Charging gun short circuit fault detection method

Abstract

The invention relates to the technical field of electric vehicle charging, in particular to a charging gun short-circuit fault detection method, and aims to solve the problems that in the prior art, charging gun short-circuit fault detection mainly depends on two technical paths of current threshold triggering and single parameter monitoring, when it is detected that the current exceeds a preset value, a circuit is cut off, detection hysteresis exists, and the detection time is shortened. And pre-judgment cannot be carried out before short circuit occurs, so that the equipment is damaged before the protection action is executed. According to the method, the digital twinning pre-diagnosis model is constructed by fusing multiple physical quantity data, early warning and graded protection of the short-circuit fault of the charging gun are realized by combining multi-dimensional feature fusion and a dynamic risk assessment mechanism, and the method has the advantages of improving the detection timeliness, reducing the misjudgment and missed judgment rate and enhancing the environmental adaptability.

Description

Technical Field

[0001] This application relates to the field of electric vehicle charging technology, and more specifically, to a method for detecting short-circuit faults in charging guns. Background Technology

[0002] Current short-circuit fault detection methods for charging guns face the following technical bottlenecks: Traditional detection methods employ a "current threshold triggering" mechanism, a passive protection approach with significant detection lag. By the time the short-circuit current reaches the preset protection threshold, the fault has already occurred and damaged the equipment, making fault warning impossible. More seriously, within the 50-100ms time window from the generation of the short-circuit current to the triggering of protection, excessively large short-circuit currents may have already burned out the internal terminals of the charging gun, damaged the cable insulation, or even caused overcurrent damage to the battery pack.

[0003] Existing technologies generally suffer from the problem of relying on a single parameter for monitoring, depending solely on the charging current. This leads to a high rate of false positives and false negatives. In practical applications, instantaneous current fluctuations caused by cable voltage drops during charging are easily misdiagnosed as short circuits rather than actual short circuit faults. On the other hand, for progressive faults such as "micro-short circuits" (e.g., increased local resistance and slow current rise due to terminal oxidation), single current monitoring methods are completely ineffective, often missing the opportunity for early intervention and ultimately evolving into serious short circuit accidents.

[0004] Furthermore, existing detection solutions have poor environmental adaptability and struggle to handle complex industrial application scenarios. Charging guns need to operate in a wide temperature range from -30℃ to 60℃, as well as in harsh conditions such as humidity and dust. Traditional detection circuits have a temperature drift coefficient as high as ±5% / ℃, causing the actual current threshold value to drift with temperature changes: protection delays occur in low-temperature environments, while false triggering is common in high-temperature environments, severely restricting the reliable application of charging piles in diverse scenarios such as outdoor use and underground parking garages.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] (a) Technical problems to be solved To address the aforementioned issues, this invention proposes a method for detecting short-circuit faults in charging guns. This method aims to solve the problem that existing technologies for detecting short-circuit faults in charging guns mainly rely on two technical paths: current threshold triggering and single-parameter monitoring. When the detected current exceeds a preset value, the circuit is cut off, resulting in detection lag and an inability to predict short circuits before they occur, leading to damage to the equipment before the protection action is executed.

[0007] (II) Technical Solution The present invention discloses a method for detecting short-circuit faults in a charging gun, the technical solution of which is as follows: Real-time acquisition of multi-physical quantity data during the operation of the charging gun, including current data, contact resistance data, temperature distribution data, and voltage change rate data; processing of the multi-physical quantity data to obtain fault feature data; inputting the fault feature data into a digital twin pre-diagnosis model; calculating the probability of a short-circuit fault occurring in the charging gun using the digital twin pre-diagnosis model; the digital twin pre-diagnosis model integrating a data fusion algorithm and a fault evolution model; determining the risk level based on the probability of the short-circuit fault occurring; and implementing corresponding protection measures based on the risk level.

[0008] Furthermore, this application also proposes that the real-time acquisition of multiple physical quantity data during the operation of the charging gun includes: acquiring current data through a current sensor at a sampling frequency of not less than 10kHz to identify precursors of current spikes; detecting the voltage at the charging gun terminals in real time through a voltage sampling chip and calculating the contact resistance value in combination with the current data; acquiring cable temperature distribution data through distributed temperature sensors arranged at preset intervals along the charging gun cable; and detecting the instantaneous rate of change of the charging gun output voltage through a voltage fluctuation detection chip.

[0009] Furthermore, this application also proposes that the process of processing multi-physical quantity data to obtain fault feature data includes: preprocessing different types of multi-physical quantity data, unifying the dimensions, extracting key features that can characterize the precursors of short-circuit faults in charging guns from the preprocessed multi-physical quantity data, forming a feature set containing multi-dimensional features; and using a multi-evidence body synthesis algorithm to fuse and calculate the extracted key features to obtain fault feature data that can be directly input into the digital twin pre-diagnosis model.

[0010] Furthermore, this application also proposes that preprocessing of the different types of multi-physical quantity data includes: The current data is denoised using sliding window filtering and Kalman filtering to control the fluctuation deviation within a certain range. Within; adopt After removing outliers from the contact resistance data, the data is smoothed using an exponentially weighted moving average algorithm to eliminate resistance value jumps caused by transient interference. Linear spatial interpolation is used to supplement the temperature distribution data for the sensor interval regions, and compensation is made based on the ambient temperature. Wavelet-based three-level decomposition combined with soft thresholding is used to denoise the voltage change rate data, eliminating spurious fluctuations caused by power grid interference; all preprocessed multi-physical quantity data are normalized to... Range, unified data format.

[0011] Furthermore, this application proposes to extract three types of features from current data: current peak frequency, current rise rate, and current fluctuation amplitude; two types of features from contact resistance data: resistance mean drift and resistance change trend slope; three types of features from temperature distribution data: local maximum temperature, temperature gradient difference, and proportion of high-temperature areas; and two types of features from voltage change rate data: peak instantaneous rate of change of voltage and duration of voltage instantaneous rate of change anomaly; and to summarize the above features to form a feature set.

[0012] Furthermore, this application proposes dividing the key features into multiple evidence bodies, including evidence body one, evidence body two, and evidence body three. Evidence body one includes the current peak frequency, current rise rate, peak instantaneous rate of change of voltage, and abnormal duration of the instantaneous rate of change of voltage; evidence body two includes the mean resistance drift, slope of the resistance change trend, local maximum temperature, and temperature gradient difference; evidence body three includes the current fluctuation amplitude and the proportion of high-temperature regions. The confidence level of each evidence body for the short-circuit anomaly is calculated using DS evidence theory, and the evidence is synthesized using an evidence synthesis formula. The overall confidence level is obtained, where, This indicates the overall confidence level after integrating the evidence. , , The confidence levels of evidence subjects 1, 2, and 3 respectively for the proposition that the charging gun has a short circuit abnormality; when the overall confidence level is ≥0.8, the corresponding key features are combined with the overall confidence level to form fault feature data; when the overall confidence level is <0.5, it is judged as normal data and no standardized fault feature data is generated.

[0013] Furthermore, this application proposes that the output of the digital twin pre-diagnostic model includes two parts: the first part is the probability of a short circuit fault, specifically the probability value of a short circuit occurring in the charging gun in the future. The calculation accuracy is based on historical data training and optimization of three typical faults: loose terminals, cable wear, and water ingress short circuits; the second part is the probability confidence level, which characterizes the reliability of the probability value. When the probability confidence level is <0.8, the latest fault feature data needs to be re-entered for secondary calculation.

