Charging pile nonmetal grounding fault early warning method based on residual current characteristic analysis

Abstract

The invention discloses a charging pile non-metal grounding fault early warning method based on residual current characteristic analysis, and the method comprises the steps: constructing a transformer area AC charging pile safety power utilization non-metal grounding fault diagnosis method which integrates multi-dimensional residual current information and deeply integrates solar term environment factors, and analyzing the transient characteristics and fluctuation trend of residual current; the pyrolysis ion concentration is explored to serve as an early warning index of early overheating decomposition of the insulating material; meanwhile, the solar term environment parameters serve as important correction factors of fault probability, an intelligent diagnosis model capable of recognizing and distinguishing different nonmetal grounding fault types is built, remote monitoring and management of the charging pile are achieved, monitoring accuracy and operation and maintenance efficiency are improved, and safe operation of the charging pile is ensured.

Description

Technical Field

[0001] This invention relates to the field of charging and swapping technology, specifically to a method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis. Background Technology

[0002] Currently, safety monitoring of AC charging piles mainly employs methods such as electrical parameter monitoring, temperature monitoring, and insulation monitoring. Electrical parameter monitoring involves real-time monitoring of parameters such as voltage, current, and power of the charging pile to determine if there are faults such as overload or short circuit. Temperature monitoring uses temperature sensors installed inside the charging pile to monitor its temperature in real time, preventing safety accidents such as fires caused by excessive heat. Insulation monitoring measures the insulation resistance of the charging pile to determine its insulation performance and prevent leakage accidents.

[0003] Non-metallic grounding faults in AC charging piles, also known as high-impedance grounding faults, are characterized by a high impedance to ground at the fault point. This results in a fault current that is much lower than the operating threshold of conventional overcurrent protection devices, making it difficult for traditional circuit breakers to cut off the fault in time. This type of fault is highly latent and poses a significant hazard.

[0004] However, existing safety monitoring technologies still have some shortcomings. On the one hand, some monitoring technologies can only monitor a single parameter, failing to comprehensively reflect the operating status of charging piles and easily leading to missed or false alarms. On the other hand, most existing monitoring technologies rely on manual analysis and judgment, which is inefficient and makes it difficult to achieve real-time and accurate monitoring of charging piles. Furthermore, with the continuous increase in the number of charging piles and their increasing level of intelligence, traditional monitoring technologies also face significant challenges in data processing and transmission.

[0005] To address the above issues, Chinese Patent CN109190670B discloses a charging pile fault prediction method based on a scalable boosting tree. The method mainly comprises two stages. The first stage includes the following steps: First, fault characteristic state signals of various components of the charging pile and fault information of the charging pile are collected and sent to a data analysis platform as model training sample data. The data analysis platform uses the Xgboost algorithm to train the large amount of received training sample data to obtain a charging pile fault prediction model with a certain level of accuracy. In the second stage, when the data analysis platform receives the charging pile's state signal again, it uses the trained Xgboost model to determine the charging pile's fault state.

[0006] For example, Chinese patent CN118608120A discloses a charging pile fault prediction method and device, including: constructing a charging pile fault prediction model based on charging pile data using a stacking multi-model fusion method, the charging pile fault prediction model including a first-level pre-selected base learner and a second-level meta-learner; training the first-level pre-selected base learners and optimizing the hyperparameters of the base learners using a sparrow search algorithm to obtain the first-level output results; performing tree model importance selection and correlation analysis on the first-level output results; the second-level meta-learner selects the optimal base learner combination based on the first-level output results to construct an improved stacking multi-model fusion fault prediction optimization model; and predicting charging pile faults based on the improved stacking multi-model fusion fault prediction optimization model.

[0007] Currently, existing charging pile fault early warning technologies still have shortcomings: both patents have improved the reliability of charging pile operation to a certain extent, but they cannot realize the automatic analysis and processing of charging pile operation data, thus they cannot comprehensively monitor the electrical performance, mechanical performance, environmental adaptability, etc. of charging piles. The charging piles and monitoring centers have not yet achieved interconnection, and the existing technologies still need to be improved. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis, in order to solve the aforementioned problems.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis, comprising the following steps: S1: Electrical Data Analysis and Feature Extraction: Analyze electrical data features such as residual current amplitude and duration, and charging power when there is a non-metallic grounding fault in an AC charging pile. Use data mining and machine learning algorithms to analyze the correlation and change patterns between the data.

[0010] S1-1: Residual Current Data Characteristic Analysis Grounding faults in AC charging piles are most directly reflected in changes in the "information dimension" of the residual current in the line. The characteristics of the residual current are the "fingerprint" of the fault. By interpreting this fingerprint, accurate fault identification, classification, and early warning can be achieved.

[0011] S1-2: Residual Current Data Feature Extraction The residual current value in the AC charging pile line is extracted from the time domain characteristics, frequency domain characteristics, and time-frequency characteristics.

[0012] S2: Pyrolysis Ion Data Analysis and Feature Extraction: Analyze the amplitude and duration characteristics of ambient pyrolysis ion concentration under different types of non-metallic grounding faults in AC charging piles. Construct a weighted analysis of the proportion of pyrolysis ion data in the diagnosis of non-metallic grounding fault types based on big data mining.

[0013] S2-1: Characteristics of pyrolysis ion concentration The characteristics of pyrolysis ion concentration can reflect the overheating condition of electrical equipment and can be used to identify faults such as contact overheating in non-metallic grounding faults.

[0014] The formula for calculating the mean is: (14) of which The average concentration of pyrolysis ions is represented by C_, where M is the sampling period and C_ is the average concentration of pyrolysis ions. k This is the kth sample value.

[0015] The formula for calculating variance is: (15) where C_var is the variance, M is the sampling period, and C_ k For the Kth sample value, C_mean is the mean of the pyrolysis ion concentration. It measures the dispersion of the pyrolysis ion concentration around the mean.

[0016] Concentration change trend: C_trend (linear regression slope) is the slope obtained through linear regression, which is the rate of change of concentration over time, and is used to represent the change trend.

[0017] S2-2: Environmental Feature Extraction and Correction Factors Environmental factors such as temperature and humidity have a significant impact on the operating status of electrical equipment and should not be ignored in the identification of non-metallic grounding fault types.

[0018] You can use the current measurements of temperature (T) and humidity (H) directly, or you can use the average over a period of time.

[0019] Temperature correction factor: (16)

[0020] Humidity correction factor: (17) Where Tenv represents the average temperature over a period of time, H - env represents the average humidity over a period of time.

