Machine learning-based new energy equipment charging anomaly detection method and system

Abstract

This invention proposes a machine learning-based method and system for detecting charging anomalies in new energy equipment, relating to the field of charging pile testing technology. The method includes: acquiring the original test data of the charging pile under test; identifying the internal auxiliary circuit located after the residual current sensor within the charging pile and forming an unexpected leakage path between the charging circuit and ground; determining the original test deviation introduced by the internal auxiliary circuit; constructing an environmental coupling correction function based on environmental characteristic parameters to correct the original test deviation and obtain the final test deviation; obtaining the corrected residual current action value; calibrating the associated performance parameters of the charging pile under test based on the corrected residual current action value to obtain a corrected set of performance parameters; and generating a performance test report for the charging pile under test based on the corrected set of performance parameters. This achieves a leap from single-point isolated measurement in charging pile residual current testing to a high-precision, full-lifecycle adaptive, and globally comparable performance evaluation system.

Description

Technical Field

[0001] This application relates to the field of charging pile performance testing technology, and in particular to a method and system for detecting charging anomalies in new energy equipment based on machine learning. Background Technology

[0002] With the rapid development of the new energy vehicle industry, charging piles, as critical infrastructure, directly impact user experience and power grid safety in terms of performance and reliability. In the daily operation and maintenance of charging piles, the residual current operating value is one of the important indicators for measuring their safety performance. Currently, testing the residual current of charging piles mainly relies on directly collecting data from built-in sensors and combining this data with performance evaluation in a standard testing environment.

[0003] However, existing testing methods have the following problems in practical applications: First, charging piles integrate various functional modules, which are typically considered ideal components with no impact on testing during routine testing. However, under actual operating conditions, especially under high-frequency switching or external electromagnetic interference, these modules can have a significant impact on the measurement results of the residual current sensor. Existing testing standards and methods are mostly designed based on steady-state or ideal operating conditions, making it difficult to effectively eliminate this measurement deviation caused by the internal circuitry's operating state, resulting in discrepancies between the test results and the intrinsic performance of the charging pile.

[0004] Secondly, charging stations are mostly deployed in outdoor or semi-outdoor environments, and their operating environment exhibits significant time-varying characteristics. Environmental conditions vary considerably across different regions, seasons, and even different times of day. Current technologies typically employ fixed environmental compensation parameters or simplified linear correction models for performance testing, often resulting in data that fluctuates unexpectedly. This fluctuation leads to a lack of comparability in test results for the same charging station at different times, making horizontal performance comparisons of charging stations in different areas difficult and impacting the accuracy of operational and maintenance decisions.

[0005] Furthermore, with the large-scale network operation of charging piles, multiple charging piles within the same power supply area are electrically coupled through the power grid topology. When a charging pile is tested, its test results may be affected by the operating status of adjacent charging piles. However, current testing methods still mainly focus on isolated single-point testing, failing to effectively utilize the collaborative information of multiple charging piles within the area, resulting in a lack of dynamic and global reference benchmarks for evaluating the performance of individual charging piles. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a machine learning-based method and system for detecting charging anomalies in new energy equipment. By identifying and quantifying internal circuit parasitic parameter interference, constructing a temperature and humidity coupled nonlinear correction mechanism, and introducing regional electrical collaborative analysis, it achieves a leap from single-point isolated measurement of charging pile residual current testing to a high-precision, full-lifecycle adaptive, and globally comparable performance evaluation system, providing accurate data support for the efficient operation and maintenance of charging networks.

[0007] In a first aspect, the present invention provides a method for detecting charging anomalies in new energy devices based on machine learning, comprising: Acquire the raw test data of the charging pile under test directly collected by the residual current sensor, wherein the raw test data includes at least the raw residual current action value; Identify the internal auxiliary circuit located after the residual current sensor inside the charging pile under test, which forms an unexpected leakage path between the charging circuit and ground; based on the influence of the internal auxiliary circuit on the test process under complex electromagnetic environment, and combined with its parasitic parameters, determine the original test deviation introduced by the internal auxiliary circuit. Obtain the environmental characteristic parameters of the charging pile under test; construct an environmental coupling correction function based on the environmental characteristic parameters, and perform nonlinear correction on the original test deviation to obtain the final test deviation. Based on the final test deviation, the original residual current action value is corrected to obtain the corrected residual current action value. Based on the corrected residual current action value, the associated performance parameters of the charging pile under test are calibrated to obtain the corrected performance parameter set. Based on the revised set of performance parameters, a performance test report for the charging pile under test is generated.

[0008] Preferably, the internal auxiliary circuit includes at least a switch-type isolation circuit connected between the charging circuit and ground; the determination of the original test deviation introduced by the internal auxiliary circuit based on its influence on the test process under complex electromagnetic environment and its parasitic parameters includes: A high-frequency equivalent model of the internal auxiliary circuit is constructed, and its parasitic capacitance and parasitic inductance parameters are extracted. Under the first operating condition, an interference signal with a preset spectrum is applied to obtain the first residual current action value; Under the second condition that the internal auxiliary circuit is deactivated by physical disconnection or software shielding, an interference signal of the preset spectrum is applied to obtain a second residual current action value. Based on the parasitic parameters, the difference between the first residual current operating value and the second residual current operating value, the original test deviation function, which is introduced by the internal auxiliary circuit and varies with the interference frequency, is calculated.

[0009] Preferably, constructing the high-frequency equivalent model of the internal auxiliary circuit includes: A wideband impedance test was performed on the internal auxiliary circuit to obtain its impedance characteristic curve; Based on the impedance characteristic curve, the parasitic capacitance, parasitic inductance, and parasitic resistance that vary with frequency are extracted by the equivalent circuit fitting method. The extracted parasitic parameters are substituted into the equivalent circuit topology to construct a lumped parameter equivalent model containing parasitic elements, and the model is verified by comparing it with the measured impedance curve.

[0010] Preferably, the environmental characteristic parameters include the average humidity and average temperature within a preset time period; the step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation includes: Based on a pre-built device aging database, the multi-physics field attenuation coefficient of key semiconductor devices in the internal auxiliary circuit is determined according to the average humidity and average temperature. The multi-physics field attenuation coefficient is obtained by performing high temperature and high humidity reverse bias accelerated aging tests on devices of the same model and fitting the results with a power law relationship model. Based on the multiphysics attenuation coefficient, an environmental coupling correction function is constructed, which includes a humidity-temperature cross-coupling term. The original test deviation is input into the environmental coupling correction function, and the final test deviation is obtained through nonlinear calculation.

[0011] Preferably, the method further includes: Obtain the corrected residual current action values ​​of multiple charging piles within the area where the charging pile under test is located; Based on the physical location of each charging pile in the area, the power grid topology connection relationship and historical load data, calculate the electrical coupling weight matrix between charging piles in the area; Based on the electrical coupling weight matrix, the corrected residual current action values ​​of each charging pile in the region are weighted and fused to generate dynamic statistical reference values ​​for the region. The corrected residual current action value of the charging pile under test is compared with the dynamic statistical reference value to generate a trend performance test report that includes horizontal comparative analysis.

[0012] Preferably, the step of calculating the electrical coupling weight matrix between charging piles within the region based on the physical location of each charging pile, the power grid topology, and historical load data includes: Obtain the physical coordinates of each charging pile in the area, the length of the cable connected to the same busbar, and the impedance parameters; Obtain the load current data of each charging pile in the area in a continuous time series, and calculate the load fluctuation similarity between any two charging piles; Based on the cable length, impedance parameters, and load fluctuation similarity, an electrical coupling weight matrix element is constructed. Among them, the more similar the load fluctuations of any two pile pairs are and the closer their electrical distance is, the greater their corresponding weight.