[0014] Furthermore, this application proposes that the risk level be determined based on the probability of a short-circuit fault, including: normal state, short-circuit fault occurrence probability < 50%, and probability confidence level ≥ 0.8; low-risk state, 50% ≤ short-circuit fault occurrence probability < 80%, and probability confidence level ≥ 0.8; high-risk state, 80% ≤ short-circuit fault occurrence probability < 95%, and probability confidence level ≥ 0.8; emergency state, short-circuit fault occurrence probability ≥ 95%, and probability confidence level ≥ 0.8; if the probability confidence level < 0.8, it is temporarily designated as a pending confirmation state, and no protection measures are implemented, only data re-collection is triggered.

[0015] Furthermore, this application also proposes a computing device, comprising: at least one processor; a memory storing executable instructions; wherein when the instructions are executed by the at least one processor, the at least one processor implements the above-described charging gun short-circuit fault detection method.

[0016] Furthermore, this application also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described charging gun short-circuit fault detection method.

[0017] (III) Beneficial Effects Compared with the prior art, the beneficial effects of the present invention are as follows: (2) In this invention, a digital twin pre-diagnosis model is constructed by integrating multiple physical quantity data. Combined with multi-dimensional feature fusion and dynamic risk assessment mechanism, early warning and graded protection of charging gun short circuit faults are realized. It has the advantages of improving detection timeliness, reducing false judgment and missed judgment rate, and enhancing environmental adaptability.

[0018] (3) This invention solves the problem of lag in traditional detection methods, enabling the identification of fault precursors and triggering early warnings before short-circuit current forms. Collaborative monitoring of multiple physical quantity data significantly improves detection coverage, avoiding missed detections due to the failure of a single parameter. The combination of digital twin models and data fusion algorithms enhances environmental adaptability under complex operating conditions, ensuring stable detection accuracy under different temperature and humidity conditions. The risk-level protection mechanism achieves precise matching of protection strategies, reducing unnecessary charging interruptions while ensuring safety. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of the logic structure of the charging gun short circuit fault detection method Detailed Implementation

[0021] like Figure 1As shown in the accompanying drawings, the technical solutions of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] Short-circuit fault detection in charging guns primarily relies on two technical approaches: current threshold triggering and single-parameter monitoring. Traditional methods trigger protection actions by setting a fixed current threshold; when the detected current exceeds the preset value, the circuit is cut off. However, this mode suffers from detection lag, failing to predict short circuits before they occur, leading to damage to the equipment before the protection action is executed. Furthermore, single-parameter monitoring schemes struggle to distinguish between genuine short circuits and transient interference, easily resulting in false alarms or missed alarms under complex operating conditions. For example, in extreme temperature environments, sensor drift can cause the threshold to deviate from the set value, resulting in protection delays at low temperatures or false triggers at high temperatures.

[0023] To address these issues, the inventors first analyzed various causes of short-circuit faults in charging guns, including terminal oxidation, cable wear, and environmental temperature variations. The study revealed that a single current parameter cannot fully reflect the precursory characteristics of different fault modes, and changes in parameters such as contact resistance and temperature distribution often precede the occurrence of current anomalies. Based on this, the inventors proposed capturing early fault signals through the fusion of multi-physical quantity data and combining this with a dynamic prediction model to achieve risk assessment.

[0024] Furthermore, to address environmental interference issues, digital twin technology is introduced to construct a real-time interaction mechanism between physical entities and virtual models, improving prediction accuracy through fault evolution simulation. This ultimately forms a complete technical path from multi-dimensional data acquisition to dynamic risk grading.

[0025] Therefore, this application proposes a method for detecting short-circuit faults in charging guns, comprising: S100: Real-time acquisition of multiple physical quantity data during the operation of the charging gun, including current data, contact resistance data, temperature distribution data, and voltage change rate data. S200. Process the multi-physical quantity data to obtain fault feature data, input the fault feature data into the digital twin pre-diagnosis model, calculate the probability of short circuit fault in the charging gun through the digital twin pre-diagnosis model, and integrate the data fusion algorithm and fault evolution model into the digital twin pre-diagnosis model. S300: Determine the risk level based on the probability of a short-circuit fault occurring, and implement corresponding protection measures based on the risk level.

[0026] Among them, the multi-physical quantity data refers to four types of parameters synchronously acquired during the operation of the charging gun: current, contact resistance, temperature distribution, and voltage change rate. Current data can be acquired through a high-frequency current sensor to capture instantaneous current spikes; contact resistance data is obtained through voltage sampling and current calculation to detect changes in terminal contact status; temperature distribution data is acquired through a distributed temperature sensor array to identify localized overheating areas in the cable; and voltage change rate data is acquired through a voltage fluctuation detection chip to capture the characteristics of a sudden voltage drop before a short circuit. The combination of these parameters can cover multiple precursor modes of short circuit faults.

[0027] A digital twin pre-diagnostic model is a predictive system that maps the physical state of a charging gun in real time using a virtual model. This model integrates data fusion algorithms to calculate confidence levels for multi-source features, and incorporates a fault evolution algorithm to simulate short-circuit development trends. Specifically, it can be trained using a deep learning framework, learning the mapping relationship between different precursor features and short-circuit probabilities through historical fault data. Its core function is to combine real-time data with fault development patterns to achieve dynamic prediction of fault probabilities.

[0028] Risk level refers to the warning level divided according to the probability of short circuit occurrence. Normal state corresponds to a safe operating condition with low probability and high confidence; low-risk state triggers power reduction operation; high-risk state activates audible and visual alarms; and emergency state directly cuts off the circuit. This classification mechanism can match differentiated protection strategies according to the prediction results, avoiding the limitations of a single protection action.

[0029] Specifically, this method first collects charging gun operation data synchronously using multiple types of sensors to eliminate blind spots in single-parameter monitoring. A current sensor captures instantaneous anomalies through high-frequency sampling, a contact resistance calculation module identifies micro-short-circuit trends caused by terminal oxidation, a distributed temperature sensor array locates overheated areas in the cable insulation layer, and a voltage change rate detection module captures the characteristics of a sudden voltage drop before a short circuit. The collected raw data undergoes noise reduction, compensation, and normalization processing to eliminate the influence of environmental interference.

[0030] The feature extraction stage filters key indicators such as current peak frequency, resistance drift, and local temperature gradient from the preprocessed data to form a multi-dimensional feature set. A data fusion algorithm synthesizes confidence scores for these features, filtering out noise interference from single-source data. Based on the fused feature data and fault evolution patterns, the digital twin model predicts the probability of short circuits in future periods. Finally, risk levels are classified according to probability values ​​and confidence scores, triggering corresponding protection actions.

[0031] Traditional solutions rely solely on current threshold triggering for protection, failing to identify early fault signals such as increased contact resistance or localized overheating. This method, through the fusion of multi-physical quantity data, can simultaneously capture four types of precursory features: abnormal current, poor contact, insulation aging, and sudden voltage drop. The application of a digital twin model overcomes the limitations of static threshold determination, enabling early triggering of protection actions through dynamic probability prediction. Furthermore, the risk level classification mechanism, combining probability values ​​and confidence levels, effectively avoids malfunctions caused by environmental interference.