[0021] S3: Data Normalization and Weighted Fusion: Normalize and weight data of different types and dimensions to form data parameters of a uniform magnitude, construct feature vectors, integrate feature data of multiple dimensions, and finally establish an initial charging pile non-metallic grounding fault diagnosis model.

[0022] S3-1: Feature Normalization and Weighted Fusion Because the dimensions and value ranges of different characteristics of residual current, such as time-domain characteristics, frequency-domain characteristics, pyrolysis ion concentration characteristics, and environmental factor characteristics, are quite different, directly combining these characteristics can lead to some characteristics dominating the analysis while the role of others is ignored, thus affecting the accuracy of fault type identification. To solve this problem, the Min-Max normalization method is used to process each characteristic.

[0023] S3-2: Constructing Feature Vectors Constructing a feature vector involves integrating features from multiple dimensions so that subsequent classification algorithms can comprehensively utilize this information to determine the fault type. All the normalized features mentioned above are combined into a single feature vector X, which has the following form:

[0024] By combining different types of features into a feature vector, the operating status of the power system and potential fault information can be comprehensively reflected. Each feature represents a specific dimension in the feature vector, and they complement each other to provide a basis for fault type determination.

[0025] S4: Quantitative Correction of Seasonal Environmental Factors: By introducing seasonal environmental parameters to optimize the diagnostic model for non-metallic grounding faults in charging piles, a series of environmental factors are deeply integrated to improve the diagnostic accuracy of non-metallic grounding fault types.

[0026] S4-1 Solar Term Environmental Quantitative Analysis Non-metallic grounding faults in AC charging piles are strongly correlated with seasonal environmental factors. Unlike sudden metallic short circuits, non-metallic grounding faults are essentially a gradual electrochemical-physical process driven by the environment.

[0027] Seasonal environmental factors are not independent diagnostic indicators, but rather serve as "physical mechanism catalysts" and "diagnostic logic correctors," improving the diagnostic accuracy of non-metallic grounding fault types through the deep integration of a series of environmental factors.

[0028] The deep integration of seasonal environmental factors can dynamically change the residual current non-metallic grounding fault threshold. By using environmental factors as the context and correction factor for the diagnostic model, the diagnosis becomes more consistent with the actual physical world. Based on real-time environmental conditions, the warning thresholds for residual current and ion concentration are adaptively adjusted. First, a large amount of data under normal conditions is collected to establish a regression model between environmental factors and background leakage current. The formula for normal background leakage current is: (19) of which This represents the predicted normal background leakage current, where RH is the relative humidity, T is the temperature, SeasonCode is the seasonal code, and β0, β1, β2, and β3 are regression coefficients. Dynamic threshold calculation formula: (20) in It is a dynamic alarm threshold. This represents the predicted normal background leakage current, where K is the safety factor (usually taken as 2-3), and σ is the standard deviation of the normal data and the residuals of the regression model. The most important aspect of deeply integrating seasonal environmental factors is that it can improve the accuracy of diagnosing non-metallic grounding fault types in AC charging piles. First, an environmental severity index is constructed, integrating multiple environmental parameters into a comprehensive indicator to quantify the severity of the current environment. Calculation formula: (twenty one) Where T, H, P pollutant These represent the current temperature, humidity, and pollution concentration, respectively; T ref H ref ,P ref ωT represents the environmental reference baseline value (e.g., 25°C, 60%, clean air); ωT, ωH, and ωP represent weighted indices, determined by fitting historical data, typically ωH>ωT>ωP; Sseason represents the seasonal factor (rainy season=1.3, summer=1.1, winter=0.7), and the values ​​will be updated in the future based on historical big data analysis.

[0029] Secondly, a dynamic weight allocation strategy is implemented to automatically adjust the weights of different features in the diagnostic model based on ESI. The formula for calculating the electrical feature weights is as follows: (twenty two) Formula for calculating the weight of environmental features: (twenty three) Where We0 and Wenv0 are the basic weights (e.g., 0.7 and 0.3), and their values ​​will be updated and adjusted in the future based on historical big data analysis; λ and μ are the adjustment coefficients (e.g., 0.2), which control the magnitude of weight changes, and their values ​​will be updated and adjusted in the future based on historical big data analysis.

[0030] The environmental adaptive diagnostic threshold is applied again, allowing it to change dynamically with the environment, thus addressing the poor adaptability of fixed thresholds. Diagnostic threshold formula: (twenty four) Among them TH base This is the baseline threshold under standard conditions; k: adaptability coefficient. For damp creepage, k > 0 (threshold increases to avoid alarming for normal fluctuations in humid weather), for overheating faults, k < 0 (threshold decreases to improve sensitivity in hot weather); ESI ref It is an environmental index.

[0031] Finally, a Bayesian correction is performed on the failure probability, using the environment as prior knowledge to calculate the posterior probability of a specific failure. The calculation formula is as follows: (25) Where P(FaultType|Data) is the probability of a certain fault given the data (the final output of the model); P(Data|FaultType) represents the probability of data characteristics (based on historical big data analysis). P(FaultType|Season) is the prior probability, that is, the general probability of this type of fault occurring in a certain season. It needs to be derived from historical big data analysis of different geographical locations.

[0032] S4-2 Fault Type Diagnosis First, an integrated learning framework is used, employing Gradient Boosting Decision Tree (GBDT) as the base classifier: inputting normalized feature data, outputting the probability of occurrence of eight typical non-metallic grounding faults: normal state, gun body oil contamination, dry tree branch contact, damp non-metallic contact, water tree inside insulation, condensation creepage, chemical corrosion leakage, and poor contact overheating.

[0033] Secondly, for sequence data, LSTM-attention mechanism is used for deep learning. Hidden state formula: (26) Where ht represents the hidden state of the LSTM at time t. It is the output at time t.

[0034] The optimization strategy employs loss function design and imbalanced data handling. A multi-task learning loss function is used: (27) Where λ1, λ2, and λ3 are task weight coefficients that balance the importance of different loss terms; Lclassification is the classification loss, Lseverity is the severity assessment loss, and Learned warning is the early warning loss.

[0035] Finally, a large-scale data training process (error backpropagation) is performed. Specifically, different types of non-metallic grounding fault data are analyzed and processed, then fed into the model for fault probability diagnosis analysis. The differences between the diagnosed type parameters and the actual fault type diagnostic parameters are determined, and the weights and correction coefficients are adjusted. This process is repeated until the difference between the model's diagnosed type and the actual non-metallic grounding fault type probabilities converges to an acceptable range, at which point training is complete.