[0013] Preferably, the dynamic statistical reference value is obtained by the following formula; including:

[0014] t represents the dynamic statistical reference value at time t; m represents the number of charging piles in the area. This is the corrected current action value for the nth charging pile; The dynamic simultaneity coefficient function is calculated based on the real-time load current of the nth and jth charging piles at time t.

[0015] Optionally, the step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation further includes: Obtain the historical test data and corresponding historical environmental parameters of the charging pile under test; Using the historical environmental parameters as input and the deviation correction amount extracted from the historical test data as output, a deep neural network model is trained. The current environmental feature parameters are input into the trained deep neural network model to predict the correction coefficient under the current environment, and the final test deviation is dynamically generated by combining the original test deviation.

[0016] Preferably, the method further includes: The generated final test deviation is compared with a preset deviation alarm threshold; If the final test deviation exceeds the deviation alarm threshold, an early warning message containing the charging pile identifier, the deviation item, and the recommended maintenance time is generated and pushed to the operation and maintenance terminal.

[0017] Secondly, the present invention provides a machine learning-based new energy equipment charging anomaly detection system, used in the machine learning-based new energy equipment charging anomaly detection method of the embodiments of the present invention, including: The data acquisition module is used to acquire the raw test data of the charging pile under test directly collected by the residual current sensor and the environmental characteristic parameters of the charging pile under test. The interference identification and modeling module is used to identify internal auxiliary circuits and build their high-frequency equivalent models to determine the original test deviation. The environmental coupling correction module is used to correct the original test deviation based on environmental characteristic parameters using a nonlinear correction model to obtain the final test deviation. The parameter correction module is used to correct the original residual current action value based on the final test deviation, and to calibrate the associated performance parameters based on the corrected residual current action value. The regional collaborative analysis module is used to generate dynamic statistical reference values ​​based on the electrical coupling relationship of charging piles within the region; The report generation and early warning module is used to generate performance test reports and issue early warnings when performance parameters exceed acceptable limits.

[0018] The beneficial effects of this invention include at least the following: By identifying and performing high-frequency equivalent modeling on the internal auxiliary circuits of the charging pile, test interference introduced by parasitic parameters can be effectively removed, and its impact on the residual current sensor can be quantified. Compared with existing technologies that rely solely on raw data acquisition or simple filtering, this invention achieves refined elimination of internal circuit interference, making the residual current action value closer to the true intrinsic characteristics of the charging pile, and providing a more accurate data foundation for subsequent performance evaluation. For the actual working conditions of charging piles exposed to complex outdoor environments for extended periods, this invention introduces a nonlinear correction mechanism based on the coupling of multiple physical fields such as temperature and humidity. By modeling the aging characteristics of devices and dynamically compensating for environmental features, the problem of test data fluctuations caused by environmental factors in existing technologies is effectively solved. Using the method of this invention, test results obtained from the same charging pile under different seasons and geographical conditions have better consistency and comparability, and can truly reflect the performance change patterns of the charging pile throughout its entire life cycle. By constructing an electrical coupling weight matrix between charging piles within a region and combining it with dynamic simultaneity coefficients for weighted fusion, this invention generates dynamic statistical reference values ​​with global reference significance. By comparing the results of individual pile tests with the reference value, abnormal pile positions that deviate from the normal level of the area can be identified, providing maintenance personnel with a basis for trend analysis. Compared with the existing isolated single-point testing mode, this invention realizes the leap from point evaluation to surface analysis, and enhances the decision-making value of test data. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a new energy equipment charging anomaly detection method based on machine learning, according to an embodiment of this application. Detailed Implementation

[0020] The present application will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0021] Example 1, see Figure 1 This embodiment provides a machine learning-based method for detecting charging anomalies in new energy devices, including: Acquire the raw test data of the charging pile under test directly collected by the residual current sensor, wherein the raw test data includes at least the raw residual current action value; Identify the internal auxiliary circuit located after the residual current sensor inside the charging pile under test, which forms an unexpected leakage path between the charging circuit and ground; based on the influence of the internal auxiliary circuit on the test process under complex electromagnetic environment, and combined with its parasitic parameters, determine the original test deviation introduced by the internal auxiliary circuit. Obtain the environmental characteristic parameters of the charging pile under test; construct an environmental coupling correction function based on the environmental characteristic parameters, and perform nonlinear correction on the original test deviation to obtain the final test deviation. Based on the final test deviation, the original residual current action value is corrected to obtain the corrected residual current action value. Based on the corrected residual current action value, the associated performance parameters of the charging pile under test are calibrated to obtain the corrected performance parameter set. Based on the revised set of performance parameters, a performance test report for the charging pile under test is generated.

[0022] Through the above steps, the root cause of test deviation is accurately identified. The test deviation caused by unexpected leakage paths in the internal auxiliary circuit after the residual current sensor, complex electromagnetic environment, and parasitic parameters is quantified and corrected in a targeted manner. This solves the problem of test data distortion caused by neglecting such internal circuit interference in traditional testing, and greatly improves the accuracy of residual current action value testing.

[0023] An environmental coupling correction function is introduced to perform nonlinear correction on the original deviation, so that the deviation correction fits the environmental characteristics of the actual service of the charging pile, avoids correction errors caused by environmental factors, and further ensures the matching degree between the test data and the actual working conditions.

[0024] By using the corrected residual current operating value as a benchmark to uniformly calibrate the associated performance parameters, the core test parameters and associated performance parameters are synergistically corrected, forming a complete and accurate set of performance parameters, so that the test report can comprehensively and truthfully reflect the actual performance status of the charging pile.

[0025] The entire testing method revolves around internal circuit interference and environmental influences to construct a full-process deviation correction system, which makes up for the lack of consideration of such hidden interference factors in traditional charging pile performance testing, and provides accurate and reliable technical basis for charging pile performance evaluation, fault diagnosis and operation and maintenance optimization.

[0026] In one possible implementation, the internal auxiliary circuit includes at least a switch-type isolation circuit connected between the charging circuit and ground; the determination of the original test deviation introduced by the internal auxiliary circuit based on its influence on the test process under complex electromagnetic environments, combined with its parasitic parameters, includes: A high-frequency equivalent model of the internal auxiliary circuit is constructed, and its parasitic capacitance and parasitic inductance parameters are extracted. Under the first operating condition, an interference signal with a preset spectrum is applied to obtain the first residual current action value I1(f); Under the second condition that the internal auxiliary circuit is deactivated by physical disconnection or software shielding, an interference signal with the preset spectrum is applied to obtain the second residual current action value I2(f). Based on the parasitic parameters, the difference between the first residual current operating value and the second residual current operating value, the original test deviation function ΔI(f), which is introduced by the internal auxiliary circuit and varies with the interference frequency, is calculated.

[0027] In one possible implementation, constructing a high-frequency equivalent model of the internal auxiliary circuit and extracting its parasitic capacitance and inductance parameters includes: A wideband impedance test was performed on the internal auxiliary circuit to obtain its impedance characteristic curve; Based on the impedance characteristic curve, the parasitic capacitance C, which varies with frequency, is extracted using the equivalent circuit fitting method. p (f) Parasitic inductance L p (f) and parasitic resistance R p (f); The extracted parasitic parameters are substituted into the equivalent circuit topology to construct a lumped parameter equivalent model containing parasitic elements, and the model is verified by comparing it with the measured impedance curve.

