geometric average of the normalized parameter values. The health index values at each time point are calculated sequentially according to the time series, generating a continuous degradation trajectory curve.
[0059] Step 4: Calculate the slope of the sliding window and determine the stage for the degradation trajectory, identify the changes in degradation rate, divide the degradation process into steady-state degradation stage, accelerated degradation stage and critical failure stage, and at the same time perform correlation matching between the current trajectory features and the pre-stored degradation pattern library.
[0060] In this step, the degradation pattern recognition module performs in-depth analysis and pattern matching on the degradation trajectory data output by the degradation trajectory construction module. The module divides the degradation trajectory curve into several analysis windows according to the time series, with the window length set to 120 time units. Within each analysis window, a linear regression algorithm is used to calculate the slope value of the degradation trajectory within that window. The slope value represents the average degradation rate within that time period. Based on the slope sequence of the continuous analysis window, the module performs stage determination and divides the degradation process into three states according to the slope change: steady-state degradation stage, accelerated degradation stage, and critical failure stage. The module performs correlation matching between the current degradation trajectory and the pre-stored degradation pattern library. By calculating the correlation coefficient between the current trajectory and each benchmark pattern, the module determines the pattern affiliation and identifies the degradation pattern type to which the current degradation trajectory belongs.
[0061] Step 5: Select the corresponding lifetime prediction model parameters according to the identified degradation mode type. Based on the probabilistic prediction method, extrapolate the current degradation trajectory to the failure threshold crossing time to calculate the point estimate and confidence interval of the remaining lifetime. Then, dynamically adjust the prediction model parameters according to the duration and rate of degradation.
[0062] In this step, the remaining lifetime prediction module performs remaining lifetime prediction calculations based on the recognition results output by the degradation pattern recognition module. The module selects the corresponding lifetime prediction model parameter set according to the identified degradation pattern type: 800 hours for the contact malfunction type and 2.5 for the shape parameter; 1200 hours for the insulation aging type and 3.0 for the shape parameter; and 600 hours for the terminal loosening type and 2.0 for the shape parameter. The prediction calculation is performed using a Weibull distribution probability prediction method. Based on the time extrapolated from the current degradation trajectory to the failure threshold, the module calculates the point estimate and confidence interval of the remaining lifetime, outputting the remaining lifetime prediction value, the failure probability distribution curve, and the suggested maintenance time window. The module dynamically corrects the scale parameters based on the duration and rate of the accelerated degradation phase, achieving adaptive improvement in prediction accuracy.
[0063] Example 2: This embodiment provides an alternative configuration scheme for the system described in Embodiment 1, mainly by making adaptive adjustments to the sensor configuration and sampling strategy of the multi-dimensional feature acquisition module to meet the monitoring needs of different application scenarios.
[0064] In some industrial environments, the installation space for connectors is limited, and traditional contact temperature sensor arrays may not be able to be fully installed. In this embodiment, the multi-dimensional feature acquisition module adopts a configuration scheme that combines infrared temperature measurement technology with contact temperature measurement. The infrared temperature sensor is set about 50 cm directly above the connector, and non-contact measurement of the connector surface temperature is achieved by collecting infrared radiation energy. The viewing angle of the infrared temperature sensor is set to 30 degrees to ensure that the temperature data of the connector body can be collected. In terms of contact resistance measurement, this embodiment adopts a six-wire measurement scheme, adding two shielded wires to the four-wire system to eliminate the influence of electromagnetic interference on the measurement results, improving the measurement accuracy to above 0.01 milliohms.
[0065] In terms of partial discharge signal acquisition, the configuration of the ultra-high frequency sensor in this embodiment has been optimized and adjusted; the operating frequency band of the sensor is set to 500 MHz to 1500 MHz, which is narrower than that in Embodiment 1 to improve the detection sensitivity for specific discharge types; the sensor probe adopts a disc-shaped metal substrate design, and a ring antenna array is set at the edge of the substrate to achieve wide-band reception of partial discharge electromagnetic waves; after the acquired partial discharge signal is processed by bandpass filtering and amplification, pulse sequence analysis and statistical feature extraction are performed, and the extracted feature parameters include discharge frequency, average discharge amplitude, standard deviation of discharge amplitude, and phase distribution characteristics.
