Electromagnetic compatibility fault prediction method for pulse withstand voltage generator

Abstract

The invention discloses an electromagnetic compatibility fault prediction method for a pulse withstand voltage generator, and relates to the technical field of electromagnetic compatibility evaluation, and the method comprises the steps: collecting original signals of different circuit states of the pulse withstand voltage generator, extracting voltage ripple, current phase and magnetic field radiation frequency spectrum, and generating a time sequence monitoring data set; based on the time sequence monitoring data set, electromagnetic disturbance characteristics are extracted through deep analysis, a rule engine is built, and a fault probability and a fault type are output; triggering a plurality of suppression rules based on different fault types; after any suppression rule is executed, collecting the original signal again, and inputting the original signal into a pre-constructed risk assessment model to generate an interference suppression efficiency index; comparing the interference suppression efficiency index with a preset efficiency threshold value; if the interference suppression efficiency index is lower than the efficiency threshold value, generating a strategy compensation instruction and feeding the strategy compensation instruction back to the corresponding suppression step to iterate and adjust parameters; the electromagnetic interference suppression effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic compatibility assessment, and in particular to an electromagnetic compatibility fault prediction method for a pulse withstand voltage generator. Background Art

[0002] With the rapid development of new energy vehicles and other fields, the application of electronic equipment such as pulse withstand voltage generators is becoming more and more extensive, and electromagnetic compatibility issues are becoming more and more prominent. Pulse withstand voltage generators are mainly used to test the insulation performance and withstand voltage capacity of related components of new energy vehicles to ensure the safety and reliability of vehicles in the operation of high-voltage electrical systems. Their working process has the characteristics of high voltage, high frequency, and strong electromagnetic radiation. With the expansion of power grid scale and intelligent development, power system fault prediction technology continues to develop, such as the use of time series analysis, machine learning, deep learning and other methods to predict power system faults; however, in the following aspects: On the one hand, the equipment generates various electromagnetic interferences during operation and is also susceptible to the external electromagnetic environment. The electromagnetic characteristics of the equipment under different internal circuit states are complex and changeable, making it difficult to conduct in-depth feature mining and comprehensively capture the multi-dimensional electromagnetic disturbance characteristics. This results in low fault prediction accuracy, making it impossible to output reliable fault probabilities and types at an early stage, and thus missing the best opportunity for fault prevention. On the other hand, when it comes to dealing with electromagnetic interference, existing suppression strategies are usually fixed parameters or static designs, relying on a single threshold alarm mechanism. They are unable to distinguish fault types and find it difficult to effectively cope with complex and changeable electromagnetic interference environments, resulting in untimely fault suppression and reduced system stability and reliability. For example, although most vehicles are currently equipped with filtering circuits, when the on-board charger operating mode is switched, the center frequency of the conducted interference will jump, and it will be impossible to match the interference frequency in time, resulting in a significant decrease in filtering efficiency, seriously affecting the normal operation of the vehicle's intelligent system. Summary of the Invention

[0003] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides an electromagnetic compatibility fault prediction method for pulse withstand voltage generators. By collecting the original signals of the pulse withstand voltage generator in different circuit states in real time, and extracting the voltage ripple, current phase and magnetic field radiation spectrum, a time-series monitoring data set is generated, and in-depth analysis is performed to capture several suppression rules. Targeted automatic adjustments are made according to actual conditions to find the optimal solution, continuously improve the electromagnetic interference suppression effect, enhance the stability and reliability of the system, and solve the problems raised in the background technology.

[0004] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: The present application provides a method for predicting electromagnetic compatibility faults of a pulse withstand voltage generator, the method comprising the following steps: S1: Real-time acquisition of raw signals from the pulse withstand voltage generator in different circuit states, extraction of voltage ripple, current phase, and magnetic field radiation spectrum, and generation of a time-series monitoring data set. Different circuit states include charging, resonance, and discharging. S2: Based on the time-series monitoring data set, we extract electromagnetic disturbance characteristics through in-depth analysis and build a rule engine to output fault probability and fault type. Based on different fault types, we trigger several suppression rules. S3: After executing any suppression rule, the original signal of the pulse withstand voltage generator is collected again and input into the pre-built risk assessment model to generate an interference suppression effectiveness index; the interference suppression effectiveness index is compared with the preset effectiveness threshold: if the interference suppression effectiveness index is lower than the effectiveness threshold, a strategy compensation instruction is generated and fed back to step S2 to iteratively adjust the parameters; if the interference suppression effectiveness index exceeds the effectiveness threshold, the parameters corresponding to the current suppression rule are output as the optimal solution.