[0032] This application further proposes a specific implementation scheme for real-time acquisition of multiple physical quantity data during the operation of the charging gun, including acquiring current data through a current sensor at a sampling frequency of not less than 10kHz, detecting the voltage of the charging gun terminals in real time and calculating the contact resistance value through a voltage sampling chip, acquiring cable temperature distribution data through distributed temperature sensors arranged at preset intervals along the charging gun cable, and detecting the instantaneous rate of change of the charging gun output voltage through a voltage fluctuation detection chip.

[0033] Among them, the current sensor collects current data at a sampling frequency of not less than 10kHz, which means that high-frequency sampling technology is used to capture the transient changes of the current signal. Specifically, a Hall effect current sensor can be used in conjunction with a high-speed analog-to-digital converter. This feature can identify the precursor signal of current spikes.

[0034] Real-time voltage sampling chip detection of charging gun terminal voltage refers to continuous measurement of the voltage at the charging gun terminals. This can be achieved using a high-precision differential voltage sampling chip. This feature, combined with current data, allows for dynamic monitoring of changes in contact resistance.

[0035] Distributed temperature sensors are arranged at preset intervals along the charging gun cable, which means that multiple temperature measurement points are set along the cable axis. Specifically, digital temperature sensors can be welded at equal intervals inside the cable sheath. This feature can obtain the spatial distribution characteristics of the temperature field along the cable axis.

[0036] The instantaneous rate of change of the output voltage detected by the voltage fluctuation detection chip refers to the real-time calculation of the time derivative of the voltage signal. Specifically, it can be implemented using a dedicated integrated circuit with differential operation function. This feature can capture the precursors of voltage sudden changes caused by insulation degradation.

[0037] Specifically, the high-frequency sampling capability of the current sensor can cover the current fluctuation characteristics during the micro-short circuit induction stage; the joint calculation of voltage sampling chip and current data can dynamically reflect the abnormal drift of terminal contact resistance; the spatial arrangement scheme of distributed temperature sensor can identify the temperature gradient change in the local overheated area of ​​the cable; and the instantaneous change rate monitoring of voltage fluctuation detection chip can detect the trend of insulation degradation in advance before current anomalies.

[0038] The four types of physical quantity acquisition devices form a progressive monitoring chain in the time dimension, construct a multi-point temperature field distribution model in the spatial dimension, and achieve cross-verification of electro-thermal multi-physics fields in the parameter dimension, thereby breaking through the limitations of single current parameter monitoring.

[0039] Existing charging gun short-circuit detection mainly relies on current monitoring with a fixed threshold, while this solution achieves fault precursor identification through a multi-parameter collaborative acquisition mechanism. Current technologies cannot correlate contact resistance and temperature distribution data, but this solution constructs a fault evolution model through joint voltage-current calculations and distributed temperature measurement.

[0040] Existing technologies using single-point temperature monitoring are prone to missing localized overheating, while this solution achieves complete monitoring of the cable's axial temperature field by arranging sensors at preset intervals. Existing technologies lack specific detection of voltage change rate, while this solution uses instantaneous change rate analysis to detect insulation degradation characteristics in advance.

[0041] This application effectively reduces the probability of false positives and false negatives, and can identify latent fault precursors such as increased contact resistance and localized overheating, avoiding misjudging current fluctuations caused by power grid interference as short-circuit faults. It also achieves full-cycle monitoring of the fault evolution process, triggering an early warning signal 500 milliseconds before the short-circuit current reaches the protection threshold, providing sufficient response time for protection measures. Furthermore, the combination of spatial distribution characteristics and time-series characteristics of multi-physical quantity data enhances detection stability in complex industrial environments and solves the threshold drift problem of traditional solutions under extreme temperatures.

[0042] This application further proposes a method for processing multi-physical quantity data to obtain fault feature data, including preprocessing different types of multi-physical quantity data and unifying their dimensions, extracting key features characterizing the precursors of short-circuit faults in charging guns from the preprocessed data to form a feature set, and using a multi-evidence body synthesis algorithm to fuse and calculate the key features to obtain fault feature data.

[0043] Preprocessing refers to denoising and format standardization of current, contact resistance, temperature distribution, and voltage change rate data. Specifically, it can be achieved by using sliding window filtering combined with Kalman filtering to denoise current data, exponential weighted moving average algorithm to smooth contact resistance data, linear spatial interpolation combined with ambient temperature compensation to correct temperature distribution data, and wavelet decomposition combined with soft thresholding to denoise voltage change rate data. This operation eliminates the impact of sensor noise and environmental interference on data quality.

[0044] Key feature extraction refers to screening physical quantity change indicators that are strongly correlated with the short-circuit fault evolution process from the preprocessed data. Specifically, it can be achieved by using ten types of feature indicators such as current peak frequency, average resistance drift, local maximum temperature, and voltage change rate peak. This operation constructs a multi-dimensional feature set covering circuit dynamic anomalies and equipment condition deterioration.

[0045] The multi-evidence body synthesis algorithm refers to a data fusion method that divides features into different evidence bodies and calculates the comprehensive confidence level. Specifically, it can be implemented by using the DS evidence theory to perform orthogonal sum operations on three types of evidence bodies: current dynamics, equipment status, and global risk. This operation improves the credibility of feature data through a cross-validation mechanism.

[0046] Specifically, after the current data is filtered by a sliding window to eliminate high-frequency noise, the actual current value is predicted using a Kalman filter, keeping fluctuations within a set range. Contact resistance data, after outlier removal, is used to generate a smooth curve using an exponentially weighted moving average algorithm, eliminating jumps caused by momentary poor contact. Temperature distribution data is supplemented with temperature values ​​in the sensor interval areas through linear interpolation, and measurement errors are corrected using an ambient temperature compensation coefficient.

[0047] After removing grid fluctuation noise from the voltage change rate data using wavelet decomposition, the effective fluctuation signal is extracted. All preprocessed data are normalized to a unified interval to form a standardized input. When extracting ten key features from the preprocessed data, the current peak frequency is calculated by counting the number of fluctuations exceeding the threshold per unit time, the mean resistance drift is obtained by comparing the difference between the current mean and the initial normal value, and the local maximum temperature is determined by traversing the maximum value in the temperature distribution matrix.

[0048] In the process of synthesizing multiple evidence bodies, the current dynamic evidence body includes the current peak frequency and the peak voltage change rate; the equipment status evidence body includes the average resistance drift and the local maximum temperature; and the global risk evidence body includes the current fluctuation amplitude and the proportion of high temperature area. The confidence level of each evidence body for short circuit anomaly is calculated by DS evidence theory, and finally the comprehensive confidence level is synthesized to determine the validity of fault feature data.

[0049] Traditional methods only perform simple filtering on single physical quantities without establishing a correlation mechanism between multi-dimensional data, resulting in high false positive and false negative rates. This solution retains effective fault characteristics through differentiated preprocessing, extracts key indicators based on fault evolution mechanisms, and uses a classification evidence fusion mechanism to achieve cross-validation of multi-source data, thus solving the problem of the one-sidedness of single-parameter monitoring.

[0050] This application effectively reduces the false positives and false negatives in short-circuit fault detection of charging guns. Through multi-dimensional feature extraction and fusion verification mechanisms, it can accurately identify early fault precursors such as micro-short circuits and local overheating, providing highly reliable input data for digital twin pre-diagnosis models and significantly improving the accuracy of short-circuit probability calculation.