[0066] The sampling strategy in this embodiment employs an adaptive adjustment mechanism. The sampling frequency is no longer fixed but dynamically adjusted according to the connector's operating conditions. When the connector's current carrying capacity is less than 50% of the rated load, the sampling frequency of each parameter follows the basic setting. When the current carrying capacity is between 50% and 80% of the rated load, the sampling frequency for temperature and contact resistance is increased to 1.5 times the basic setting. When the current carrying capacity exceeds 80% of the rated load, the sampling frequency for all parameters is increased to twice the basic setting. This adaptive sampling strategy effectively reduces data transmission volume and storage overhead while ensuring monitoring accuracy.
[0067] Regarding the degradation trajectory construction module, the weight coefficients in this embodiment adopt an online adaptive adjustment mechanism. The initial settings of the weight coefficients are consistent with those in Embodiment 1: contact resistance 0.4, temperature 0.25, insulation resistance 0.25, and partial discharge 0.1. The adjustment mechanism is based on the correlation analysis results between each parameter and the actual fault event, and performs a correlation analysis every 72 hours. The specific analysis process is as follows: extract the change sequence of each parameter in the most recent analysis period and perform correlation calculation with the actual fault event to obtain the correlation coefficient of each parameter. When a parameter shows strong fault indication capability (i.e., a high correlation coefficient) in recent fault cases, the weight coefficient of that parameter in feature fusion is automatically increased by 0.05, while the weight coefficients of other parameters are proportionally decreased to ensure that the sum of the weight coefficients remains one. The adaptive adjustment mechanism enables the degradation trajectory construction module to continuously optimize the weight allocation strategy based on actual operating data, thereby improving the accuracy of the degradation trajectory in representing the actual degradation process.
[0068] Regarding the degradation pattern recognition module, the degradation pattern library in this embodiment is dynamically updated using an incremental learning mechanism. The initial state of the degradation pattern library includes the three benchmark degradation patterns described in Embodiment 1. When the degradation pattern recognition module identifies a new degradation trajectory pattern and the correlation coefficients with the existing benchmark patterns are all below 0.6, the system stores the trajectory feature in the pending review queue and sends a new trajectory discovery notification to the operation and maintenance personnel. Operations and maintenance personnel can review and confirm new trajectory features through the management interface and specify the fault type corresponding to the trajectory. After the review and confirmation are completed, the system will include the trajectory shape and its corresponding fault type into the degradation pattern library as a new baseline pattern, and subsequent identification tasks can match this type of degradation pattern.
[0069] Regarding the remaining life prediction module, the failure threshold in this embodiment adopts a dynamic adjustment strategy. The initial setting of the failure threshold is a health index value of 0.2. When the corresponding health index drops to 20% of the initial value, it is judged as a critical failure state. The dynamic adjustment strategy corrects the failure threshold according to the actual operating conditions of the connector. The specific correction coefficient is related to three factors: the load rate of the connector's current carrying capacity, the temperature level of the operating environment, and the historical failure rate. When the load rate exceeds 80%, the failure threshold correction factor is set to 0.95; when the operating ambient temperature exceeds 40 degrees Celsius, the failure threshold correction factor is set to 0.9; when the historical failure rate is higher than the average of similar equipment, the failure threshold correction factor is set to 0.85; the final failure threshold correction factor is obtained by multiplying the various correction factors together, and then multiplied by the initial failure threshold to obtain the current failure threshold; the dynamic adjustment strategy ensures that the failure threshold setting meets the actual needs of different application scenarios and avoids inaccurate warnings due to differences in operating conditions.
[0070] Comparative Example 1: Comparative Example 1 provides a smart power distribution equipment connector monitoring system using existing technology. This system does not have the multi-dimensional feature time-series alignment processing, degradation trajectory construction, degradation pattern recognition, and remaining lifetime prediction functions disclosed in this invention. By comparing and analyzing Comparative Example 1 with Embodiment 1 of this invention, the technical effects and technological progress of this invention can be clearly demonstrated.