[0005] Furthermore, through in-depth analysis, electromagnetic disturbance characteristics are extracted and a rule engine is built to output fault probability and fault type, including: Data processing: preprocessing of time series monitoring data sets, including denoising, normalization, and modal decomposition; Feature analysis: Based on the preprocessed results, a support vector machine is used to determine if the system is qualified. Electromagnetic disturbance features are extracted from the preprocessed results, and anomaly determination is made using the isolation forest algorithm. The results are summarized and, based on a preconfigured rule engine, the failure rate of the current time series is obtained. Compared to the failure rate of the previous time series, the failure trend is analyzed to obtain the failure trend. At the same time, the failure rate and failure change trend are marked as monitoring feature vectors; Fault output: Multiple time series are randomly selected as target variables, and the failure rate and failure change trend are substituted into the pre-established logistic regression model to calculate the operating probability of the corresponding operating state at each time node, including the first probability of normal state, the second probability of excessive conducted interference fault, the third probability of excessive radiated interference fault, the fourth probability of ground loop resonance state, and the fifth probability of imbalance state.

[0006] Furthermore, during the fault output process, fault assessment indicators are generated, including: The fault change trend is abstracted into a fault change curve. Through cluster analysis, the fault diffusion rate is obtained, and several state cycles are obtained based on the fault diffusion rate. Based on the state cycle, the initial weight is defined; the fault assessment model is defined by minimizing the failure rate and minimizing the fault diffusion speed. The particle swarm algorithm is used to update the weight and optimize the fault assessment model. The fitness value corresponding to the individual is calculated through the fault assessment model, and the weight combination of the individual with the highest fitness value in the population is selected. The current failure rate and fault diffusion speed are weighted and summed to generate the fault assessment index.

[0007] Furthermore, fault types include excessive conducted interference, excessive radiated interference, and ground loop resonance. Based on different fault types, several suppression rules are triggered, including: If the fault type is excessive conducted interference, the dynamic filtering strategy is triggered; If the fault type is excessive radiation interference, the shielding and reconstruction strategy is triggered; If the fault type is ground loop resonance, the impedance tuning strategy is triggered.

[0008] Furthermore, if the fault type is excessive conducted interference, a dynamic filtering strategy is triggered, including: Characteristic identification: Extract the characteristics of conducted interference, which are characterized by the center frequency of the interference spectrum; Switching analysis: Based on the characteristics of conducted interference, the topology of the currently running LC filter group is switched. The LC filter group includes a third-order π-type filter circuit and a second-order band-stop filter circuit. The switching logic is: If the center frequency of the interference spectrum is less than or equal to the preset frequency threshold, the third-order π-type filter circuit is activated; If the center frequency of the interference spectrum is greater than the preset frequency threshold, the second-order band-stop filter circuit is activated; Reverse compensation current injection: Common-mode noise is offset by injecting reverse compensation current to achieve conducted interference suppression.

[0009] Furthermore, if the fault type is excessive radiation interference, the shielding and reconstruction strategy is triggered, including: Radiation source positioning: extracting radiation interference characteristics, which are characterized by radiation spatial distribution and intensity; Shielding layer adjustment: Based on the radiation interference characteristics, the opening and closing degree of the adjustable shielding layer is controlled; Grounding optimization: Dynamically switch the grounding method of the shielding layer based on the characteristics of radiation interference.

[0010] Furthermore, if the fault type is ground loop resonance, the impedance tuning strategy is triggered, including: Impedance location: Extract the ground loop characteristics, which are characterized by the resonant frequency and impedance amplitude of the loop; Adjustment analysis: Based on the ground loop characteristics, the proportional term, integral term, and differential term are dynamically adjusted through the PID control algorithm until the resonance phenomenon is eliminated.

[0011] Furthermore, the risk assessment model is established based on the LSTM model architecture.

[0012] Furthermore, if the interference suppression effectiveness index is lower than the effectiveness threshold, a strategy compensation instruction is generated and fed back to step S2 to iteratively adjust the parameters, including: Obtaining a difference between an interference suppression effectiveness index and an effectiveness threshold, and obtaining a strategy compensation instruction based on the difference; The strategy compensation instruction is passed to the parameter adjustment link corresponding to the suppression rule, the parameters are reconfigured and executed cyclically until the efficiency threshold is met.

[0013] Furthermore, the state cycle includes an early stage, a middle stage, and a late stage; based on the state cycle, the initial weight is defined, including: the initial weight in the early stage is smaller than the initial weight in the middle stage, which is smaller than the initial weight in the late stage.