[0051] This application further proposes methods for preprocessing different types of multi-physical quantity data, including: noise reduction of current data using sliding window filtering and Kalman filtering to control current data fluctuation deviation within ±0.1A; smoothing of contact resistance data by removing outliers using criteria and then applying an exponentially weighted moving average algorithm; compensation of temperature distribution data by using linear spatial interpolation to supplement temperature values ​​in the sensor interval area and combining it with ambient temperature; noise reduction of voltage change rate data by using wavelet basis for three-level decomposition combined with soft thresholding; and normalization of all preprocessed multi-physical quantity data to... The intervals use a uniform data format.

[0052] Sliding window filtering refers to local smoothing of current data by setting a data window of fixed length. Specifically, it can be implemented using a moving average algorithm with a window length of 50 sampling points to suppress high-frequency noise interference.

[0053] Kalman filtering refers to dynamic error correction of current data based on a state-space model. Specifically, it can be implemented using a linear Kalman filter to eliminate dynamic measurement errors of sensors.

[0054] Outlier removal criteria involve identifying and removing outliers from contact resistance data based on statistical rules. Specifically, the 3σ criterion can be used to eliminate false resistance jumps caused by momentary poor terminal contact. The exponentially weighted moving average algorithm performs weighted smoothing on time-series data. This can be achieved using an exponential weighting calculation with an attenuation factor of 0.2, preserving the trend of resistance changes while eliminating transient interference. Linear spatial interpolation calculates the temperature value of the intermediate region based on temperature measurements from adjacent sensors. This can be implemented using a one-dimensional linear interpolation algorithm to construct a complete axial temperature distribution for the cable.

[0055] Ambient temperature compensation refers to adjusting the measured temperature value based on the external ambient temperature. This can be achieved by multiplying the ambient temperature collected by a temperature sensor by a compensation coefficient, thus eliminating the impact of ambient temperature variations on measurement accuracy. Wavelet basis three-level decomposition refers to removing interference components from voltage change rate data through multi-scale decomposition. This can be achieved using a db4 wavelet basis for decomposition and reconstruction, eliminating spurious voltage fluctuations caused by power grid harmonics. Soft thresholding denoising involves thresholding the high-frequency coefficients after wavelet decomposition. This can be achieved by using a threshold of 0.2V / ms for signal reconstruction, preserving the true voltage sag characteristics. Normalization to... An interval refers to mapping physical quantities with different dimensions to a unified numerical range. Specifically, it can be implemented using a minimum-maximum normalization algorithm to eliminate the impact of differences in data magnitude on subsequent fusion calculations.

[0056] Specifically, the current data first undergoes sliding window filtering to remove spurious fluctuations caused by high-frequency noise while retaining the current spike characteristics of short-circuit precursors. Further dynamic estimation using Kalman filtering corrects sensor measurement errors, stabilizing current fluctuations within a controllable range. Contact resistance data undergoes outlier removal using the 3σ criterion and is then smoothed using an exponentially weighted moving average algorithm, preserving the slow trend of resistance changes while eliminating instantaneous jumps caused by poor contact. Temperature distribution data is supplemented with linear interpolation to fill in temperature values ​​in the sensor intervals, forming a complete axial temperature distribution curve. This is combined with an ambient temperature compensation coefficient to correct measurement deviations, ensuring accurate identification of high-temperature areas. Voltage rate of change data undergoes wavelet decomposition and soft thresholding to remove grid harmonic interference, preserving true voltage abrupt changes. All preprocessed data is then normalized to convert physical quantities of different dimensions into a unified numerical range, providing standardized input for subsequent multi-evidence fusion.

[0057] Existing technologies typically employ a single filtering algorithm for all data types, such as using only moving average or Kalman filtering to process all physical quantity data, resulting in uneven noise suppression effects. This solution, however, designs matching preprocessing algorithms for different noise sources and feature extraction requirements of different physical quantities. For example, a dual filtering combination is used for current data to retain pre-fault characteristics while reducing noise; contact resistance data employs a combination of outlier removal and trend smoothing to avoid feature distortion caused by a single algorithm. Existing technologies do not address the problem of missing temperature data due to sensor spacing, while this solution constructs a complete temperature distribution through linear interpolation and environmental compensation, significantly improving the reliability of high-temperature region detection.

[0058] This application effectively solves the problems of noise interference, abnormal fluctuations, missing data, and inconsistent formats in the preprocessing of multi-physical quantity data. Noise reduction processing of current and voltage data avoids interference from spurious fluctuations in feature extraction; smoothing of contact resistance ensures accurate identification of resistance change trends; temperature interpolation and compensation improve the detection integrity of overheated areas; and normalization processing eliminates compatibility issues in multi-source data fusion. Therefore, the preprocessed data can accurately reflect the characteristics of fault precursors, providing a reliable data foundation for subsequent fault probability calculations.

[0059] This application further proposes to extract three types of features from current data: current peak frequency, current rise rate, and current fluctuation amplitude; two types of features from contact resistance data: resistance mean drift and resistance change trend slope; three types of features from temperature distribution data: local maximum temperature, temperature gradient difference, and proportion of high-temperature area; and two types of features from voltage change rate data: peak instantaneous rate of change of voltage and duration of voltage instantaneous rate of change anomaly; and to summarize the above features to form a feature set.

[0060] The current spike frequency refers to the number of times the current exceeds a preset fluctuation threshold per unit time. This can be achieved using a sliding window statistical method to capture high-frequency abnormal signals before a short circuit. The current rise rate refers to the slope of the current change over time, which can be calculated using a linear regression algorithm to identify phenomena of accelerated current growth.

[0061] Current fluctuation amplitude refers to the maximum deviation of the current within a preset time period, which can be calculated using the range method and is used to characterize the degree of continuous current anomaly. Resistance mean drift refers to the offset of the average contact resistance relative to its initial state, which can be calculated using a moving average algorithm and is used to reflect the degree of degradation of the terminal contact surface.

[0062] The slope of the resistance change trend refers to the rate at which the contact resistance changes over time. It can be obtained using least squares fitting and is used to determine the accelerating trend of resistance change. The local maximum temperature refers to the maximum value in the axial temperature distribution of the cable. It can be detected using a distributed temperature sensor array and is used to identify localized overheating areas. The temperature gradient difference refers to the maximum temperature difference between adjacent temperature measurement points along the cable's axial direction. It can be calculated using the difference in data from adjacent sensors and is used to reflect abnormal heat diffusion.

[0063] The percentage of high-temperature regions refers to the proportion of sensors exceeding a preset temperature threshold. This can be achieved using threshold comparison and counting methods to assess the overheating range. The peak value of the instantaneous rate of change of voltage refers to the extreme value of the voltage change rate. This can be obtained using a differentiating circuit combined with a peak detection module to detect voltage drops. The duration of the abnormal instantaneous rate of change of voltage refers to the length of time the voltage change rate exceeds a preset threshold. This can be achieved using a combination of a timer and a comparator circuit to determine the persistence of the voltage anomaly.