[0028] In one specific implementation, the internal auxiliary circuit is a switch-type isolation circuit, and the construction of its high-frequency equivalent model includes the following steps: Step 1: Circuit Topology Deconstruction. Deconstruct the physical structure of the switch-type isolation circuit, clarify the connection relationships of core components (such as optocouplers, MOSFETs, isolation transformers, and filter capacitors / resistors), draw the topology diagram, and label the nominal parameters of each component (such as capacitance, resistance, and transformer turns ratio). By reverse engineering the circuit board or consulting the design drawings, clarify the electrical connection relationships between the components, providing a topological foundation for the subsequent extraction of parasitic parameters and the construction of equivalent models.

[0029] Step 2: Use a network analyzer or impedance analyzer to perform a wideband impedance scan test on the internal auxiliary circuit in the frequency range of 1kHz to 1GHz; set the charging circuit end as the test port and the ground end as the reference ground, and measure the input impedance of the port relative to ground Z(f) = R(f) + jX(f).

[0030] Based on the measured impedance characteristics, parasitic parameters are extracted using an equivalent circuit fitting method: Parasitic capacitance C p Based on the capacitive characteristics of the impedance in the low-frequency band (<1MHz), the equivalent capacitance value is calculated from the imaginary part X(f), or the lumped capacitance parameter is obtained by fitting. Parasitic inductance L p Based on the inductive characteristics of impedance in the high-frequency band (>10MHz), the equivalent inductance value is calculated from the slope of the imaginary part X(f), or the lumped inductance parameter is obtained by fitting. Parasitic resistance R p Based on the frequency response characteristics of the real part of the impedance R(f), the equivalent series resistance corresponding to the high-frequency skin effect and dielectric loss is extracted.

[0031] The extraction process needs to be combined with open circuit / short circuit / load calibration to remove the parasitic effects of test fixtures and cables, ensuring that the obtained parameters reflect the intrinsic characteristics of the circuit.

[0032] Step 3: Based on the extracted parasitic parameters, construct a high-frequency equivalent model of the internal auxiliary circuit in electromagnetic simulation software or circuit simulation software. The modeling principle is: The core active devices (such as optocouplers and MOSFETs) use the SPICE model provided by the manufacturer, and externally connect the extracted pin parasitic capacitance and inductance. Passive devices (such as transformers) are modeled using a multi-winding equivalent model that includes distributed capacitance and leakage inductance. PCB traces and solder joints are connected to the model in a distributed π-type or T-type equivalent circuit according to their physical size and location.

[0033] After the model is built, frequency domain simulation is performed to obtain its impedance characteristic curve. The simulated curve is compared with the measured impedance curve, and the parasitic parameter values ​​are adjusted to make them match within the target frequency band, thus completing the model verification and calibration.

[0034] In one possible implementation, determining the original test deviation function ΔI(f) includes: Calculate the basic deviation value ΔI0(f) = I1(f) - I2(f); Based on the high-frequency equivalent model, the transmission characteristics of the interference signal transmitted through the internal auxiliary circuit to the residual current sensor are simulated and calculated to obtain the transfer function H(f). The basic deviation value ΔI0(f) is corrected according to the transfer function H(f) to obtain the original test deviation function ΔI(f) = ΔI0(f) × η(f), where η(f) is the frequency correlation correction coefficient determined based on H(f).

[0035] In one specific implementation, the determination of the original test deviation function ΔI(f) includes the following steps: First, under the first operating condition, a frequency sweep interference signal with a preset spectrum is applied to the charging pile under test, and the first residual current action value corresponding to different frequency points f is recorded. Then, the switch-type isolation circuit is deactivated by software shielding or physical disconnection, and the same frequency sweep interference signal is applied under the same second operating condition, and the second residual current action value is recorded. The basic deviation value at each frequency point is calculated, which reflects the direct impact of the presence or absence of the internal auxiliary circuit on the test results.

[0036] Secondly, based on the high-frequency equivalent model constructed in step 1, a frequency-sweeping interference signal consistent with the actual measurement conditions is applied to the model in the simulation software. The simulation yields the response current transmitted from the interference to the residual current sensor via the internal auxiliary circuit, and the transfer function is calculated accordingly. The transfer function describes the transmission characteristics of the internal auxiliary circuit under interference at different frequencies.

[0037] Then, the basic deviation value is corrected according to the transfer function. Since there may be complex factors that cannot be fully simulated under actual test conditions, an empirical correction coefficient η(f) needs to be introduced to make the corrected deviation best match the measured data. The empirical correction coefficient η(f) is obtained by least squares fitting to minimize the sum of squared errors of ΔI(f) = ΔI0(f) × η(f) at each test frequency point.

[0038] The final obtained ΔI(f) is the original test deviation function that continuously varies with the interference frequency f. This function takes the interference frequency f as the independent variable and the deviation of the residual current action value as the dependent variable. Its domain covers the typical interference frequency band that the charging pile may encounter (e.g., 1kHz~1GHz), and its unit is mA. At non-test frequencies, the corresponding deviation value can be obtained through interpolation or function fitting (such as polynomial fitting or power function fitting), which is used for subsequent fine-tuning of the original residual current action value under interference of different frequency components.

[0039] The above scheme constructs a high-frequency equivalent model of the internal auxiliary circuit and extracts its parasitic capacitance, inductance, and resistance parameters that vary with frequency, accurately characterizing the high-frequency parasitic characteristics of the circuit in complex electromagnetic environments. Based on this, by applying interference signals of the same spectrum under both the active and deactivation of the internal auxiliary circuit, the first and second residual current action values ​​are obtained, and the fundamental deviation value is calculated, effectively eliminating the direct impact of the presence or absence of the internal auxiliary circuit on the test results. The transfer function is obtained based on the high-frequency equivalent model simulation, describing the amplitude-frequency and phase-frequency characteristics of the interference signal transmitted from the internal auxiliary circuit to the residual current sensor. An empirical correction coefficient is obtained by fitting using the least squares method, ultimately constructing a function of the original test deviation that continuously varies with the interference frequency. This function covers the typical interference frequency bands of charging piles and can finely correct deviations introduced by interference of different frequency components, rather than simply providing overall compensation. This significantly improves the accuracy and reliability of subsequent environmental coupling correction and residual current action value calibration, providing a data foundation closer to the true intrinsic characteristics of charging pile performance testing.

[0040] In one possible implementation, the environmental characteristic parameters include average humidity and average temperature over a preset time period; the step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation includes: Based on a pre-built device aging database, the multiphysics attenuation coefficients of key semiconductor devices in the internal auxiliary circuit are determined according to the average humidity and average temperature. ; Based on the multiphysics attenuation coefficient, an environmental coupling correction function Fcorr=F(ΔI, α, RH, T) is constructed, which includes a humidity-temperature cross-coupling term; ΔI is the original test deviation. The original test deviation is input into the environmental coupling correction function, and the final test deviation is obtained through nonlinear calculation.

[0041] The preset time period can be one year, six months, three months, etc.

[0042] In one specific implementation, the method for constructing the environmental coupling correction function F(ΔI,α,RH,T) and determining the multiphysics attenuation coefficient α is as follows: The multiphysics attenuation coefficient α is used to characterize the impact of long-term temperature and humidity on the parasitic parameters of key semiconductor devices in the internal auxiliary circuit. This coefficient is obtained by performing high-temperature and high-humidity reverse-bias accelerated aging tests on devices of the same model and fitting the results with a power-law relationship model.