[0036] Furthermore, in S1-1, transient residual current refers to the leakage current to ground that occurs suddenly in a circuit and lasts for an extremely short time (typically from microseconds to milliseconds). Any event that causes a drastic change in circuit voltage within a very short period of time will generate transient residual current. In any electrical circuit and equipment, the phase and neutral wires are not completely insulated from the ground; a tiny "distributed capacitance" exists. Under steady-state power frequency, the capacitive reactance of this capacitor is very large, and the leakage current is very small, as shown in the formula: (1) Where Ic is the current through the capacitor, C is the distributed capacitance to ground, and dV / dt is the rate of change of voltage with respect to time, i.e., how fast the voltage changes. When a voltage surge occurs and dV / dt is very large, even if the distributed capacitance to ground C is small, a huge high-frequency charging or discharging current Ic will be generated instantaneously. This current will flow back to the system through the grounding path, thus forming a transient residual current. Furthermore, in S1-2, the time-domain characteristics describe the characteristics of the residual current signal in the time domain, which can intuitively reflect the basic characteristics and changing trends of the charging pile's residual current. The analysis focuses on the trend of the residual current signal changing with time, its steady-state value, sudden changes (such as during startup and shutdown), and pulse characteristics. The following are some common time-domain characteristics: Average value: The average value is the residual current over one cycle, reflecting the DC component of the residual current. The calculation formula is: (2) in, This represents the average leakage current, where N is the number of sampling points. Let be the residual current value at the i-th sampling point.

[0037] RMS value: For residual current with period N The formula for calculating its effective value is: (3) Electrical equipment is usually designed based on the effective value. When the effective value of the residual current exceeds a certain threshold, it may damage the equipment or cause a safety accident.

[0038] Peak value: peak value This is the maximum value of the residual current within one cycle. It reflects the instantaneous impact of the residual current and is crucial for assessing its stress on electronic components and identifying pulse-type residual currents. The calculation formula is: (4) During the operation of a charging station, external interference such as lightning strikes or the start-up of nearby large equipment may cause a momentary peak in the residual current. By monitoring the peak value, these abnormalities can be detected in a timely manner, and their impact on the safe operation of the charging station can be assessed.

[0039] Waveform factor: Waveform factor It is the ratio of the effective value to the average value of the residual current. It describes the sharpness of the waveform and is used to measure the degree of distortion of the current waveform. The calculation formula is: (5) Waveform factor can serve as an important basis for determining whether a charging pile has harmonic interference or other faults.

[0040] Amplitude and frequency of pulse current: Pulse current refers to transient current that occurs within a short period of time. Its amplitude and frequency can reflect potential problems such as arcing faults in charging piles. By setting an appropriate threshold, the residual current signal is detected. When the current exceeds the threshold, it is considered a pulse current, and its amplitude and frequency are recorded. When an arcing fault occurs, a series of high-frequency pulse currents are generated, and their amplitude and frequency will increase significantly.

[0041] Furthermore, in S1-2, frequency domain feature extraction mainly uses Fast Fourier Transform (FFT) to convert the complex time-domain residual current signal into a set of different frequencies, amplitudes, and sine waves, analyzes its frequency components, and thus obtains information related to the fault.

[0042] The basic principle of the FFT transform is to decompose a time-domain signal into a superposition of multiple sine and cosine waves of different frequencies. For a discrete residual current signal I(n), its Discrete Fourier Transform (DFT) is defined as: (6) in It is a discrete sampled sequence of a time-domain signal, where N is the total number of sample points. It corresponds to frequency The complex spectral components. Where K represents the frequency index (spectral line number), the sequence number of the Kth frequency component, Indicates the actual frequency of the conversion. Indicates the sampling frequency.

[0043] The Fast Fourier Transform (FFT) is a highly efficient computational algorithm for the Distributed Fourier Transform (DFT) and a standard tool in engineering practice. It can significantly reduce computational load and improve computational efficiency. Through the FFT, we can obtain the spectrum of the residual current, intuitively seeing the residual current amplitude at the 50Hz power frequency and its various harmonics (100Hz, 150Hz, ...) as well as higher switching frequencies.

[0044] The main focus is on the frequency domain characteristics of the fundamental frequency (50Hz), harmonic components, and high-frequency components.

[0045] The formula for calculating the fundamental frequency (50Hz) amplitude is: (7) F represents the spectral magnitude (complex amplitude) at the spectral line index corresponding to a frequency of 50Hz, and df represents the frequency resolution. The actual index is 50 / df, which may need to be rounded down. The fundamental amplitude reflects the magnitude of the 50Hz component in the residual current. Under normal circumstances, the fundamental amplitude is relatively stable; when a fault occurs, it may affect the fundamental component, causing its amplitude to change.

[0046] The amplitude of harmonic components is calculated similarly for the second harmonic (100Hz), third harmonic (150Hz), etc. The occurrence of harmonic components is often related to faults or abnormal operating conditions of electrical equipment.

[0047] The high-frequency components divide the spectrum into several frequency bands, such as the low-frequency band (0-50Hz), the mid-frequency band (50-500Hz), and the high-frequency band (500-2000Hz). The energy of each frequency band is calculated using the following formulas: Formula for low-frequency energy: (8) Mid-frequency energy formula: (9) High-frequency energy formula: (10) Where f represents frequency, df represents frequency resolution, and F[f] represents spectral amplitude. The changes in energy in each frequency band can reflect the fault characteristics in different frequency ranges.

[0048] Furthermore, in S1-2, time-frequency domain feature extraction utilizes wavelet transform to analyze the transient residual current in order to obtain its feature information in both time and frequency dimensions. Wavelet transform is a time-frequency analysis method that overcomes the limitation of Fourier analysis, which only has frequency resolution but no time resolution. It possesses both frequency and time resolution, enabling it to characterize the local features of the signal in both the time and frequency domains.

[0049] The basic principle of wavelet transform is to use a mother wavelet function... A series of wavelet basis functions are constructed by scaling and translation. (11) Where a is the scaling factor and b is the translation factor. For a given residual current signal i(t), its continuous wavelet transform (CWT) is defined as: (12) in, These are wavelet transform coefficients. yes The conjugate function of the transient residual current. When analyzing the energy distribution characteristics of the transient residual current, the residual current signal is first decomposed using wavelet decomposition to obtain wavelet coefficients at different scales. Each scale corresponds to a different frequency range. By calculating the energy of the wavelet coefficients at each scale, the energy distribution of the transient residual current in different frequency bands can be obtained. For example, for a residual current signal i(n) containing N sampling points, after wavelet decomposition, wavelet coefficients djn (j = 1, 2, ..., M) at M scales are obtained. Then, the energy Ej at the j-th scale is: (13) Furthermore, in S3-1, the principle of the Min-Max normalization method is to map the eigenvalues ​​to a specified interval, typically [0,1]. Its formula is: (18) Where x is the original feature value. and These are the minimum and maximum values ​​of the feature in the training set, respectively. This formula compresses the value range of each feature to the [0,1] interval, making different features comparable. In the model, for the mean of the residual current Valid value Peak Waveform factor K f The fundamental amplitude A_50, energy values ​​E_low, E_medium, and E_high in the residual current frequency domain characteristics, the mean C_mean, variance C_var, and trend C_trend in the pyrolysis ion concentration characteristics, and the temperature T and humidity H in the environmental factor characteristics are all normalized according to the above formulas. After normalization, all features are at the same dimensional level, which allows them to participate more effectively in subsequent feature fusion and fault type classification, improving the performance and accuracy of the algorithm.