[0064] Specifically, the frequency of current spikes can be used to effectively identify instantaneous pulse signals before a short circuit by statistically analyzing the number of times the current data exceeds the normal fluctuation range; the rate of current rise can be used to detect abnormal current acceleration in advance by analyzing the current growth trend; and the amplitude of current fluctuations can be used to reflect persistent abnormal conditions by calculating the degree to which the current deviates from the rated value.

[0065] The drift of the average contact resistance can be detected by monitoring changes in the average resistance value, which can capture resistance degradation caused by terminal oxidation or loosening; the slope of the resistance change trend can be used to determine whether the contact surface degradation is accelerating by quantifying the rate of resistance change. The local maximum temperature can be located by detecting the extreme values ​​in the cable temperature distribution; the temperature gradient difference can be used to identify heat diffusion obstruction by calculating the differences between adjacent temperature measurement points; the proportion of high-temperature areas can be used to assess the overall thermal state of the cable by statistically analyzing the proportion of overheated areas.

[0066] The peak value of the instantaneous rate of change of voltage can detect short-circuit precursors by capturing the magnitude of voltage drops; the abnormal duration of the instantaneous rate of change of voltage can distinguish between transient disturbances and actual faults by recording the cumulative duration of abnormal voltage changes. These ten types of features extract key abnormal information of electrical and thermodynamic parameters from different dimensions, forming a complementary verification mechanism and providing comprehensive data support for short-circuit fault prediction.

[0067] Existing solutions typically extract only a single feature from a single physical quantity, such as monitoring only current peaks or single-point temperature values. These methods cannot distinguish between genuine short circuits and transient interference, nor can they cover the multi-stage characteristics of fault evolution. This solution designs ten features from four categories of physical quantities to construct a multi-dimensional feature system covering abrupt changes in electrical parameters, contact surface degradation, and abnormal thermal diffusion. This system can cross-verify abnormal signals from different dimensions, effectively solving the problems of misjudgment and missed detection caused by single-parameter monitoring. For example, when the frequency of current peaks increases, if an increase in the average drift of contact resistance and a rise in local maximum temperature are simultaneously detected, it can be confirmed that the anomaly originates from a true short circuit precursor rather than grid interference.

[0068] This application can accurately distinguish between real short-circuit precursors and transient interference signals based on feature extraction and fusion analysis of multi-physical quantity data, significantly improving the reliability of fault detection. Through multi-dimensional feature cross-validation of current, resistance, temperature, and voltage change rate data, misjudgments caused by single parameter anomalies can be avoided; through full-cycle feature design covering the fault initiation to the critical state, micro-short-circuit causes and insulation degradation phenomena can be identified in advance, reducing the risk of missed detection.

[0069] This application further proposes dividing the key features into multiple evidence bodies, including evidence body one, evidence body two, and evidence body three. Evidence body one includes the current peak frequency, current rise rate, peak instantaneous rate of change of voltage, and abnormal duration of instantaneous rate of change of voltage. Evidence body two includes the mean resistance drift, slope of resistance change trend, local maximum temperature, and temperature gradient difference. Evidence body three includes the current fluctuation amplitude and the proportion of high-temperature areas. The confidence level of each evidence body for short-circuit anomalies is calculated using DS evidence theory, and the comprehensive confidence level is obtained through evidence synthesis formula. When the comprehensive confidence level is ≥0.8, the corresponding key features are combined with the comprehensive confidence level to form fault feature data. When the comprehensive confidence level is <0.5, it is judged as normal data and no standardized fault feature data is generated.

[0070] Among them, the current spike frequency refers to the number of times the current exceeds a preset threshold per unit time. This can be implemented using a sliding window counting algorithm to capture transient current surges caused by poor contact inside the charging gun. The current rise rate refers to the amount of current change per unit time. This can be implemented using differential calculation combined with low-pass filtering to identify abnormal current growth trends in the initial stage of a short circuit. The peak instantaneous rate of change of voltage refers to the absolute value of the maximum rate of change during voltage fluctuations. This can be implemented using a peak detection circuit combined with a sample-and-hold module to reflect voltage drops caused by sudden changes in terminal contact resistance.

[0071] The duration of voltage anomaly refers to the length of time during which the voltage rate of change exceeds the normal fluctuation range. Specifically, it can be achieved by using a timer and a comparator working together to determine whether the voltage anomaly persists.

[0072] The mean resistance drift refers to the deviation of the contact resistance relative to a reference value. This can be achieved using a moving average algorithm combined with reference value calibration, and is used to detect resistance anomalies caused by terminal oxidation or loosening. The slope of the resistance change trend refers to the rate at which the contact resistance changes over time. This can be achieved using linear regression analysis combined with time series data, and is used to assess the dynamic process of resistance degradation. The local maximum temperature refers to the maximum surface temperature of the charging gun cable. This can be achieved using a distributed temperature sensor array combined with a maximum value screening algorithm, and is used to identify the risk of insulation damage caused by localized overheating.

[0073] Temperature gradient difference refers to the maximum temperature difference between adjacent sensors. This can be achieved by calculating the temperature difference between adjacent nodes combined with a sorting algorithm, and is used to detect temperature surges caused by localized short circuits in the cable. Current fluctuation amplitude refers to the maximum fluctuation range of the current within a set time window. This can be achieved by using range calculation combined with sliding window statistics, and is used to reflect abnormal current stability. High-temperature region percentage refers to the proportion of sensors whose temperature exceeds a preset threshold out of the total number of sensors. This can be achieved by using threshold comparison combined with counter accumulation, and is used to assess the overall overheating degree of the cable.

[0074] DS evidence theory refers to a mathematical method for processing multi-source uncertain information based on evidence synthesis rules. Specifically, it can be implemented using a combination of basic probability assignment functions and orthogonal sum operators to resolve conflicts and uncertainties among multiple pieces of evidence. The overall confidence level refers to the degree of support for the short-circuit anomaly proposition after the fusion of multiple pieces of evidence. It can be achieved by combining the DS evidence synthesis formula with normalization processing to quantify the credibility of fault characteristic data.

[0075] Specifically, the peak frequency of current spikes and the peak instantaneous rate of change of voltage are categorized into evidence body one. By capturing the abrupt changes in transient electrical parameters, potential poor contact or arcing phenomena in the early stages of a short circuit can be identified. The mean resistance drift and temperature gradient difference are categorized into evidence body two. By analyzing the drift trend of contact resistance and the spatial differences in temperature distribution, progressive degradation caused by terminal oxidation or cable insulation aging can be identified.

[0076] The amplitude of current fluctuations and the proportion of high-temperature regions are categorized into Evidence Entity Three. By correlating the global anomalies in current stability and temperature distribution, the risk of persistent short circuits due to insulation failure is identified. Evidence Entities One, Two, and Three correspond to the three stages of short-circuit faults: transient anomalies, progressive degradation, and global anomalies, respectively, forming a multi-dimensional feature coverage. The DS evidence theory calculation formula is as follows: The overall confidence level is obtained, where, This represents the overall confidence level after fusing the evidence, comprehensively reflecting the degree to which the three types of evidence jointly support the conclusion that "the charging gun has a short circuit anomaly." It is the core basis for determining whether to generate standardized fault characteristic data. , , The values ​​represent the confidence levels of evidence 1, evidence 2, and evidence 3 in supporting the proposition that the charging gun has a short circuit anomaly. The higher the value, the more the evidence supports the "short circuit anomaly".