[0014] (3) Beneficial effects The present invention provides a method for predicting electromagnetic compatibility faults of a pulse withstand voltage generator, which has the following beneficial effects: 1. This invention collects raw signals from the pulse withstand voltage generator in real time under different circuit states, such as charging, resonance, and discharge, and extracts voltage ripple, current phase, and magnetic field radiation spectrum signals to generate a time-series monitoring data set and conduct in-depth analysis. This multi-dimensional, full-state data collection and analysis method can more comprehensively capture the characteristics of electromagnetic disturbances. Compared with traditional single-parameter monitoring, it can significantly improve the accuracy of fault probability calculation and fault type judgment, and detect potential electromagnetic compatibility faults in advance. 2. This invention achieves early identification of fault types (conduction / radiation / resonance) through the fusion of multi-dimensional electromagnetic features, avoiding misjudgments and missed judgments of traditional threshold alarms. By matching optimal suppression rules for different fault types, such as dynamic filtering, shielding reconstruction, and impedance tuning, precise management is achieved, improving the efficiency and effectiveness of fault handling, and making it more targeted and effective. 3. After executing the suppression rules, the present invention generates an interference suppression efficiency index through a risk assessment model and compares it with a preset threshold. If the requirements are not met, feedback is given to iteratively adjust the parameters. This closed-loop feedback mechanism can continuously optimize the suppression strategy, so that the system can automatically adjust according to actual conditions, find the optimal solution, continuously improve the electromagnetic interference suppression effect, enhance the stability and reliability of the system, and deal with electromagnetic interference problems in a timely manner, which can reduce the damage of electromagnetic interference to electronic components. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a schematic diagram showing steps of an electromagnetic compatibility fault prediction method according to an exemplary embodiment; Figure 21 is a control timing diagram of angle proportions of different circuit states according to an exemplary embodiment. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example: The embodiment of the present invention provides a method for predicting electromagnetic compatibility faults of a pulse withstand voltage generator; Figure 1 is a schematic diagram showing steps of an electromagnetic compatibility fault prediction method according to an exemplary embodiment; Figure 2 is a schematic diagram of control timing of different circuit state angle proportions according to an exemplary embodiment; please refer to Figures 1 to 2 , the method comprises the following steps: S1: State Monitoring: This collects raw signals from the pulse withstand voltage generator in real time at different circuit states, extracts voltage ripple, current phase, and magnetic field radiation spectrum, and generates a multi-dimensional time-series monitoring data set. Different circuit states include charging, resonant, and discharging. It should be noted that different circuit states describe different stages of the pulse withstand voltage generator; First, set the pulse withstand voltage generator parameters based on the design indicators in Table 1: Table 1 Design indicators

[0018] Then, the original signals under different circuit states are monitored based on the parameters of the pulse withstand voltage generator, and the original signals include current signals, voltage signals and magnetic field radiation signals; For example: Assume that in one working cycle, the circuit is set as follows: Charging state: If the charging trigger signal arrives, the charging switch tube is turned on, the discharging switch tube and the resonant switch tube are turned off, the input power charges the parallel main capacitors (such as: C1 to Cn) through the current limiting inductor, and the resonant power supply charges the resonant capacitor Cr; Resonant state: After the charging state is completed, the resonance trigger signal arrives; the charging switch tube is turned off, and the resonant switch tube is turned on; the main capacitors (such as: C1 to Cn) maintain the current potential and wait for the discharge switch tube to turn on; the resonant capacitor Cr and the resonant inductor Lr in the resonant circuit begin to resonate in parallel; Discharge state: When the discharge trigger signal arrives, the resonant switch tube remains on, the discharge switch tube is turned on, the main capacitor (such as: C1 to Cn) and the resonant capacitor Cr are connected in series to discharge the equivalent load Rl; the main capacitor generates a pulse drop when discharging; Then we have: Voltage ripple: A high-voltage differential probe is used, connected in parallel with the main capacitor group (e.g., C1 to Cn). The input signal amplitude is reduced through the principle of capacitive voltage division to match the range requirements of the subsequent acquisition circuit. A 16-bit high-precision ADC module with a sampling frequency of 1MHz is used to capture the high-frequency components of the ripple signal. The ADC module digitizes the collected voltage signal to obtain the original time domain data. The digitized signal is processed using a sliding window algorithm with the window size set to a certain number of sampling points (e.g., 1024). The standard deviation of the signal within each window is calculated to obtain the quantized value of the voltage ripple. Current phase: A Rogowski coil is used to collect current signals. A series resonant circuit (e.g., the Cr to Lr path) is connected to the Rogowski coil. The resonant characteristics of the inductor Lr and capacitor Cr are used to compensate for the coil's frequency response. A capacitive voltage divider is connected in parallel with the resonant capacitor Cr to achieve precise signal attenuation and phase calibration. The processed current signal and reference voltage signal are collected, and the phase difference between the two signals is calculated using a cross-correlation algorithm. Magnetic field radiation spectrum: A near-field probe array is used, including three magnetic field probes in orthogonal directions (X, Y, and Z axes), covering a frequency range of 10kHz-1GHz. It is equipped with a low-noise preamplifier (noise figure ≤3dB) and a bandpass filter group to enhance weak signal detection capabilities. A high-speed ADC (sampling rate 2.5GS / s) is used to digitize broadband signals. The near-field probe array synchronously acquires magnetic field radiation signals in three dimensions. A fast Fourier transform (FFT) is performed on the time-domain signal to convert it into a frequency-domain energy distribution. For example, if the spectrum resolution is set to 1kHz and the dynamic range is ≥80dB, the output result is a magnetic field radiation spectrum, including the frequency (Hz) and the energy value of the corresponding frequency band. The transitions between charging, resonant, and discharging states are completed in microseconds. The voltage ripple, current phase, and magnetic field radiation spectrum are time-aligned and aggregated to obtain a multi-dimensional time-series monitoring data set. It should be noted that the digital settings mentioned above are examples only and should be set up according to actual conditions. The above devices or sensors are not shown in the figure and should be installed adaptively according to actual conditions. For example, the high-voltage differential probe should be installed near the main capacitor bank to reduce lead interference; the Rogowski coil should be wrapped around the outside of the current bus to ensure uniform coupling; the near-field probe array should cover the key electromagnetic radiation source area of ​​the equipment and is recommended to be installed at the same height as the radiation source; all signal cables should be shielded and single-ended grounded to avoid introducing additional electromagnetic interference. In addition, in a complete working cycle, the three stages of charging, resonance, and discharge of the pulse withstand voltage generator have different proportions within a cycle, which are usually quantified in the form of angles. Similarly, a complete working cycle is regarded as 360°, and the corresponding angle value of each stage reflects its time proportion; for example: t1=330°, t2=335°, t1=347°, t1=351°, t1=355°, t1=360° By real-time acquisition of original signals from the pulse withstand voltage generator under different circuit states, such as charging, resonance, and discharge, and extracting voltage ripple, current phase, and magnetic field radiation spectrum signals, a time-series monitoring data set is generated and in-depth analysis is performed. This multi-dimensional, full-state data acquisition and analysis method can more comprehensively capture electromagnetic disturbance characteristics. Compared with traditional single-parameter monitoring, it can greatly improve the accuracy of fault probability calculation and fault type judgment, and detect potential electromagnetic compatibility faults in advance.