[0050] Furthermore, the root cause of non-metallic grounding faults in S4-1 is the decrease in the impedance of the insulation system. Environmental factors affect the impedance in the following ways: (1) Humidity: Water is an excellent electrolyte that can dissolve ions in pollutants and form conductive channels. (2) Temperature: High temperatures accelerate material aging and chemical reactions, while low temperatures lead to condensation and material embrittlement. (3) Pollutants: Pollen, salt spray, dust, etc. provide charge carriers for leakage current. (4) Biological activities: Plant growth, small animal activities, etc., directly change the external insulation conditions.

[0051] These factors all exhibit significant seasonal and cyclical patterns.

[0052] Compared with existing technologies, the beneficial effects of this invention are as follows: A method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis is proposed. This method constructs a diagnostic method for non-metallic grounding faults in AC charging piles in transformer substations that integrates multi-dimensional residual current information and deeply integrates seasonal environmental factors. It analyzes the transient characteristics and fluctuation trends of residual current; explores pyrolysis ion concentration as an early warning indicator for early overheating decomposition of insulating materials; and uses seasonal environmental parameters as an important correction factor for fault probability, building an intelligent diagnostic model capable of identifying and distinguishing different types of non-metallic grounding faults (such as damp creepage, dry tree branch contact, and oil contamination of the charging gun body). This enables remote monitoring and management of charging piles, improves monitoring accuracy and operation and maintenance efficiency, and ensures the safe operation of charging piles. Attached Figure Description

[0053] Figure 1 This is a flowchart of the algorithm for a non-metallic grounding fault early warning method for charging piles based on residual current characteristic analysis, according to the present invention. Figure 2 This invention presents a method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis, and a system architecture diagram for diagnosing and warning of non-metallic grounding fault types. Figure 3 This invention provides a big data training method for identifying fault types in a non-metallic grounding fault early warning method for charging piles based on residual current characteristic analysis. Figure 4 This is a flowchart illustrating the generation of the prior probability for a non-metallic grounding fault early warning method for charging piles based on residual current characteristic analysis, as per the present invention. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0055] This invention provides a technical solution: a method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis. The main causes of non-metallic grounding faults in AC charging piles are: (1) The charging pile cable, internal wiring harness, charging gun head, etc. are exposed to the outdoors for a long time and are affected by sunlight, rain, ozone, and temperature cycles, which causes the insulation material (such as PVC, TPE, rubber) to harden, become brittle, and crack, and the insulation resistance slowly decreases. This is a gradual process. The residual current is very small at the beginning, and the current gradually increases as the crack expands and becomes damp. (2) In environments with high air humidity (such as rainy season, coastal areas), in places where the temperature difference between day and night is large and condensation is easy to occur, or in industrial areas or coastal areas with serious dust and salt spray pollution. (3) Frequent bending, pulling, and crushing of the charging cable cause minor damage or permanent deformation of the internal insulation layer. Improper insertion and removal of the gun head may also cause internal insulation damage. This damage point may not immediately lead to a short circuit, but it will become a high impedance fault starting point. (4) Under the simultaneous action of electric field and moisture, a tree-like conductive channel will grow inside the insulation material. This is a typical failure mode for polymer insulation materials. Dendration is initially very insidious, with minimal residual current, but it continues to develop and eventually leads to breakdown.

[0056] Based on the above analysis of the causes of non-metallic grounding faults, the main scenarios for non-metallic grounding faults in AC charging piles include oil contamination of the charging gun body, contact with dry tree branches, contact with damp non-metallic materials, water treeing inside the insulation, condensation-type creepage, chemical corrosion leakage, and overheating due to poor contact. This invention primarily focuses on conducting big data research and analysis to construct a fault type diagnostic model for these AC charging pile non-metallic grounding fault scenarios. The steps include: S1: Electrical Data Analysis and Feature Extraction: Analyze electrical data characteristics such as residual current amplitude and duration, and charging power during non-metallic grounding faults in AC charging piles. Utilize data mining and machine learning algorithms to analyze the correlations and patterns of change among the data.

[0057] S1-1: Residual Current Data Characteristic Analysis Grounding faults in AC charging piles are most directly reflected in changes in the "information dimension" of the residual current in the line. The characteristics of the residual current are the "fingerprint" of the fault. By interpreting this fingerprint, accurate fault identification, classification, and early warning can be achieved.

[0058] The study and analysis of the characteristics of residual current mainly focuses on the transient characteristics of residual current through different dimensions such as time domain characteristic extraction, frequency domain characteristic extraction, and time-frequency domain characteristic extraction.

[0059] Transient residual current refers to a sudden, extremely short-duration (typically from microseconds to milliseconds) leakage current to ground in a circuit. It is characterized by being "non-steady-state," with its waveform containing abundant high-frequency oscillations, rapid rise times, and decay processes, which is distinctly different from the stable power frequency (50 / 60Hz) leakage current.

[0060] The generation of transient residual current is essentially the result of the combined effects of sudden voltage changes in the circuit and the system's distributed parameters to ground. Any event that causes a drastic change in circuit voltage within a very short time will generate transient residual current. In any electrical circuit and equipment, the phase and neutral wires are not completely insulated from the ground; a tiny "distributed capacitance" exists. Under steady-state power frequency, this capacitance has a large capacitive reactance, resulting in a very small leakage current.

[0061] Core principle formula: .

[0062] Where Ic is the current through the capacitor. C is the distributed capacitance of the line to ground. dV / dt is the rate of change of voltage with respect to time (i.e., how fast the voltage changes).

[0063] When a voltage surge occurs (dV / dt is very large), even if the capacitance to ground C is small, a huge high-frequency charging or discharging current Ic will be generated instantaneously. This current will flow back to the system through the grounding path, thus forming a transient residual current. The waveform of the transient residual current is usually not a sine wave, but rather a very sharp and rapid current spike.