[0077] In the calculation of DS evidence theory, the confidence weight of each piece of evidence is allocated differently according to the characteristics of the directional intensity of the short-circuit anomaly. For example, the weight of the current peak frequency is higher than the duration of the voltage instantaneous rate of change anomaly.

[0078] The confidence scores of multiple evidence bodies are synthesized using orthogonal methods and operators. When the overall confidence score exceeds 0.8, it is judged as a high-confidence anomaly, and standardized fault feature data is generated. When the overall confidence score is below 0.5, random interference under normal operating conditions is directly excluded to avoid invalid calculations. This mechanism ensures that subsequent diagnostic processes are triggered only when multi-dimensional features collaboratively point to a short-circuit anomaly.

[0079] Existing charging gun short-circuit detection typically relies on a single current parameter threshold, failing to distinguish between grid interference and genuine short circuits, and unable to identify micro-short circuits caused by progressive degradation. This solution divides evidence into three categories to capture transient anomalies, equipment condition degradation, and global anomaly features, covering the entire lifecycle of short-circuit faults. It employs DS evidence theory to fuse the confidence levels of multiple evidence categories, addressing the problem of misjudgment based on a single parameter. A dual-reset confidence threshold is set to improve detection accuracy while reducing invalid data processing. Existing multi-feature fusion methods often involve direct weighting without classification, neglecting the characteristics of different fault stages. In contrast, this solution achieves deep matching between features and fault mechanisms through evidence category classification and weight allocation.

[0080] This application effectively reduces misjudgments caused by fluctuations in a single parameter or environmental interference. For example, if a current spike caused by instantaneous fluctuations in the power grid is not accompanied by an abnormal rate of voltage change or a temperature increase, the overall confidence level cannot reach the threshold. Simultaneously, it avoids missing micro-short circuits caused by progressive degradation. For instance, the synergistic effect of slow drift in contact resistance and local temperature increases can be accurately identified through the fusion of evidence body two and evidence body three. Through multi-evidence body collaborative judgment and confidence threshold filtering, high-precision short-circuit anomaly detection is achieved.

[0081] This application further proposes that the output of the digital twin pre-diagnostic model includes two parts: the first part is the probability of short circuit fault occurrence, specifically the probability value of the charging gun short circuit in the future. The calculation accuracy is based on historical data training and optimization of three typical faults: terminal loosening, cable wear, and water ingress short circuit; the second part is the probability confidence level, which characterizes the reliability of the probability value. When the probability confidence level is less than 0.8, the latest fault feature data needs to be re-entered for secondary calculation.

[0082] The short-circuit fault probability refers to the likelihood of the charging gun short-circuiting within the next 500 milliseconds. This can be achieved using a neural network model trained on historical fault data, generating the probability value by inputting multi-dimensional fault feature data. The probability confidence level is the model's reliability assessment of the currently calculated short-circuit fault probability. This can be achieved using a Bayesian confidence network algorithm, generating the confidence level value by analyzing the completeness of the input data and the logical consistency between features.

[0083] Specifically, the digital twin pre-diagnostic model is trained by fusing historical data from three typical fault types, enabling the model to identify the characteristic patterns of different fault modes. When real-time collected fault feature data is input into the model, it first calculates the short-circuit probability within the next 500 milliseconds, and then evaluates the confidence level of this probability. If the confidence level is below 0.8, it indicates that the current input data contains noise interference or missing features. At this point, a data re-acquisition process is triggered to reacquire the latest fault feature data and recalculate until the confidence level reaches the reliability threshold. Through this dual output mechanism, both quantitative prediction of fault risk and reliable verification of the prediction results are achieved.

[0084] Existing fault detection models only output a single judgment result and lack a self-verification mechanism, failing to distinguish the characteristic patterns of different fault modes. This solution establishes a two-way correlation between probability calculation and data quality through targeted training on historical data and dynamic confidence assessment. It limits the prediction window to 500 milliseconds in the time dimension, satisfying the time requirements for protection measure execution while avoiding environmental interference errors in long-term predictions.

[0085] This application effectively solves the problem of misjudgment and missed judgment caused by single parameter monitoring. It achieves accurate risk assessment by quantifying probability output, and at the same time uses a confidence verification mechanism to ensure the reliability of the detection results, significantly improving the accuracy of charging gun short circuit fault detection and the timeliness of protection.

[0086] This application further proposes determining the risk level based on the probability of a short-circuit fault, including normal state, low-risk state, high-risk state, emergency state, and pending confirmation state. Normal state is defined as a short-circuit fault occurrence probability of less than 50% with a probability confidence level greater than or equal to 0.8; low-risk state is defined as a short-circuit fault occurrence probability between 50% and 80% with a probability confidence level greater than or equal to 0.8; high-risk state is defined as a short-circuit fault occurrence probability between 80% and 95% with a probability confidence level greater than or equal to 0.8; emergency state is defined as a short-circuit fault occurrence probability greater than or equal to 95% with a probability confidence level greater than or equal to 0.8; if the probability confidence level is less than 0.8, it is temporarily designated as pending confirmation state, and no protection measures are implemented, only data re-acquisition is triggered.

[0087] Risk level classification refers to combining the probability of short-circuit fault occurrence with probability confidence level to form multi-level judgment conditions. Specifically, it can be implemented by using probability interval division algorithm and confidence threshold comparison algorithm. The probability value reflects the possibility of the fault, and the confidence value verifies the reliability of the judgment.

[0088] Probability confidence refers to the statistical reliability index of the model's output results. It can be implemented using Bayesian posterior probability calculation methods to filter out false high-probability judgments caused by data noise or model errors. The pending confirmation state refers to pausing the protection action and re-verifying the data when the confidence level is insufficient. This can be achieved using a data re-sampling loop mechanism, which re-executes the data preprocessing, feature extraction, and model calculation processes to generate new judgment results, avoiding erroneous operations caused by a single instance of abnormal data.

[0089] Specifically, during the charging gun's operation, the digital twin pre-diagnostic model continuously outputs the probability of a short-circuit fault and its corresponding confidence level. When the confidence level reaches 0.8, it matches four preset risk level intervals based on the probability value. For example, if the probability is 65% and the confidence level is 0.85, it is determined to be a low-risk state, triggering a power reduction operation and prompting a terminal inspection; if the probability is 90% and the confidence level is 0.82, it is determined to be a high-risk state, activating an audible and visual alarm and notifying maintenance personnel. When the confidence level is below 0.8, the judgment result is marked as pending confirmation. At this time, the system automatically re-collects multi-physical quantity data, preprocesses and fuses it, and then re-inputs it into the model for calculation until a judgment result with a satisfactory confidence level is obtained. This mechanism ensures that each protection action is executed based on highly reliable data, and simultaneously achieves dynamic matching between protection strength and fault severity through a multi-level response strategy.

[0090] Traditional methods rely solely on a single probability threshold to trigger protection actions, such as directly cutting off the circuit when the probability exceeds 90%, failing to differentiate fault development stages and implement tiered measures. This solution effectively eliminates misjudgments caused by transient sensor interference by introducing a confidence verification mechanism; by dividing the fault into four probability intervals and binding differentiated protection measures, it implements early warning intervention in the early stages of the fault and performs emergency power cut-off in the critical stage, avoiding the problems of over-protection or protection lag in existing technologies.