[0043] For example, multiple stress conditions, such as 40℃ / 90%RH, 60℃ / 90%RH, and 80℃ / 90%RH, were selected to conduct a 1000-hour accelerated aging test on the device, and the changes in parameters such as parasitic capacitance and parasitic inductance were measured periodically. Based on the experimental data, the expression for α was obtained by fitting the data using the least squares method:

[0044] Wherein, RH is the average relative humidity of the preset time period area, representing the average humidity level of the charging pile's service environment; T is the average temperature of the preset time period area, representing the average temperature level of the charging pile's service environment. This is the threshold for humidity effects (e.g., 60%RH; below this value, the effect of humidity is negligible). The threshold for temperature influence; and The critical value is determined based on accelerated aging tests; when RH≤RH0 and T≤T0, α=1, indicating that there is no attenuation when the environmental stress does not exceed the device's tolerance limit; k is the device sensitivity coefficient, reflecting the overall sensitivity of the device to temperature and humidity stress; β is the power exponent of the humidity effect, characterizing the intensity of the nonlinear effect of humidity; γ is the exponential coefficient of the temperature effect, characterizing the exponential acceleration effect of temperature stress.

[0045] The environmental coupling correction function F(ΔI,α,RH,T) is used to nonlinearly correct the original test deviation ΔI. Its core lies in introducing a humidity-temperature cross-coupling term to characterize the amplification effect of the synergistic effect of temperature and humidity on the test deviation. The physical mechanism is that high-temperature environments accelerate moisture penetration, leading to changes in the dielectric constant of the PCB substrate and an increase in leakage current on the device surface. Conversely, high-humidity environments exacerbate the electrochemical migration process at high temperatures. These two factors mutually promote each other, jointly affecting the drift rate of parasitic parameters.

[0046] As an exemplary implementation, the environment coupling correction function can be expressed as: F(ΔI,α,RH,T)=ΔI×α×[1+λ·( )·( )] This expression is only one possible implementation. In practical applications, more complex function forms can be obtained by using multivariate nonlinear regression or neural network fitting based on experimental data in the device aging database, as long as it contains humidity-temperature cross-coupling terms. Through this environmental coupling correction function, the original test deviation ΔI is processed by nonlinear operation to obtain the final test deviation, thereby more accurately reflecting the impact of environmental factors on the performance test of charging piles under real working conditions.

[0047] ΔI represents the original test deviation (i.e., the value of the frequency-varying deviation function ΔI(f) determined in the previous step at a specific frequency point); λ is the humidity-temperature cross-coupling coefficient, obtained by fitting the aging test data through multivariate nonlinear regression; the part within square brackets is the humidity-temperature cross-coupling term, when... and When both are non-zero, this term produces a nonlinear increment greater than the sum of its parts alone.

[0048] It should be noted that the above expressions are merely examples and are not intended to limit the scope of protection of this invention. In practical applications, more complex function forms can be obtained by using experimental data from the device aging database, such as multivariate nonlinear regression, support vector regression, or deep neural networks, as long as the function can describe the nonlinear coupling effect of temperature and humidity on the original test deviation.

[0049] The original test deviation ΔI determined in the previous step (which can be a function that varies with frequency or a value at a specific frequency point) is input into the above environmental coupling correction function. After nonlinear calculation, the final test deviation ΔI is obtained. final =F(ΔI,α,RH,T).

[0050] The final test deviation has taken into account the parasitic characteristics of the internal auxiliary circuit, the interference frequency, and the long-term effects and short-term coupling effects of environmental stress such as temperature and humidity. It can be used for subsequent accurate correction of the original residual current operating value.

[0051] The effects of the above technical solution are as follows: Based on a pre-built device aging database and high-temperature and high-humidity reverse-bias accelerated aging test, this solution fits and obtains a multi-physics field attenuation coefficient α, which can scientifically reflect the degree of attenuation of parasitic parameters of key semiconductor devices in the internal auxiliary circuit under long-term temperature and humidity. This allows the correction of test deviations to not only consider instantaneous environmental conditions, but also take into account the performance drift of the device throughout its entire life cycle.

[0052] By introducing a humidity-temperature cross-coupling term into the environmental coupling correction function, the limitation of traditional linear correction models in failing to characterize the mutual promotion and amplification of test deviations by temperature and humidity is overcome. When the charging pile is in an environment with both high temperature and high humidity for a long time, this cross-coupling term can generate a nonlinear correction amount greater than the sum of the individual effects of the two, making the final test deviation closer to the actual working conditions.

[0053] This improves the comparability of test data under different geographical and seasonal conditions. After the modification of this solution, the test results obtained by the same charging pile in the dry and cold north and the hot and humid south, or the test results obtained under the conditions of high temperature and high humidity in summer and low temperature and low humidity in winter, can all be normalized based on a unified environmental benchmark. This eliminates the data fluctuations introduced by environmental differences and provides reliable data support for the horizontal comparison and vertical tracking of charging pile performance.

[0054] In a preferred embodiment of the present invention, the modified set of performance parameters includes, but is not limited to: the modified residual current operating value, the modified residual current operating time, the modified protection operating threshold, and the insulation resistance reference value re-evaluated based on the modified residual current operating value.

[0055] The specific calibration process is as follows: After obtaining the final test deviation ΔIfinal, the original residual current action value I is... raw Correction to be made: I corr= I raw -ΔIfinal; Where Icorr is the corrected residual current operating value, in milliamperes (mA).

[0056] The residual current operating characteristic curve function built into the charging pile is obtained; this function reflects the standard operating time corresponding to different residual current values. The corrected residual current operating value is substituted into this function to obtain the first operating time; if there is a deviation between the original operating time recorded by the charging pile and the first operating time, the first operating time is used as the corrected operating time. The residual current operating characteristic curve, known as the inverse-time characteristic curve in electrical engineering, is characterized by the fact that the larger the residual current, the shorter the allowable operating time. The residual current operating characteristic curve function is a function or data table reflecting the inverse-time relationship between residual current and standard operating time, based on the operating time limits specified in national standards and type test data, through mathematical fitting or directly preset in the charging pile control system as a parameter table.

[0057] Obtain the preset protection action threshold of the charging pile (e.g., 30mA). Calculate the current deviation ratio before and after correction; multiply the preset protection action threshold by this deviation ratio to obtain the calibrated protection action threshold.

[0058] Obtain the applied voltage U (in volts V) during the insulation resistance test of the charging pile. Recalculate the insulation resistance based on the corrected residual current operating value.

[0059] The final set of corrected performance parameters contains accurate parameters for multiple dimensions (current value, action time, protection threshold, insulation resistance), enabling the performance test report to fully reflect the real status of the charging pile, providing a more reliable basis for operation and maintenance decisions, effectively extending the service life of the charging pile and ensuring power safety.

[0060] In one possible implementation, the method further includes: Obtain the corrected residual current action values ​​of multiple charging piles within the area where the charging pile under test is located; Based on the physical location of each charging pile in the area, the power grid topology connection relationship and historical load data, calculate the electrical coupling weight matrix between charging piles in the area; Based on the electrical coupling weight matrix, the corrected residual current action values ​​of each charging pile in the region are weighted and fused to generate dynamic statistical reference values ​​for the region. The corrected residual current action value of the charging pile under test is compared with the dynamic statistical reference value to generate a trend performance test report that includes horizontal comparative analysis.

[0061] The calculation of the electrical coupling weight matrix among the charging piles within the region, based on the physical location of each charging pile, the power grid topology, and historical load data, includes: Obtain the physical coordinates of each charging pile in the area, the length of the cable connected to the same busbar, and the impedance parameters; Obtain the load current data of each charging pile in the area over a continuous time series, and calculate the load fluctuation similarity between any two charging piles. ; Based on the cable length, impedance parameters, and load fluctuation similarity, an electrical coupling weight matrix element is constructed. ;

[0062] This is the equivalent electrical distance between the nth and jth charging piles, calculated based on cable length and impedance.