[0064] S1-2: Residual Current Data Feature Extraction Residual current is not merely a quantitative concept, but also a carrier of information. It requires feature extraction from multiple dimensions. The project extracts features from the residual current values ​​in AC charging pile lines from the perspectives of time domain characteristics, frequency domain characteristics, and time-frequency characteristics.

[0065] Time-domain characteristics describe the properties of the residual current signal in the time domain, intuitively reflecting the basic characteristics and trends of the residual current in a charging pile. The core idea is to analyze the trend, steady-state value, abrupt changes (such as during startup and shutdown), and pulse characteristics of the residual current signal over time. The following are some common time-domain characteristics and their calculation formulas and principles: Mean: The mean is the average value of the residual current over one cycle, reflecting the DC component of the residual current. The calculation formula is: in, This represents the mean, and N is the number of sampling points. This represents the residual current value at the i-th sampling point. Under normal operating conditions, the average residual current of a charging pile is usually close to zero. A significant deviation in the average value may indicate leakage or other faults, such as minor leakage caused by insulation aging, which can gradually increase the average residual current. This measures the DC offset of the signal. It is crucial for determining the presence of smoothed DC residual current (related to the operation of a Type B residual current device). The daily average leakage current reflects the average leakage level of the charging pile's lines over a period of time. If this value continues to rise, it may indicate a gradual decline in the insulation performance of the lines, posing a safety hazard.

[0066] Valid values: Valid values It is a standardized parameter of alternating current or voltage, reflecting the actual work capacity of the residual current and also used to measure the thermal effect of the current. It is the core basis for determining whether a traditional residual current device will operate. For a residual current with a period of N... The formula for calculating its effective value is: .

[0067] Electrical equipment is typically designed based on RMS values. When the RMS value of residual current exceeds a certain threshold, it may damage the equipment or cause a safety accident. For example, when a short circuit occurs inside a charging station, the RMS value of the residual current will increase sharply, far exceeding the normal range.

[0068] Peak value: peak value This is the maximum value of the residual current within one cycle. It reflects the instantaneous impact of the residual current and is crucial for assessing its stress on electronic components and identifying pulse-type residual currents. The calculation formula is: ; During the operation of a charging station, external interference such as lightning strikes or the startup of nearby large equipment may cause a momentary peak in the residual current. By monitoring the peak value, these abnormalities can be detected in a timely manner, and their impact on the safe operation of the charging station can be assessed.

[0069] Waveform factor: The waveform factor is the ratio of the effective value to the average value of the residual current. It describes the sharpness of the waveform and can be used to measure the degree of distortion in the current waveform. The calculation formula is: .

[0070] For a standard sine wave, the waveform factor If K f > 1.11 indicates that the waveform is "sharper" than a sine wave, and may contain abundant higher harmonics or pulses; if K fA value < 1.11 indicates that the waveform is "flatter" than a sine wave, potentially indicating the presence of phenomena such as clipping at the top. When the residual current waveform of a charging pile is distorted, such as due to increased harmonic content caused by the nonlinear characteristics of power electronic devices, the waveform factor will deviate from the standard value. Therefore, the waveform factor can serve as an important basis for determining whether a charging pile has harmonic interference or other faults.

[0071] Amplitude and frequency of pulse current: Pulse current refers to transient current that occurs within a short period of time. Its amplitude and frequency can reflect potential problems such as arcing faults in charging piles. By setting an appropriate threshold, residual current signals are detected. When the current exceeds the threshold, it is considered a pulse current, and its amplitude and frequency are recorded. When an arcing fault occurs, a series of high-frequency pulse currents are generated, and their amplitude and frequency increase significantly. Therefore, monitoring the amplitude and frequency of pulse current helps to detect arcing fault risks in a timely manner, take corresponding protective measures, and ensure the safety of charging piles and personnel.

[0072] Frequency domain feature extraction mainly involves converting complex time-domain residual current signals into sets of different frequencies, amplitudes, and sine waves using Fast Fourier Transform (FFT), analyzing their frequency components, and thus obtaining information related to the fault.

[0073] The basic principle of the FFT transform is to decompose a time-domain signal into a superposition of multiple sine and cosine waves of different frequencies. For a discrete residual current signal I(n), its Discrete Fourier Transform (DFT) is defined as: ; Where I(n) is the discrete sampling sequence of the time-domain signal, N is the total number of sampling points, and X(k) corresponds to the frequency... The complex spectral components. The Fast Fourier Transform (FFT) is a highly efficient computational algorithm for the Distributed Fourier Transform (DFT), a standard tool in engineering practice that significantly reduces computational load and improves efficiency. Through the FFT, we can obtain the spectrum of the residual current, visually observing the residual current amplitude at the 50Hz power frequency and its harmonics (100Hz, 150Hz, ...) as well as higher switching frequencies (such as several kiloHz to tens of kHz).

[0074] We are primarily concerned with the frequency domain characteristics of the fundamental frequency (50Hz), harmonic components, and high-frequency components.

[0075] The formula for calculating the amplitude of the fundamental frequency (50Hz) is: , The actual index is 50 / df, which may need to be rounded. The fundamental frequency amplitude reflects the magnitude of the 50Hz component in the residual current. Under normal circumstances, the fundamental frequency amplitude is relatively stable; however, when a fault occurs, it may affect the fundamental frequency component, causing its amplitude to change.

[0076] The amplitude of harmonic components is calculated similarly for the second harmonic (100Hz), third harmonic (150Hz), etc. The occurrence of harmonic components is often related to faults or abnormal operating conditions of electrical equipment.

[0077] High-frequency components divide the spectrum into several frequency bands, such as the low-frequency band (0-50Hz), the mid-frequency band (50-500Hz), and the high-frequency band (500-2000Hz). The energy of each frequency band is calculated using the following formulas: Formula for low-frequency energy: , Mid-frequency energy formula: , High-frequency energy formula: .

[0078] Changes in energy across different frequency bands can reflect fault characteristics within different frequency ranges.

[0079] Time-frequency domain feature extraction utilizes wavelet transform to analyze transient residual currents, obtaining their characteristic information in both time and frequency dimensions. Wavelet transform is a time-frequency analysis method that overcomes the limitation of Fourier analysis, which only has frequency resolution but no time resolution. It possesses both frequency and time resolution, enabling the characterization of local signal features in both the time and frequency domains.

[0080] The basic principle of wavelet transform is to use a mother wavelet function... A series of wavelet basis functions are constructed by scaling and translation. ; Where a is the scaling factor and b is the translation factor. For a given residual current signal i(t), its continuous wavelet transform (CWT) is defined as: .