[0091] This application addresses the mismatch in protection measures caused by ambiguous risk level classification, achieving precise correspondence between fault prediction and protection actions. A dual-constraint-based judgment mechanism effectively improves the accuracy of risk identification, a gradient response strategy reduces unnecessary charging interruptions, and a closed-loop resampling process enhances the system's fault tolerance under complex operating conditions, ensuring a balance between safety and continuity during the charging process.

[0092] This application further proposes a computing device, including at least one processor and a memory. The memory stores executable instructions. When the instructions are executed by at least one processor, the processor acquires multi-physical quantity data in real time during the operation of the charging gun. The multi-physical quantity data includes current data, contact resistance data, temperature distribution data, and voltage change rate data. The multi-physical quantity data is processed to obtain fault characteristic data. The fault characteristic data is input into a digital twin pre-diagnosis model. The digital twin pre-diagnosis model calculates the probability of a short-circuit fault occurring in the charging gun. The digital twin pre-diagnosis model integrates a data fusion algorithm and a fault evolution model. Based on the probability of a short-circuit fault occurring, the risk level is determined, and corresponding protection measures are implemented based on the risk level.

[0093] At least one processor refers to a hardware unit capable of performing parallel computing, specifically a multi-core CPU or GPU cluster, used to synchronously process four heterogeneous data streams: current, resistance, temperature, and voltage. The memory refers to a non-volatile storage medium, specifically flash memory chips or solid-state drives, used to permanently store the instruction set containing multi-physical quantity acquisition rules, feature fusion algorithms, and digital twin model parameters. The executable instructions refer to pre-compiled program code, specifically implemented in machine language or bytecode format, used to define the entire process control logic from data acquisition to the triggering of protection measures. The digital twin pre-diagnostic model refers to a virtual-real mapping simulation system, specifically implemented using a hybrid modeling approach combining neural networks and physical equations, used to fuse real-time data with historical fault evolution patterns to predict short-circuit probabilities.

[0094] Specifically, after the instructions stored in memory are loaded by the processor, they first drive the sensor interface module to collect current, voltage, temperature, and contact resistance data at different sampling rates. The multi-core processor performs sliding window filtering, outlier removal, spatial interpolation compensation, and wavelet denoising on the four types of data to eliminate noise caused by environmental interference. The processed data is input to the feature extraction module, which synchronously extracts ten types of features, including current peak frequency, resistance mean drift, temperature gradient difference, and voltage change rate peak, through parallel computing threads. The feature fusion module calls a multi-evidence body synthesis algorithm to divide the features into three evidence bodies for confidence calculation. When the comprehensive confidence reaches a threshold, standardized fault feature data is generated. After receiving the feature data, the digital twin pre-diagnostic model combines fault evolution parameters such as cable wear rate and terminal oxidation process to dynamically calculate the probability value and confidence of future short circuits. Based on the risk level range of the probability value, the processor triggers the corresponding level of protection instructions, such as only logging in low-risk situations, initiating current limiting in high-risk situations, and cutting off the main circuit in emergency situations.

[0095] Traditional charging gun testing equipment only uses a single current sensor and threshold comparator, which cannot distinguish between real short circuits and transient interference, and the protection action lags behind the occurrence of the fault. This solution, through a collaborative architecture of processor and memory, achieves parallel processing and fusion analysis of multi-source heterogeneous data, incorporating the spatiotemporal variation characteristics of four parameters—current, resistance, temperature, and voltage—into a unified evaluation system. The digital twin model breaks through the dependence of traditional detection circuits on preset thresholds; through digital modeling of fault evolution patterns, it can identify micro-short circuit precursors caused by terminal oxidation. The data processing flow embedded in memory ensures stable algorithm operation under extreme temperature environments, avoiding malfunctions caused by temperature drift in traditional analog circuits.

[0096] This application effectively solves the problem of false positives and false negatives caused by single-parameter monitoring, and improves the detection accuracy to meet industrial application requirements through the fusion of multi-physical quantity data. The digital twin pre-diagnostic model realizes probabilistic prediction before the fault occurs, enabling the protection measures to be triggered in the early stage of short-circuit current formation. The parallel computing architecture of the processor supports fast response and can complete the circuit disconnection operation before the cable insulation layer is damaged. The physical isolation design between the storage medium and the processing module ensures stable operation in a temperature range of -30℃ to 60℃, avoiding the protection threshold drift caused by environmental temperature changes in traditional detection circuits.

[0097] This application further proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor implements a method for detecting short-circuit faults in a charging gun. This method includes real-time acquisition of multi-physical quantity data during the operation of the charging gun, including current data, contact resistance data, temperature distribution data, and voltage change rate data. The multi-physical quantity data is processed to obtain fault feature data. The fault feature data is then input into a digital twin pre-diagnosis model. The probability of a short-circuit fault occurring in the charging gun is calculated using the digital twin pre-diagnosis model, which integrates a data fusion algorithm and a fault evolution model. The risk level is determined based on the probability of the short-circuit fault, and corresponding protection measures are implemented based on the risk level.

[0098] Computer-readable storage media refers to non-volatile storage media capable of persistently storing program code, specifically solid-state drives (SSDs) or flash memory chips, used to solidify the execution logic of the detection algorithm. A computer program is a set of instructions containing data processing flows, specifically written in C++ or Python, used to implement programmed control for multi-physical quantity data acquisition and model calculation. A processor is the arithmetic unit that executes program instructions, specifically implemented using embedded microcontrollers or industrial PLC chips, used to run the detection algorithm in real time and output control signals. Multi-physical quantity data refers to a set of multi-dimensional parameters reflecting the operating status of the charging gun, specifically acquired synchronously by current sensors, voltage sampling chips, and temperature sensors, used to construct multi-dimensional fault judgment criteria. Fault feature data refers to structured data that has undergone feature extraction and fusion processing, specifically generated using sliding window filtering, Kalman filtering, and wavelet denoising algorithms, used to eliminate the impact of environmental interference on detection accuracy. A digital twin pre-diagnostic model is a predictive model that integrates physical laws and data-driven approaches, specifically constructed using LSTM neural networks and DS evidence theory, used to simulate the evolution of short-circuit faults. Risk level refers to the warning level based on probability values. It can be categorized using preset threshold ranges to trigger graded protection mechanisms. Protection measures refer to the operational instructions executed according to the risk level, which can be implemented using relay control signals or communication alarm messages to prevent short-circuit faults from escalating.

[0099] Specifically, after the computer program stored in the storage medium is loaded into the processor, it first controls the sensor array to synchronously collect current, voltage, and temperature data at a sampling frequency of no less than 10kHz. A sliding window filter is used to eliminate current spike noise, and an exponentially weighted moving average algorithm is used to smooth contact resistance jumps. Wavelet denoising is then applied to process voltage fluctuation data. The preprocessed multi-physical quantity data is normalized to a unified dimension. A feature extraction module extracts the rise rate feature from the current data, the mean shift from the contact resistance data, and the gradient difference from the temperature distribution data, forming a feature set containing 12 dimensions. The DS evidence theory is used to fuse multiple evidence bodies into the feature set. Standardized fault feature data is generated when the overall confidence level reaches 0.8. After receiving the fault feature data, the digital twin pre-diagnosis model simulates the development trends of typical faults such as loose terminals and cable wear using a fault evolution model. A probability prediction module trained with historical data outputs the probability of a future short circuit. When the probability value exceeds 95%, the processor immediately generates an emergency shutdown command and sends it to the charging pile control system via the CAN bus. At the same time, when the probability confidence level is lower than 0.8, the data re-sampling process is automatically triggered to eliminate false judgments.