[0063] In one possible implementation, Let Pearson correlation coefficients be the load current time series of the nth and jth charging piles within a preset time window (e.g., 72 consecutive hours); the value range is [-1, 1], representing the synchronicity of load changes between the two; the results can be normalized, letting... '=( +1) / 2, with a value range of [0,1], eliminates the influence of negative correlation on weight calculation. The final weight matrix uses the normalized value. Participate in the calculation.

[0064] The equivalent electrical distance is based on the cable length of the nth and jth charging piles connected to the same busbar. and impedance per unit length (Including resistance and reactance) Calculated, the impedance magnitude is taken: ; If the two piles are connected to the busbar via different paths, the equivalent impedance value needs to be calculated based on the actual topology.

[0065] In one possible implementation, the dynamic statistical reference value is obtained by the following formula; including:

[0066] t represents the dynamic statistical reference value at time t; m represents the number of charging piles in the area. This is the corrected current action value for the nth charging pile; The dynamic simultaneity coefficient function is calculated based on the real-time load current of the nth and jth charging piles at time t.

[0067] Wherein, the dynamic simultaneous coefficient function The calculation is based on the real-time load current of the nth and jth charging piles at time t and their statistical characteristics within a preset time window.

[0068] It should be noted that, The intermediate variable is calculated from the load similarity and electrical distance ratio, and its dimension is conductance (1 / Ω). Since both the numerator and denominator contain [variables] during the weighted fusion process... When summed, the dimensions are automatically canceled out, resulting in a dynamic statistical reference value. Dimensions and Consistent (mA); therefore, When used as a weight in normalization calculations, its absolute value determines the contribution ratio of each stake and does not affect the dimensional accuracy of the final result.

[0069] In one possible implementation, the dynamic simultaneous coefficient function The following data is obtained by using the real-time load current of the nth and jth charging piles at time t, and the corresponding rated current of the nth and jth charging piles:

[0070] in, Let be the real-time load current of the nth charging pile at time t; Let be the real-time load current of the j-th charging pile at time t; Let n be the rated current of the nth charging pile; Let be the rated current of the j-th charging pile.

[0071] In another possible implementation, the dynamic simultaneous coefficient function The Pearson correlation coefficient is obtained by calculating the load current sequence of the nth and jth charging piles within a preset time window. The preset time window is one hour before time t, i.e., the load current sequence from t-1h to time t is used for calculation. If the Pearson correlation coefficient is used as the dynamic simultaneity coefficient function, then the absolute value or non-negative part of the coefficient is taken, i.e.:

[0072] or;

[0073] This is the load current correlation coefficient calculated based on a real-time sliding window. Using the absolute value preserves the correlation strength information, while using non-negative truncation only considers the positive correlation effect; the choice can be made according to the actual scenario.

[0074] The dynamic simultaneity coefficient function product ratio is suitable for scenarios where charging piles of the same model and rated current are consistent within a region. It is simple to calculate, has strong real-time performance, and can quickly reflect the load synchronization status of the two at time t. The value range is [0,1]. The closer the value is to 1, the closer the load of the two at time t is to the rated value and the higher the synchronization. The Pearson correlation coefficient is suitable for scenarios where charging piles of different models and rated currents vary greatly within a region. It can eliminate the influence of rated current differences and accurately characterize the trend consistency of load fluctuations of the two. The value range is [-1,1], and it is used after normalization to [0,1].

[0075] The principle of the above technical solution is as follows: First, the residual current action values ​​of multiple charging piles within the area where the charging pile under test is located are obtained after correction using the aforementioned method, serving as the intrinsic performance parameters of each pile. Second, an electrical coupling weight matrix is ​​constructed based on the physical location of the charging piles within the area, the grid topology connections, and historical load data. The weight between any two piles n and j is determined by the ratio of their load fluctuation similarity to their equivalent electrical distance: the more similar the load fluctuations and the closer the electrical distance between the pile pairs, the greater their weight, indicating a tighter electrical coupling between them.

[0076] Then, by combining the real-time load current, a dynamic simultaneity coefficient function is calculated to characterize the correlation between the load states of the two piles at the current moment. This coefficient can be obtained by the product ratio of the real-time load and the rated current, or by the load correlation coefficient within a sliding window, and its value is dynamically adjusted as the load changes.

[0077] Next, the electrical coupling weight and dynamic simultaneity coefficient are integrated, and the corrected residual current action values ​​of all charging piles in the area are weighted and averaged to obtain a dynamic statistical reference value. This reference value comprehensively reflects the statistical benchmark of the overall charging pile performance in the area under the current load coupling state.

[0078] Finally, the corrected residual current action value of the charging pile under test is compared with the dynamic statistical reference value to analyze its deviation degree and trend, and a trend performance test report including horizontal comparative analysis is generated to provide decision-making basis for operation and maintenance personnel.

[0079] The effects of the above technical solution are as follows: Abandoning the traditional model of testing individual charging piles independently, this study considers the electrical coupling of the power grid topology and the correlation of load fluctuations between charging piles in the region, making the performance evaluation more in line with the actual working conditions of large-scale network operation of charging piles, and avoiding the one-sidedness of single-pile test results.

[0080] Dynamic statistical reference values ​​are more accurate and closely reflect real-time conditions: Physical connection associations are quantified through electrical coupling weight matrix, and the real-time load synchronization at time t is reflected by dynamic simultaneous coefficient function. This makes the regional reference values ​​both statistically representative and real-time dynamic. Compared with fixed reference values, they can more accurately reflect the overall performance level of the region.

[0081] By comparing the performance of the pile under test with the regional dynamic reference value, the performance ranking, deviation degree and trend change of a single pile in the region can be clarified. This allows the test report to not only reflect the true performance of a single pile, but also provide data support for regional operation and maintenance of charging piles, fault prediction and resource optimization (such as early investigation of piles with excessive deviation and development of maintenance plans for piles with deteriorating performance trends).

[0082] In one possible implementation, the step of comparing the corrected residual current operating value of the charging pile under test with the dynamic statistical reference value to generate a trend-based performance test report including cross-sectional comparative analysis includes: Calculate the absolute and relative deviations between the corrected residual current action value of the charging pile under test and the dynamic statistical reference value; Based on the absolute and relative deviations, and combined with a preset deviation threshold range, the performance deviation status of the charging pile under test is determined. The performance deviation status includes levels such as normal, attention, and abnormal; for example: Normal threshold: relative deviation ≤ 5% and absolute deviation , This refers to the instrument's basic error limit; Attention threshold: 5% < relative deviation ≤ 15% or absolute deviation exceeds the basic error but is less than the alarm threshold; Abnormal threshold: relative deviation > 15% or absolute deviation exceeds the alarm threshold; The judgment logic not only considers a single measurement, but can also be combined with sliding window filtering. For example, if the threshold is exceeded three times in a row, a state change is triggered to avoid misjudgment caused by transient interference. The performance deviation status and corrected residual current action value of the charging pile under test are obtained in multiple consecutive test cycles to form time series data; the MAC address or physical ID of the charging pile is used as an index to record two core parameters generated in each test cycle (e.g., daily or weekly) in the time series database: corrected residual current action value and performance deviation status (which can be quantified as numerical labels, such as normal=0, attention=1, abnormal=2) to form a time series. Based on the time series data, future trends are predicted using linear regression or exponential smoothing, and patterns of continuous deterioration or periodic fluctuations are identified. Linear regression models (such as least squares fitting of the slope of a straight line) are used to determine whether the performance is stable (slope ≈ 0), deteriorating (slope > 0), or improving (slope < 0). Alternatively, exponential smoothing methods (such as the Holt-Winters model) are used to predict the residual current action value for the next cycle. Through frequency domain analysis or periodic checks of historical data, the presence of seasonal fluctuations (such as generally higher values ​​during periods of high temperature and humidity in summer) or occasional spikes is identified.