[0081] Where Wf(a,b) are wavelet transform coefficients. yes The conjugate function of . In practical applications, the Discrete Wavelet Transform (DWT) is usually used, which discretizes the scaling factor a and the translation factor b.

[0082] When analyzing the energy distribution characteristics of transient residual current, the residual current signal is first decomposed using wavelet decomposition to obtain wavelet coefficients at different scales. Each scale corresponds to a different frequency range. By calculating the energy of the wavelet coefficients at each scale, the energy distribution of the transient residual current in different frequency bands can be obtained. For example, for a residual current signal i(n) containing N sampling points, after wavelet decomposition, wavelet coefficients djn (j = 1,2,..., M) at M scales are obtained. Then, the energy Ej at the j-th scale is: .

[0083] Time-frequency domain characteristics play a crucial role in charging pile fault diagnosis. Transient residual currents often contain rich fault information. For example, during sudden grounding faults or arcing faults, the residual current undergoes rapid changes, which are reflected in both the time and frequency domains. Wavelet transform can simultaneously capture these local features in both time and frequency domains. By analyzing the energy distribution characteristics of the transient residual current, the timing, duration, and type of fault can be accurately determined. For instance, during an arcing fault, the transient residual current exhibits energy concentration at high frequencies. Analyzing the energy distribution characteristics using wavelet transform allows for timely detection and diagnosis of arcing faults, providing strong assurance for the safe operation of charging piles.

[0084] S2: Pyrolysis Ion Data Analysis and Feature Extraction: Analyze the amplitude and duration characteristics of ambient pyrolysis ion concentration under different types of non-metallic grounding faults in AC charging piles. Construct a weighted analysis of the proportion of pyrolysis ion data in the diagnosis of non-metallic grounding fault types based on big data mining.

[0085] S2-1: Characteristics of pyrolysis ion concentration The characteristics of pyrolysis ion concentration can reflect the overheating condition of electrical equipment and play an important role in judging faults such as contact overheating in non-metallic grounding faults.

[0086] The formula for calculating the mean is: , Where C_mean represents the mean concentration of pyrolysis ions, M is the sampling period, and C_k is the k-th sample value. It represents the average level of pyrolysis ion concentration over a period of time. When electrical equipment is operating normally, the mean concentration of pyrolysis ions is usually at a low level and relatively stable; if the mean increases, it may mean that the equipment is overheating, leading to an increase in the generation of pyrolysis particles.

[0087] The formula for calculating variance is: , Where C_var is the variance, M is the sampling period, C_k is the Kth sample value, and C_mean is the mean of the pyrolysis ion concentration. It measures the dispersion of the pyrolysis ion concentration around the mean. A larger variance indicates greater fluctuations in the pyrolysis ion concentration, potentially suggesting unstable overheating or intermittent overheating in the equipment. For example, loose contact points in the equipment can lead to unstable contact resistance, thus increasing the variance of the pyrolysis ion concentration.

[0088] Concentration trend: C_trend (linear regression slope) is the slope obtained through linear regression, representing the rate of change of concentration over time, thus indicating the trend. If the pyrolysis ion concentration shows an upward trend, it indicates that the overheating of the equipment is gradually worsening, which may be a signal of a developing fault. For example, during the aging process of equipment insulation, the pyrolysis ion concentration will gradually increase over time. By monitoring its trend, signs of fault development can be detected in time, allowing for preventive and handling measures to be taken in advance.

[0089] S2-2: Environmental Feature Extraction and Correction Factors Environmental factors such as temperature and humidity have a significant impact on the operating status of electrical equipment and should not be ignored in the identification of non-metallic grounding fault types.

[0090] You can use current measurements of temperature (T) and humidity (H) directly, or you can use averages over a period of time. In practical applications, averages over a period of time can more comprehensively reflect the long-term effects of environmental factors and reduce the interference of instantaneous fluctuations. For example, when analyzing the insulation performance of equipment, considering the average humidity over a period of time can more accurately assess the impact of humidity on insulation performance.

[0091] Temperature correction factor: .

[0092] Humidity correction factor: S3: Data Normalization and Weighted Fusion: Normalize and weight data of different types and dimensions to form data parameters of a uniform magnitude, and finally construct feature vectors. Integrate feature data of multiple dimensions to finally establish an initial non-metallic grounding fault diagnosis model for charging piles.

[0093] S3-1: Feature Normalization and Weighted Fusion Because the dimensions and value ranges of different characteristics of residual current, such as time-domain characteristics, frequency-domain characteristics, pyrolysis ion concentration characteristics, and environmental factor characteristics, vary significantly, directly combining these characteristics can lead to some characteristics dominating the analysis while the role of others is ignored, thus affecting the accuracy of fault type identification. To address this issue, we employ the Min-Max normalization method to process each characteristic.

[0094] The principle of the Min-Max normalization method is to map eigenvalues ​​to a specified interval, typically [0,1]. Its formula is: , Where x is the original feature value, and x_min and x_max are the minimum and maximum values ​​of the feature in the training set, respectively. This formula compresses the value range of each feature to the interval [0,1], making different features comparable. In the model, for the mean I(avg) and effective value of the residual current... Peak The waveform factor Kf, the fundamental amplitude A_50, and the energies E_low, E_medium, and E_high in the residual current frequency domain characteristics, the mean C_mean, variance C_var, and trend C_trend in the pyrolysis ion concentration characteristics, and the temperature T and humidity H in the environmental factor characteristics are all normalized according to the above formulas. After normalization, all features are at the same dimensional level, which can more effectively participate in subsequent feature fusion and fault type classification, improving the performance and accuracy of the algorithm.

[0095] S3-2: Constructing Feature Vectors Constructing a feature vector involves integrating features from multiple dimensions so that subsequent classification algorithms can comprehensively utilize this information to determine the fault type. We combine all the normalized features mentioned above into a single feature vector X, which has the following form: By combining different types of features into a feature vector, the operating status and potential fault information of the power system can be comprehensively reflected. Each feature in the feature vector represents a specific dimension, and they complement each other, jointly providing a basis for fault type judgment. For example, the time-domain and frequency-domain characteristics of residual current can reflect changes in electrical quantities, the pyrolysis ion concentration characteristics can reflect equipment overheating, and environmental factor characteristics can reflect the impact of the external environment on equipment operation. Integrating these features together can more comprehensively and accurately describe the characteristics of faults, improving the accuracy and reliability of fault type identification. The construction of the feature vector provides a unified data format for subsequent fault type classification, facilitating processing and analysis by classification algorithms.