[0071] The existing technology system in Comparative Example 1 uses a fault judgment method based on a single parameter threshold comparison. In terms of data acquisition, this system only sets up a contact temperature sensor and a current transformer to collect two parameters: the surface temperature of the connector terminal and the current passing through it. The sampling frequency is uniformly set to once per minute, and it does not have the ability to collect multi-dimensional features synchronously. Regarding fault diagnosis, this system presets temperature and current thresholds. An alarm is triggered when the collected temperature value exceeds the temperature threshold or the current value exceeds the current threshold. This system does not have a time-series alignment function, and there is a lack of correlation analysis between the collected data at different times; it does not have a degradation trajectory construction function, and cannot generate a continuous curve reflecting the degradation process of the connector; it does not have a degradation mode recognition function, and cannot determine what degradation stage or degradation mode the connector is in; it does not have a remaining life prediction function, and can only detect temperature anomalies that have occurred but cannot predict future degradation trends and remaining life.
[0072] In the actual operation of Comparative Example 1, when the connector is in the process of gradual degradation, the temperature parameter in the early stage of degradation is always lower than the preset threshold, and the system will not generate any alarm prompts. Only when the degradation accumulates to the critical state, causing the temperature to rise sharply and exceed the threshold, will the system trigger an alarm. However, at this time, the connector has often been severely overheated or even shown signs of burning, and the maintenance personnel do not have enough time to respond and deal with it. This Comparative Example 1 system can only determine whether the temperature is abnormal but cannot identify the specific cause of the abnormality. When the temperature abnormality is caused by different reasons, the system gives the same handling suggestions, which can easily lead to the misallocation of maintenance resources.
[0073] Compared with the system in Comparative Example 1, the fault prediction method and system for intelligent power distribution equipment connectors of the present invention can improve the multi-parameter collaborative monitoring capability through the steps of synchronous acquisition of multi-dimensional feature signals and time alignment processing. It can synchronously acquire multi-dimensional feature data such as temperature, contact resistance, insulation resistance and partial discharge through heterogeneous sensor network, and perform time alignment processing based on piecewise cubic interpolation algorithm to eliminate phase deviation introduced by sampling rate difference, providing a complete and consistent data foundation for the systematic tracking of degradation trajectory. By constructing degradation trajectories and identifying degradation patterns, the entire degradation process can be tracked and patterns can be identified. A weighted geometric average method is used to fuse multidimensional features into a unified health indicator. Based on sliding window slope analysis, the degradation process is divided into three stages: steady state, acceleration, and critical. This addresses the technical problem of existing technologies that only focus on whether the current state is normal and cannot identify abnormal degradation trends. The remaining lifetime prediction step enables early prediction of failure time. Based on a Weibull distribution-based probability prediction method, corresponding parameters are selected according to the identified degradation pattern type. Point estimates and confidence intervals of remaining lifetime are calculated through trajectory extrapolation, allowing maintenance personnel to obtain sufficient response time windows before failures occur. The early warning feedback module enables closed-loop linkage between operation and maintenance response and data collection. It automatically triggers three-level early warnings based on the remaining life prediction results and failure probability. Differentiated early warning levels correspond to differentiated operation and maintenance response processes, which can solve the passive operation and maintenance dilemma caused by the lack of predictive capabilities of existing technologies. It can realize the paradigm shift from post-fault handling to pre-fault preventive maintenance.
[0074] Table 1 is a comparison table between the embodiments of the present invention and Comparative Example 1: 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 illustrative of the principles of 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 present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A fault prediction method for intelligent power distribution equipment connectors, applied to power distribution equipment connectors, characterized in that, Includes the following steps: Step 1: Synchronously collect multi-dimensional operating condition data, including thermal and electrical characteristics, through a sensor network set at the connector monitoring points, and construct a multi-dimensional dynamic dataset reflecting the operating status of the connector; Step 2: Perform a time window-based soft alignment operation on the multidimensional dataset to map the original data with different sampling frequencies to a unified discrete time grid. Use an interpolation algorithm to eliminate the phase deviation introduced by the sampling rate difference, and at the same time perform outlier removal and missing data imputation. Step 3: Normalize the feature signals of each dimension after time alignment, map parameters with different dimensions to a preset value range, determine the weight allocation based on the sensitivity of each parameter to the connector degradation process, and generate a continuous degradation trajectory curve through weighted fusion operation. Step 4: Calculate the slope of the sliding window and determine the stage for the degradation trajectory, identify the changes in degradation rate, divide the degradation process into steady-state degradation stage, accelerated degradation stage and critical failure stage, and at the same time perform correlation matching between the current trajectory features and the pre-stored degradation mode library. Step 5: Select the corresponding lifetime prediction model parameters according to the identified degradation mode type. Based on the probabilistic prediction method, extrapolate the current degradation trajectory to the failure threshold crossing time to calculate the point estimate and confidence interval of the remaining lifetime. Then, dynamically adjust the prediction model parameters according to the duration and rate of degradation.