[0083] The performance deviation status, trend prediction results, and horizontal ranking with other charging piles in the same area of ​​the charging pile under test are integrated to generate the trend performance test report.

[0084] The report includes: Deviation status: The current judgment level (normal / concern / abnormal).

[0085] Horizontal ranking: Within the same area, the residual current action value of the pile is sorted with other piles, and a percentile ranking is given.

[0086] Trend Forecast: Displays historical data curves and future forecast curves in chart form, and marks conclusions such as trend deterioration or trend stabilization.

[0087] Operation and maintenance recommendations: If the predicted value is about to exceed the threshold or the status remains under concern, an early warning will be automatically generated, and on-site inspection or in-depth maintenance is recommended.

[0088] The above technical solution achieves the following effect: by introducing dynamic statistical reference values ​​based on regional electrical coupling, the data of a single charging pile is placed within the entire charging pile group for horizontal comparison. Compared with the traditional method of comparing only with a fixed threshold, this method can effectively eliminate common deviations caused by regional power grid background fluctuations and environmental consistency changes (such as a region being affected by temperature and humidity simultaneously), thereby more accurately identifying whether a single individual has a true performance degradation.

[0089] By performing trend prediction (such as linear regression) on data from continuous testing cycles, the system can identify a continuously deteriorating trend before the residual current operating value exceeds the national standard's acceptable upper limit. This predictive maintenance capability allows the operation and maintenance team to intervene in advance, avoiding sudden failures and safety hazards, and significantly improving the availability and safety of charging pile clusters.

[0090] The performance deviation status classification (normal, watch out, abnormal) and horizontal ranking included in the trend test report provide operations and maintenance personnel with a clear priority list. Resources can be prioritized for abnormal devices or those at the bottom of the ranking and continuing to deteriorate, avoiding blind inspection of all devices and thus significantly reducing the labor and time costs of operations and maintenance.

[0091] Identifying periodic fluctuation patterns helps uncover design flaws in certain devices that only surface under specific environmental conditions (such as the rainy season). By feeding this trend data back to the R&D department, improvements can be made at the source to enhance the anti-interference design or multi-physical field protection capabilities of the charging pile's internal auxiliary circuitry, creating a data-driven product iteration loop.

[0092] In one possible implementation, the step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation further includes: Obtain the historical test data and corresponding historical environmental parameters of the charging pile under test; Using the historical environmental parameters as input and the deviation correction amount extracted from the historical test data as output, a deep neural network model is trained. The current environmental feature parameters are input into the trained deep neural network model to predict the correction coefficient under the current environment, and the final test deviation is dynamically generated by combining the original test deviation.

[0093] This implementation uses a deep neural network model to fit the complex nonlinear mapping relationship between environmental factors and test deviation, thereby achieving dynamic and adaptive correction of the original test deviation. Specifically, it includes the following steps: Obtain raw test data and corresponding environmental characteristic parameters of the charging pile under test in multiple historical test cycles. The historical test data includes at least the original residual current action value and the baseline value measured after physically disconnecting or software-shielding the internal auxiliary circuit; the difference between the two is the historical deviation correction amount (label value). The environmental characteristic parameters include at least the temperature, humidity, electromagnetic interference intensity, and power supply voltage fluctuation at the time of test. Arrange all historical data in chronological order to form the raw dataset.

[0094] The original dataset is cleaned to remove outliers caused by equipment malfunctions or recording errors; missing values ​​are filled using interpolation or forward imputation; environmental parameters and bias corrections are normalized to have a mean of 0 and a variance of 1 to accelerate neural network training convergence.

[0095] Design a multilayer feedforward neural network (DNN) for fitting the environment coupling correction function. The network structure includes: Input layer: The number of neurons is equal to the number of selected environmental feature parameters (such as temperature, humidity, electromagnetic interference intensity, voltage fluctuation, etc.), and each neuron corresponds to a normalized environmental parameter.

[0096] Hidden layers: Several fully connected hidden layers are set, such as 3 to 5 layers, with the number of neurons in each layer decreasing progressively (e.g., 64, 32, 16). The activation function is ReLU (Rectified Linear Unit) to introduce non-linearity, and Dropout layers (dropout rate 0.2 to 0.5) are added after some layers to prevent overfitting.

[0097] Output layer: a single neuron, with a correction coefficient k for the output prediction. pred This coefficient is a scalar representing the deviation correction factor under current environmental conditions relative to a standard environment (such as a laboratory calibration environment). The network uses mean squared error (MSE) as the loss function.

[0098] The preprocessed historical dataset was divided into a training set (70%), a validation set (15%), and a test set (15%). The neural network was trained using the training set, with Adam selected as the optimizer and an initial learning rate of 0.001. Early stopping was employed to monitor the validation set loss; training was stopped when the validation loss ceased to decrease for several consecutive epochs, and the optimal model was saved. The model's generalization ability was evaluated using the test set to ensure the correlation coefficient R between the predicted correction coefficient and the historical actual deviation correction. 2 It reaches 0.9 or higher.

[0099] The currently collected environmental feature parameters (average humidity, average temperature, etc.) are input into the trained deep neural network model. After forward propagation, the model outputs the prediction correction coefficient k. pred Then, the correction coefficient is combined with the original test deviation (multiplied) to obtain the final test deviation. If the model output correction coefficient is greater than 1, it indicates that the current environment has aggravated the deviation; if it is less than 1, it indicates that the environment has suppressed the deviation.

[0100] Environmental factors (such as temperature, humidity, and electromagnetic interference) often have highly nonlinear and coupled effects on the parasitic parameters of the auxiliary circuits inside charging piles. Deep neural networks, through multi-layer nonlinear transformations, can automatically learn and approximate this complex coupling relationship, capturing subtle interactions that are difficult to describe by traditional cross-term models, thereby significantly improving correction accuracy. As charging piles operate for longer periods, historical data accumulates. The model can automatically adapt to equipment aging, component performance drift, and environmental changes (such as seasonal changes and power grid upgrades) through incremental learning or periodic retraining, maintaining the timeliness and accuracy of the correction function and avoiding the problem of fixed-parameter models failing due to environmental changes.

[0101] In one possible implementation, the historical environmental parameters include multi-physics environmental stress spectrum data continuously collected within a preset historical period; the step of training a deep neural network model using the historical environmental parameters as input and the deviation correction amount extracted from the historical test data as output includes: The temperature time series curve, humidity time series curve and electromagnetic interference spectrum energy distribution curve are obtained within the first preset time period before the test, forming a multidimensional environmental stress spectrum matrix. Construct a deep neural network containing convolutional layers or recurrent neural network layers, and use the convolutional layers to extract the temporal variation features and frequency domain distribution features of the multidimensional environmental stress spectrum matrix; The extracted temporal variation features and frequency domain distribution features are input into the fully connected layer of the deep neural network, and the model is trained using the deviation correction amount as a label to obtain an environmental coupling correction model that can capture the dielectric hygroscopic hysteresis effect and thermal inertia effect.