[0096] S4: Quantitative Correction of Seasonal Environmental Factors: The non-metallic grounding fault diagnosis model for charging piles is optimized by introducing seasonal environmental parameters. Deep integration of a series of environmental factors improves the diagnostic accuracy of non-metallic grounding fault types.

[0097] S4-1 Solar Term Environmental Quantitative Analysis Non-metallic grounding faults in AC charging piles are strongly correlated with seasonal environmental factors. Unlike sudden metallic short circuits, non-metallic grounding faults are essentially a gradual electrochemical-physical process driven by the environment. The root cause of non-metallic grounding faults is a decrease in the impedance of the insulation system, and environmental factors affect the impedance in the following ways: (1) Humidity: Water is an excellent electrolyte that can dissolve ions in pollutants and form conductive channels.

[0098] (2) Temperature: High temperature accelerates material aging and chemical reaction, while low temperature leads to condensation and material embrittlement.

[0099] (3) Pollutants: pollen, salt spray, dust, etc. provide carriers for leakage current.

[0100] (4) Biological activities: plant growth, small animal activities, etc. directly change the external insulation conditions.

[0101] These factors all exhibit significant seasonal and cyclical patterns.

[0102] Seasonal environmental factors are not independent diagnostic indicators, but rather serve as "physical mechanism catalysts" and "diagnostic logic correctors," improving the diagnostic accuracy of non-metallic grounding fault types through the deep integration of a series of environmental factors.

[0103] Deeply integrating seasonal environmental factors can dynamically change the residual current threshold for non-metallic grounding faults. Using environmental factors as context and correction factors in the diagnostic model makes the diagnosis more consistent with the actual physical world. For example, the residual current alarm threshold for non-metallic grounding faults is usually higher during the rainy season than during the dry winter; introducing seasonal environmental factors can improve detection sensitivity. The warning thresholds for residual current and ion concentration are adaptively adjusted based on real-time environmental conditions. First, a large amount of data under normal conditions is collected to establish a regression model of environmental factors and background leakage current. The formula for normal background leakage current is: , in is the predicted normal background leakage current, RH is the relative humidity, T is the temperature, SeasonCode is the seasonal code, and β0, β1, β2, and β3 are regression coefficients.

[0104] Dynamic threshold calculation formula: , Where I_alarm is the dynamic alarm threshold, I_background is the predicted normal background leakage current, K is the safety factor (usually taken as 2-3), and σ is the standard deviation of the normal data and the residual of the regression model.

[0105] The most important benefit of deeply integrating seasonal environmental factors is that it can improve the accuracy of diagnosing non-metallic grounding fault types in AC charging piles.

[0106] First, an environmental severity index is constructed, integrating multiple environmental parameters into a comprehensive indicator to quantify the severity of the current environment. Calculation formula: , Where T, H, and Ppollutant represent the current temperature, humidity, and pollution concentration, respectively; Tref, Href, and Pref represent environmental reference values ​​(e.g., 25°C, 60%, clean air); ωT, ωH, and ωP represent weighted indices, determined by fitting historical data, typically ωH>ωT>ωP; Sseason represents the seasonal factor (rainy season=1.3, summer=1.1, winter=0.7), and the values ​​will be updated in the future based on historical big data analysis.

[0107] Secondly, a dynamic weight allocation strategy is implemented to automatically adjust the weights of different features in the diagnostic model based on ESI. The formula for calculating the electrical feature weights is as follows:

[0108] Formula for calculating the weight of environmental features: , Where We0 and Wenv0 are the basic weights (e.g., 0.7 and 0.3), and their values ​​will be updated and adjusted in the future based on historical big data analysis; λ and μ are the adjustment coefficients (e.g., 0.2), which control the magnitude of weight changes, and their values ​​will be updated and adjusted in the future based on historical big data analysis.

[0109] The environmental adaptive diagnostic threshold is applied again, allowing it to change dynamically with the environment, thus addressing the poor adaptability of fixed thresholds. Diagnostic threshold formula: , Where THbase is the baseline threshold under standard conditions; k: adaptability coefficient. For damp creepage, k > 0 (threshold is increased to avoid alarming for normal fluctuations in humid weather), and for overheating faults, k < 0 (threshold is decreased to improve sensitivity in hot weather); ESIref is the environmental index.

[0110] Finally, a Bayesian correction is performed on the failure probability, using the environment as prior knowledge to calculate the posterior probability of a specific failure. The calculation formula is as follows: , Where P(FaultType|Data) is the probability of a certain fault given the data (the final output of the model); P(Data|FaultType) represents the probability of data characteristics (based on historical big data analysis). P(FaultType|Season) is the prior probability, that is, the general probability of this type of fault occurring in a certain season. It needs to be derived from historical big data analysis of different geographical locations.

[0111] S4-2 Fault Type Diagnosis First, an integrated learning framework is used, employing Gradient Boosting Decision Tree (GBDT) as the base classifier: inputting normalized feature data, outputting the probability of occurrence of eight typical non-metallic grounding faults: normal state, gun body oil contamination, dry tree branch contact, damp non-metallic contact, water tree inside insulation, condensation creepage, chemical corrosion leakage, and poor contact overheating.

[0112] Secondly, for sequence data, LSTM-attention mechanism is used for deep learning. Hidden state formula: , Where ht represents the hidden state of the LSTM at time t. It is the output at time t.

[0113] The optimization strategy employs loss function design and imbalanced data handling. A multi-task learning loss function is used: , Where λ1, λ2, and λ3 are task weight coefficients that balance the importance of different loss terms; Lclassification is the classification loss, Lseverity is the severity assessment loss, and Learned warning is the early warning loss.

[0114] Finally, a large-scale data training process (error backpropagation) is performed. Specifically, different types of non-metallic grounding fault data are analyzed and processed, then fed into the model for fault probability diagnosis analysis. The differences between the diagnosed type parameters and the actual fault type diagnostic parameters are determined, and the weights and correction coefficients are adjusted. This process is repeated until the difference between the model's diagnosed type and the actual non-metallic grounding fault type probabilities converges to an acceptable range, at which point training is complete.