[0100] Traditional detection schemes rely on a single current threshold triggering protection mechanism, failing to correlate the impact of temperature changes and contact resistance drift on short-circuit risk. This solution, however, uses a multi-physical quantity fusion algorithm embedded in storage media, enabling the processor to simultaneously analyze multi-dimensional precursor features such as current spikes, resistance abrupt changes, and abnormal temperature gradients. Existing technologies use fixed hardware circuits for threshold comparisons, making it difficult to dynamically adjust detection parameters to adapt to different ambient temperatures. This solution, through a programmable digital twin model, can automatically compensate for temperature drift, maintaining detection accuracy within an environment range of -30℃ to 60℃. In existing technologies, fuse activation requires waiting for the current to reach a preset threshold, while this solution, based on probabilistic prediction, can trigger an early warning 50ms before the short-circuit current forms, effectively preventing terminal burnout.

[0101] This application solves the problem of misjudgment caused by single-parameter monitoring in traditional detection methods. By fusing multiple physical quantity data, the accuracy of micro-short circuit identification is improved to over 95%. It overcomes the limitation of poor environmental adaptability of hardware circuits and can maintain detection stability of ±0.5% even under extreme temperature conditions. It also realizes the fault prediction function, advancing the response time of protection actions to before the formation of short-circuit current, thus avoiding irreversible damage to the internal components of the charging gun.

[0102] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting short-circuit faults in a charging gun, characterized in that, include: Real-time acquisition of multiple physical quantity data during the operation of the charging gun, including current data, contact resistance data, temperature distribution data, and voltage change rate data; The multi-physical quantity data is processed to obtain fault feature data, and the fault feature data is input into a digital twin pre-diagnosis model. The probability of a short circuit fault occurring in the charging gun is calculated through the digital twin pre-diagnosis model. The digital twin pre-diagnosis model integrates a data fusion algorithm and a fault evolution model. The risk level is determined based on the probability of the short-circuit fault occurring, and corresponding protection measures are implemented based on the risk level.

2. The charging gun short-circuit fault detection method according to claim 1, characterized in that, The real-time acquisition of multiple physical quantity data during the operation of the charging gun includes: Current data is collected using a current sensor at a sampling frequency of no less than 10kHz to identify precursors to current spikes. The voltage of the charging gun terminals is detected in real time by a voltage sampling chip, and the contact resistance value is calculated by combining the current data. Distributed temperature sensors are arranged at preset intervals along the charging gun cable to collect cable temperature distribution data. The instantaneous rate of change of the output voltage of the charging gun is detected by a voltage fluctuation detection chip.

3. The charging gun short-circuit fault detection method according to claim 1, characterized in that, The fault characteristic data obtained by processing the multi-physical quantity data includes: The different types of multi-physical quantity data are preprocessed to unify the dimensions, and key features that can characterize the precursors of short-circuit faults in charging guns are extracted from the preprocessed multi-physical quantity data to form a feature set containing multi-dimensional features. The extracted key features are fused and calculated using a multi-evidence body synthesis algorithm to obtain the fault feature data that can be directly input into the digital twin pre-diagnosis model.

4. The charging gun short-circuit fault detection method according to claim 3, characterized in that, Preprocessing of the different types of multi-physical quantity data includes: The current data is denoised using sliding window filtering and Kalman filtering to control the fluctuation deviation within a certain range. within; use After removing outliers from the contact resistance data, the data is smoothed using an exponentially weighted moving average algorithm to eliminate resistance value jumps caused by transient interference. The temperature distribution data is supplemented by linear spatial interpolation to supplement the temperature values ​​in the sensor interval area, and compensation is made according to the ambient temperature. use Wavelet-based three-level decomposition combined with soft thresholding is used to denoise the voltage change rate data and eliminate spurious fluctuations caused by power grid interference. Normalize all preprocessed multi-physical quantity data to Range, unified data format.

5. The charging gun short-circuit fault detection method according to claim 3, characterized in that, Three types of features are extracted from the current data: current peak frequency, current rise rate, and current fluctuation amplitude. Two types of features are extracted from the contact resistance data: the mean resistance drift and the slope of the resistance change trend. Three types of features are extracted from the temperature distribution data: local maximum temperature, temperature gradient difference, and proportion of high-temperature areas. Extract two types of features from the voltage change rate data: the peak value of the instantaneous rate of change of voltage and the duration of abnormal instantaneous rate of change of voltage. The above features are summarized to form the feature set.

6. The charging gun short-circuit fault detection method according to claim 3, characterized in that, The key features are divided into multiple evidence bodies, including evidence body one, evidence body two, and evidence body three, wherein... The evidence includes the current spike frequency, current rise rate, peak value of instantaneous voltage change rate, and abnormal duration of instantaneous voltage change rate. The second piece of evidence includes the mean resistance drift, the slope of the resistance change trend, the local maximum temperature, and the temperature gradient difference. The third piece of evidence includes the amplitude of current fluctuations and the proportion of high-temperature areas; The confidence level of each piece of evidence regarding the short-circuit anomaly is calculated using the DS evidence theory, and then synthesized using the evidence combination formula. The overall confidence level is obtained, where, This indicates the overall confidence level after integrating the evidence. , , The numbers represent the confidence levels of evidence 1, evidence 2, and evidence 3 regarding the proposition that the charging gun has a short circuit abnormality. When the overall confidence level is ≥0.8, the corresponding key features are combined with the overall confidence level to form fault feature data; When the overall confidence level is less than 0.5, the data is considered normal and no standardized fault feature data is generated.

7. The charging gun short-circuit fault detection method according to any one of claims 1-6, characterized in that, The output of the digital twin pre-diagnostic model includes two parts: The first part is the probability of short circuit failure, specifically the probability value of the charging gun short circuit in the future. The calculation accuracy is based on historical data training and optimization of three typical faults: loose terminals, cable wear, and water ingress short circuit. The second part is the probability confidence level, which characterizes the reliability of the probability value. When the probability confidence level is <0.8, the latest fault characteristic data needs to be re-entered for secondary calculation.

8. The charging gun short-circuit fault detection method according to claim 7, characterized in that, The determination of the risk level based on the probability of the short-circuit fault occurring includes: Under normal conditions, the probability of a short circuit fault is <50%, and the probability confidence level is ≥0.8; In a low-risk condition, the probability of a short-circuit fault is 50% ≤ < 80%, and the probability confidence level is ≥ 0.8; High-risk condition: 80% ≤ probability of short circuit fault occurrence < 95%, and probability confidence level ≥ 0.8; In an emergency, the probability of a short circuit fault occurring is ≥95%, and the confidence level of the probability is ≥0.8; If the probability confidence level is <0.8, it is temporarily set to a pending confirmation state, and no protection measures are implemented; only data re-collection is triggered.

9. A computing device, characterized in that, include: At least one processor; A memory that stores executable instructions; When the instruction is executed by the at least one processor, the at least one processor implements the charging gun short-circuit fault detection method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the charging gun short-circuit fault detection method as described in any one of claims 1 to 8.

Images