[0102] By constructing time-series curves of temperature, humidity, and electromagnetic interference spectra over a continuous period before testing, a multi-dimensional environmental stress spectrum matrix is ​​formed, enabling the correction model to perceive the dynamic changes in the environment. For example, the moisture absorption of charging pile insulation materials differs significantly between continuous high humidity and sudden high humidity environments. This solution can accurately distinguish these scenarios through time-series feature extraction, greatly improving correction accuracy.

[0103] The performance changes of the internal insulation materials and electronic components of charging piles are not synchronized with environmental changes, exhibiting significant hysteresis (hygroscopic hysteresis) and thermal accumulation effects (thermal inertia). By introducing convolutional layers or recurrent neural network layers, the model can automatically learn deep-level characteristics such as the rate, acceleration, and duration of environmental stress changes, thereby accurately quantifying the complex impact of these hysteresis effects on the residual current action value and achieving predictive correction ahead of environmental changes. The impact of electromagnetic interference on charging piles depends not only on the interference intensity but also on the spectral distribution of the interference and its time-frequency overlap with the test signal. This solution incorporates the electromagnetic interference spectral energy distribution curve into a multidimensional stress spectrum matrix, enabling the model to learn the coupling law between interference in different frequency bands and the resonant peak values ​​of parasitic parameters in internal auxiliary circuits, significantly improving the accuracy of residual current testing under complex electromagnetic environments. The multidimensional environmental stress spectrum matrix contains rich information on environmental evolution. Through automatic feature extraction by the deep learning model, it is possible to discover hidden fault modes that cannot be manifested under static parameters. For example, some charging piles may only experience a surge in deviation during specific weather conditions with a sudden drop in temperature and high humidity. This complex mode can only be effectively identified and corrected through time-series stress spectrum analysis.

[0104] In one possible implementation, the training of the deep neural network model further includes: Obtain a pre-constructed high-frequency equivalent physical model of the internal auxiliary circuit, wherein the high-frequency equivalent physical model includes the mapping relationship between parasitic parameters and interference frequencies; A physical consistency constraint term is introduced into the loss function for training the deep neural network model. The physical consistency constraint term is used to measure whether the correction coefficients of the neural network output satisfy the preset circuit physical laws when mapped to the high-frequency equivalent physical model. The loss function includes the error between the predicted deviation correction and the historical actual deviation correction, as well as a physical consistency constraint term; by minimizing the loss function, a physical information neural network model that conforms to the physical laws of the circuit is trained.

[0105] While purely data-driven neural network models offer strong fitting capabilities, they may produce corrections that are physically impossible to achieve, posing a potential hazard in the demanding safety testing scenarios of charging stations. This solution introduces a physical consistency constraint term into the loss function, injecting fundamental physical laws of circuitry (such as impedance continuity, phase constraints, and energy conservation) as prior knowledge into the model. This ensures that the neural network, while learning data features, must adhere to objective physical laws, achieving a deep integration of physical mechanisms and data-driven approaches. The neural network trained with physical consistency constraints naturally outputs results consistent with circuit principles. Engineers can back-map the predictions to a high-frequency equivalent physical model to analyze the rationality of parasitic parameter changes corresponding to the current correction coefficients. This provides traceable and interpretable physical evidence for the test results, significantly increasing testers' trust in the AI's correction results.

[0106] In one possible implementation, the method further includes: The generated final test deviation is compared with a preset deviation alarm threshold; If the final test deviation exceeds the deviation alarm threshold, an early warning message containing the charging pile identifier, the deviation item, and the recommended maintenance time is generated and pushed to the operation and maintenance terminal.

[0107] Based on the design standards, safety regulations, and historical operating data of the charging piles, preset deviation alarm thresholds are established. Threshold settings can be implemented using one or a combination of the following two methods: Fixed threshold method: Based on national standards or enterprise technical specifications, set the maximum permissible percentage deviation of the residual current operating value (e.g., ±15%) as the absolute threshold. When the final test deviation exceeds this percentage, an alarm is triggered.

[0108] Dynamic threshold method: Based on the historical final test deviations of all charging piles of the same model in the area where the charging pile is located, the statistical process control (SPC) method is used to calculate the mean. μ and standard deviation σ are set as the upper and lower limits of the threshold, which are μ ± 3σ. When the deviation of a single pile exceeds this control limit, it is considered a statistical anomaly and an alarm is triggered. The dynamic threshold can adapt to long-term trends such as equipment aging and environmental changes, reducing false alarms.

[0109] After each charging pile performance test is completed and the final test deviation is obtained, it is compared with the preset deviation alarm threshold in real time. The judgment logic includes: Instantaneous deviation: If the current final test deviation is greater than the upper limit of the threshold or less than the lower limit of the threshold, an early warning will be triggered immediately.

[0110] Cumulative deviation: To avoid false alarms caused by noise in a single measurement, a sliding window mechanism can be introduced. If the threshold is exceeded for N consecutive tests (e.g., 3 consecutive tests), an early warning will be triggered.

[0111] Once an alert is triggered, the system automatically generates alert information containing the following key fields: Charging pile identification: includes unique identification information such as charging pile ID, geographical location, and the operating site to which it belongs.

[0112] Out-of-tolerance items: clearly indicate the parameters that are out of tolerance, such as the deviation of the residual current action value and the current deviation. Deviation trend: combine the time series data of the historical deviation of the pile to give a short-term trend description (such as the deviation continuing to increase).

[0113] Using trend prediction models (such as linear regression or exponential smoothing), predict the remaining time before the deviation reaches the critical alarm threshold (such as the national standard upper limit), and provide graded recommendations such as maintenance within X days or immediate shutdown for repair.

[0114] Through multi-channel message push service, early warning information is sent to the terminal devices of maintenance personnel in real time, including: Mobile app push notifications: Push notifications via a dedicated operations and maintenance app or WeChat mini-program in the form of notification bar messages.

[0115] SMS / email alerts: For high-risk alerts (such as those recommending immediate shutdown), both SMS and email alerts will be sent to on-duty personnel.

[0116] Work order system integration: Automatically create a "high" priority work order in the operation and maintenance work order system, assign it to the responsible person in the corresponding area, and track and process the closed loop.

[0117] All triggered warning messages are stored in a historical database, forming warning records. Regular statistical analysis of these warning records identifies charging pile models or sites that frequently exhibit similar out-of-range performance, providing data support for product improvement and operational strategy optimization.

[0118] The effects of the above technical solution are as follows: This transforms traditional post-fault maintenance into pre-fault early warning. By monitoring the final test deviation and issuing early warnings when it exceeds a threshold, maintenance personnel can intervene before actual faults occur at the charging station (such as malfunctioning or refusing to operate the residual current protection), preventing charging service interruptions and improving charging station availability and user satisfaction. The warning information includes precise fault location (exceeding tolerance) and priority (suggested maintenance time), enabling maintenance personnel to quickly identify and address problems, reducing troubleshooting time. Simultaneously, alarms based on dynamic thresholds avoid ineffective inspections of normal fluctuations caused by environmental factors, thus saving manpower and material resources and achieving refined maintenance. Exceeding the residual current operating value deviation often indicates deterioration in insulation performance or abnormal auxiliary circuits; failure to address this promptly may lead to electric shock or fire risks. The early warning mechanism ensures that such safety hazards are detected and addressed in a timely manner, significantly improving the overall safety of charging facilities.