[0115] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis, characterized in that: Includes the following steps: S1: Electrical Data Analysis and Feature Extraction: Analyze electrical data characteristics such as residual current amplitude and duration, and charging power during non-metallic grounding faults in AC charging piles. Utilize data mining and machine learning algorithms to analyze the correlations and patterns of change among the data. S2: Pyrolysis Ion Data Analysis and Feature Extraction: Analyze the amplitude and duration data characteristics of surrounding pyrolysis ion concentration under different types of non-metallic grounding faults in AC charging piles, and construct the weight of pyrolysis ion data in the diagnosis of non-metallic grounding fault types based on big data mining; S3: Data Normalization and Weighted Fusion: Normalize and weight data of different types and dimensions to form data parameters of a uniform magnitude, construct feature vectors, integrate feature data of multiple dimensions, and finally establish an initial charging pile non-metallic grounding fault diagnosis model. S4: Quantitative Correction of Seasonal Environmental Factors: By introducing seasonal environmental parameters to optimize the diagnostic model for non-metallic grounding faults in charging piles, a series of environmental factors are deeply integrated to improve the diagnostic accuracy of non-metallic grounding fault types. S4-1 Solar Term Environmental Quantitative Analysis Establish a regression model for environmental factors and background leakage current. The formula for normal background leakage current is: in This represents the predicted normal background leakage current, where RH is the relative humidity, T is the temperature, SeasonCode is the seasonal code, and β0, β1, β2, and β3 are regression coefficients. Dynamic threshold calculation formula: in It is a dynamic alarm threshold. This is the predicted normal background leakage current, K is the safety factor, usually taken as 2-3, and σ is the standard deviation of the normal data and the regression model residuals. First, an environmental severity index is constructed, integrating multiple environmental parameters into a comprehensive index to quantify the severity of the current environment. The calculation formula is: Where T, H, P pollutant These represent the current temperature, humidity, and pollution concentration, respectively; T ref H ref ,P ref ωT represents the environmental reference baseline value; ωH, ωP represent weight indices, determined through historical data fitting, typically ωH>ωT>ωP; Sseason represents the seasonal factor, which will be updated based on historical big data analysis in the future; secondly, a dynamic weight allocation strategy is implemented, automatically adjusting the weights of different features in the diagnostic model according to ESI. The formula for calculating the electrical feature weights is as follows: Formula for calculating the weight of environmental features: Where We0 and Wenv0 are the basic weights, which will be updated and adjusted based on historical big data analysis; λ and μ are adjustment coefficients that control the magnitude of weight changes, and will also be updated and adjusted based on historical big data analysis; an environmental adaptive diagnostic threshold is then applied, allowing the diagnostic threshold to change dynamically with the environment, solving the problem of poor adaptability of fixed thresholds. The diagnostic threshold formula is: Among them TH base It is the baseline threshold under standard conditions; k is the fitness coefficient; ESI ref It is an environmental index; Finally, a Bayesian correction of the failure probability is performed, using the environment as prior knowledge to calculate the posterior probability of a specific failure. The calculation formula is as follows: Where P(FaultType|Data) is the probability of a certain fault given the data, and is the final output of the model; P(Data|FaultType) represents the probability of a data feature, a value based on historical big data analysis. P(FaultType|Season) is the prior probability, which is the general probability of this type of fault occurring in a certain season. The probability value needs to be obtained based on historical big data analysis of different geographical locations. S4-2 Fault Type Diagnosis First, an integrated learning framework is used, with Gradient Boosting Decision Tree (GBDT) as the base classifier: input normalized feature data, output the probability of occurrence of 8 typical non-metallic grounding faults: normal state, gun body oil pollution, dry tree branch contact, damp non-metallic contact, water tree inside insulation, condensation creepage, chemical corrosion leakage, poor contact and overheating. Secondly, for sequence data, deep learning is performed using the LSTM-attention mechanism, with the hidden state formula as follows: Where ht: The hidden state of the LSTM at time t. It is the output at time t; The optimization strategy employs loss function design and imbalanced data processing, using a multi-task learning loss function: Where λ1, λ2, and λ3 are task weight coefficients, balancing the importance of different loss terms; Lclassification is the classification loss, Lseverity is the severity assessment loss, and Learned warning is the early warning loss; finally, a big data training process is carried out. The specific steps are to analyze and process different types of non-metallic grounding fault data and then input them into the model for fault probability diagnosis analysis, obtain the difference between the diagnostic type parameters and the actual fault type diagnostic parameters, adjust the weights and correction coefficients, and then carry out big data training again. This process is repeated until the difference between the model's diagnostic type and the actual non-metallic grounding fault type probability converges to an acceptable range, and the training is completed.

2. The method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis according to claim 1, characterized in that: S1 includes the following steps: S1-1: Residual Current Data Characteristic Analysis Grounding faults in AC charging piles are most directly reflected in changes in the "information dimension" of the residual current in the line. The characteristics of the residual current are the "fingerprint" of the fault. By interpreting this fingerprint, accurate identification, classification, and early warning of faults can be achieved. S1-2: Residual Current Data Feature Extraction The residual current value in the AC charging pile line is extracted from the time domain characteristics, frequency domain characteristics, and time-frequency characteristics.

3. The method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis according to claim 2, characterized in that: S2 includes the following steps: S2-1: Analysis of pyrolysis ion concentration characteristics The formula for calculating the mean is: in This represents the mean concentration of pyrolysis ions, where M is the sampling period. For the k-th sample value, The formula for calculating variance is: in This is the variance value, where M is the sampling period. For the Kth sample value, The mean of the pyrolysis ion concentration measures the degree of dispersion of the pyrolysis ion concentration around the mean. S2-2: Environmental feature extraction and correction factors, temperature correction factor: Where Tenv represents the average temperature over a period of time. Humidity correction factor: H-env represents the average humidity over a period of time.

4. The method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis according to claim 3, characterized in that: S3 includes the following steps: S3-1: Feature Normalization and Weighted Fusion The Min-Max normalization method is used to process each feature. Min-Max normalization maps feature values ​​to a specified interval, typically [0,1]. The formula is as follows: Where x is the original feature value, and These are the minimum and maximum values ​​of the feature in the training set, respectively. This formula compresses the value range of each feature to the interval [0,1]. S3-2: Constructing Feature Vectors Constructing a feature vector involves integrating features from multiple dimensions so that subsequent classification algorithms can comprehensively utilize this information to determine the fault type. All the normalized features are combined into a single feature vector X, which takes the form:

5. The method for early warning of non-metallic grounding faults in charging piles based on residual current characteristic analysis according to claim 4, characterized in that: The Min-Max normalization method, in the model, for the mean I(avg) and effective value I of the residual current... △(RMS) Peak I △(peak) Waveform factor K f The fundamental amplitude A_50, the energy of each frequency band E_low, E_medium, and E_high in the residual current frequency domain characteristics, the mean C_mean, variance C_var, and trend C_trend in the pyrolysis ion concentration characteristics, and the temperature T and humidity H in the environmental factor characteristics are all normalized according to the above formulas. After normalization, all features are at the same dimensional level, which can more effectively participate in subsequent feature fusion and fault type classification, thereby improving the performance and accuracy of the algorithm.

Images