2. The fault prediction method for a smart power distribution equipment connector according to claim 1, characterized in that, In step two, the alignment process uses a piecewise cubic interpolation algorithm. The time window width is set according to the period of the highest sampling frequency. Outlier removal is based on the statistical process control principle. Missing data is filled using a weighted moving average method based on adjacent period data.
3. The fault prediction method for a smart power distribution equipment connector according to claim 1, characterized in that, In step three, the normalization process uses a linear mapping method. In the weight allocation, the contact resistance weight coefficient is higher than the temperature and insulation resistance weight coefficients, and the partial discharge weight coefficient is the lowest.
4. The fault prediction method for a smart power distribution equipment connector according to claim 1, wherein, In step four, the length of the analysis window is set to a preset number of time units. Within each analysis window, a linear regression algorithm is used to calculate the trajectory slope. The standard deviation of the slope is used to determine the steady-state degradation stage, and the slope increase is used to determine the accelerated degradation stage.
5. The fault prediction method for a smart power distribution equipment connector according to claim 1, characterized in that, In step five, the lifetime prediction model uses the Weibull distribution probability prediction method to calculate the remaining lifetime based on the scale and shape parameters under different degradation modes. The output includes point estimates, failure probability distribution curves, and recommended maintenance time windows.
6. A fault prediction system for intelligent power distribution equipment connectors, applicable to the fault prediction method for intelligent power distribution equipment connectors as described in any one of claims 1-5, characterized in that, This includes a processing link, which includes: Acquire multi-dimensional operating condition data from connector monitoring points, the multi-dimensional operating condition data including thermal and electrical characteristics, and apply logical condition constraints based on sampling frequency differences to the acquired data; Interpolation is performed on a unified discrete-time grid to determine the timing alignment between data with different sampling frequencies and eliminate phase deviations between data. Normalization is performed on the time-aligned multidimensional feature data to map parameters with different dimensions to a preset numerical range. The weight allocation is determined according to the sensitivity of each parameter to the degradation process, and a weighted fusion operation is performed to generate a continuous degradation trajectory curve. Qualitative judgment criteria based on the processing results output and the state of the connector in the industrial field include degradation stage attribution, degradation mode type matching, and remaining lifetime prediction.
7. The fault prediction system for an intelligent power distribution equipment connector according to claim 6, characterized in that, When performing normalization processing on the multidimensional feature data, a linear mapping method is used, and the weight allocation is determined based on the sensitivity of contact resistance, temperature, insulation resistance and partial discharge parameters. The fused health index sequence is arranged in time to generate a degradation trajectory curve.
8. The fault prediction system for an intelligent power distribution equipment connector according to claim 6, characterized in that, When performing degradation pattern recognition, the sliding window slope is calculated, and the degradation process is divided into three stages: steady state, acceleration and critical state, based on the slope change. At the same time, the correlation coefficient between the current trajectory and each benchmark pattern in the pre-stored degradation pattern library is calculated to determine the pattern attribution.
9. A fault prediction system for an intelligent power distribution equipment connector according to claim 6, characterized in that, When performing remaining lifetime prediction, the Weibull distribution prediction model with corresponding parameters is called according to the identified degradation mode type. The remaining lifetime is calculated based on the cross-time extrapolated from the current degradation trajectory to the failure threshold, and the point estimate, failure probability distribution and maintenance time window suggestion are output.
10. A fault prediction system for an intelligent power distribution equipment connector according to claim 6, characterized in that, It also includes triggering graded early warnings based on the remaining life prediction and failure probability, with different early warning levels corresponding to different operation and maintenance response processes, and feeding back the early warning status to the data collection stage, automatically adjusting the sampling frequency under high-risk conditions.