[0119] Example 2: This example provides a machine learning-based charging anomaly detection system for new energy equipment, used in Example 1 for the machine learning-based charging anomaly detection method for new energy equipment. The system is characterized by including: The data acquisition module is used to acquire the raw test data of the charging pile under test directly collected by the residual current sensor and the environmental characteristic parameters of the charging pile under test. The interference identification and modeling module is used to identify internal auxiliary circuits and build their high-frequency equivalent models to determine the original test deviation. The environmental coupling correction module is used to correct the original test deviation based on environmental characteristic parameters using a nonlinear correction model to obtain the final test deviation. The parameter correction module is used to correct the original residual current action value based on the final test deviation, and to calibrate the associated performance parameters based on the corrected residual current action value. The regional collaborative analysis module is used to generate dynamic statistical reference values ​​based on the electrical coupling relationship of charging piles within the region; The report generation and early warning module is used to generate performance test reports and issue early warnings when performance parameters exceed acceptable limits.

[0120] The working principle and beneficial effects of the above technical solution are the same as those of Embodiment 1, and will not be repeated here.

[0121] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any embodiment of this invention.

[0122] This invention also provides a computer-readable storage medium for storing a computer program. When the computer program is executed, it implements the steps of the method described in Embodiment 1 of this invention. The specific implementation method is consistent with the implementation method and the technical effects achieved in the above method embodiments, and some contents will not be repeated.

[0123] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the invention without departing from the principles and spirit of the invention, and all such changes should fall within the protection scope of the claims of the present invention.

Claims

1. A method for detecting charging anomalies in new energy equipment based on machine learning, characterized in that, include: Acquire the raw test data of the charging pile under test directly collected by the residual current sensor, wherein the raw test data includes at least the raw residual current action value; Identify the internal auxiliary circuit located after the residual current sensor inside the charging pile under test, which forms an unexpected leakage path between the charging circuit and ground; based on the influence of the internal auxiliary circuit on the test process under complex electromagnetic environment, and combined with its parasitic parameters, determine the original test deviation introduced by the internal auxiliary circuit. Obtain the environmental characteristic parameters of the charging pile under test; Based on the environmental characteristic parameters, an environmental coupling correction function is constructed to perform nonlinear correction on the original test deviation to obtain the final test deviation. Based on the final test deviation, the original residual current action value is corrected to obtain the corrected residual current action value. Based on the corrected residual current action value, the associated performance parameters of the charging pile under test are calibrated to obtain the corrected performance parameter set. Based on the revised set of performance parameters, a performance test report for the charging pile under test is generated.

2. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 1, characterized in that, The internal auxiliary circuit includes at least a switch-type isolation circuit connected between the charging circuit and ground; the determination of the original test deviation introduced by the internal auxiliary circuit based on its influence on the test process under complex electromagnetic environment and its parasitic parameters includes: A high-frequency equivalent model of the internal auxiliary circuit is constructed, and its parasitic capacitance and parasitic inductance parameters are extracted. Under the first operating condition, an interference signal with a preset spectrum is applied to obtain the first residual current action value; Under the second condition that the internal auxiliary circuit is deactivated by physical disconnection or software shielding, an interference signal of the preset spectrum is applied to obtain a second residual current action value. Based on the parasitic parameters, the difference between the first residual current operating value and the second residual current operating value, the original test deviation function, which is introduced by the internal auxiliary circuit and varies with the interference frequency, is calculated.

3. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 2, wherein constructing the high-frequency equivalent model of the internal auxiliary circuit includes: A wideband impedance test was performed on the internal auxiliary circuit to obtain its impedance characteristic curve; Based on the impedance characteristic curve, the parasitic capacitance, parasitic inductance, and parasitic resistance that vary with frequency are extracted by the equivalent circuit fitting method. The extracted parasitic parameters are substituted into the equivalent circuit topology to construct a lumped parameter equivalent model containing parasitic elements, and the model is verified by comparing it with the measured impedance curve.

4. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 1, characterized in that, The environmental characteristic parameters include the average humidity and average temperature within a preset time period; the step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation includes: Based on a pre-built device aging database, the multi-physics field attenuation coefficient of key semiconductor devices in the internal auxiliary circuit is determined according to the average humidity and average temperature. The multi-physics field attenuation coefficient is obtained by performing high temperature and high humidity reverse bias accelerated aging tests on devices of the same model and fitting the results with a power law relationship model. Based on the multiphysics attenuation coefficient, an environmental coupling correction function is constructed, which includes a humidity-temperature cross-coupling term. The original test deviation is input into the environmental coupling correction function, and the final test deviation is obtained through nonlinear calculation.

5. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 1, characterized in that, The method further includes: Obtain the corrected residual current action values ​​of multiple charging piles within the area where the charging pile under test is located; Based on the physical location of each charging pile in the area, the power grid topology connection relationship and historical load data, calculate the electrical coupling weight matrix between charging piles in the area; Based on the electrical coupling weight matrix, the corrected residual current action values ​​of each charging pile in the region are weighted and fused to generate dynamic statistical reference values ​​for the region. The corrected residual current action value of the charging pile under test is compared with the dynamic statistical reference value to generate a trend performance test report that includes horizontal comparative analysis.

6. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 5, characterized in that, The calculation of the electrical coupling weight matrix among the charging piles within the region, based on the physical location of each charging pile, the power grid topology, and historical load data, includes: Obtain the physical coordinates of each charging pile in the area, the length of the cable connected to the same busbar, and the impedance parameters; Obtain the load current data of each charging pile in the area in a continuous time series, and calculate the load fluctuation similarity between any two charging piles; Based on the cable length, impedance parameters, and load fluctuation similarity, an electrical coupling weight matrix element is constructed. Among them, the more similar the load fluctuations of any two pile pairs are and the closer their electrical distance is, the greater their corresponding weight.

7. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 6, characterized in that, The dynamic statistical reference value is obtained by the following formula; including: t represents the dynamic statistical reference value at time t; m represents the number of charging piles in the area. This is the corrected current action value for the nth charging pile; The dynamic simultaneity coefficient function is calculated based on the real-time load current of the nth and jth charging piles at time t.

8. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 1, characterized in that, The step of constructing an environmental coupling correction function based on the environmental characteristic parameters to perform nonlinear correction on the original test deviation to obtain the final test deviation also includes: Obtain the historical test data and corresponding historical environmental parameters of the charging pile under test; Using the historical environmental parameters as input and the deviation correction amount extracted from the historical test data as output, a deep neural network model is trained. The current environmental feature parameters are input into the trained deep neural network model to predict the correction coefficient under the current environment, and the final test deviation is dynamically generated by combining the original test deviation.

9. The method for detecting charging anomalies in new energy equipment based on machine learning according to claim 1, characterized in that, The method further includes: The generated final test deviation is compared with a preset deviation alarm threshold; If the final test deviation exceeds the deviation alarm threshold, an early warning message containing the charging pile identifier, the deviation item, and the recommended maintenance time is generated and pushed to the operation and maintenance terminal.

10. A machine learning-based new energy equipment charging anomaly detection system, used to execute the machine learning-based new energy equipment charging anomaly detection method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the raw test data of the charging pile under test directly collected by the residual current sensor and the environmental characteristic parameters of the charging pile under test. The interference identification and modeling module is used to identify internal auxiliary circuits and build their high-frequency equivalent models to determine the original test deviation. The environmental coupling correction module is used to correct the original test deviation based on environmental characteristic parameters using a nonlinear correction model to obtain the final test deviation. The parameter correction module is used to correct the original residual current action value based on the final test deviation, and to calibrate the associated performance parameters based on the corrected residual current action value. The regional collaborative analysis module is used to generate dynamic statistical reference values ​​based on the electrical coupling relationship of charging piles within the region; The report generation and early warning module is used to generate performance test reports and issue early warnings when performance parameters exceed acceptable limits.

Images