[0019] S2: Fault Prediction: Based on time-series monitoring data sets, this system extracts multi-dimensional electromagnetic disturbance features through in-depth analysis and builds a rule engine to output fault probability and fault type. Fault types include excessive conducted interference, excessive radiated interference, and ground loop resonance. Based on different fault types, several suppression rules are triggered: If the fault type is excessive conducted interference, the dynamic filtering strategy is triggered; If the fault type is excessive radiation interference, the shielding and reconstruction strategy is triggered; If the fault type is ground loop resonance, the impedance tuning strategy is triggered; Through in-depth analysis, electromagnetic disturbance characteristics are extracted and a rule engine is built to output fault probability and fault type, including: Data processing: preprocessing of time series monitoring data sets, including denoising, normalization, and modal decomposition; Denoising: Wavelet threshold denoising, such as using the sym5 wavelet basis function to perform a three-layer decomposition of the signal, dividing the signal into different frequency bands; median filtering, such as setting a median filter with a window size of 5 for impulse noise, replacing the value of each point in the signal with the median of all the points in the window, eliminating isolated spikes in the signal; adaptive notch filtering: Real-time monitoring of the signal spectrum. If periodic interference of a specific frequency (such as 50Hz grid harmonics) is detected, the center frequency and bandwidth of the notch filter are dynamically adjusted, and the adaptive least mean square (LMS) algorithm is used to optimize the filter parameters to accurately suppress periodic interference; Standardization: Z-score standardization method is used to eliminate dimensional differences and amplitude fluctuations between different parameters to ensure consistency and comparability of data in subsequent analysis; Modal decomposition: The variational modal decomposition (VMD) algorithm is used to adaptively decompose the signal into multiple intrinsic mode functions (IMFs). For example, the number of modes is set to 4, and the penalty factor is set to 2000 to balance decomposition accuracy and computational efficiency. Spectral analysis and correlation coefficient calculation are performed on the decomposed IMFs to screen out modal components related to fault characteristics, including the fundamental component reflecting normal equipment operation, transient interference components including sudden interference, and harmonic components reflecting the nonlinear characteristics of the equipment. Feature analysis: Based on the pre-processed results, support vector machine (SVM) technology is used to make qualification judgments and output binary results to indicate whether the system meets the normal operation characteristics. It should be noted that SVM is a binary classification model based on statistical theory. Its basic model definition is to seek the maximum interval seen by the feature, which enhances its strong generalization ability in solving EMC fault diagnosis applications. Kernel function selection: Radial basis function (RBF) is used as the kernel function of SVM to adapt to nonlinear classification problems in high-dimensional space and effectively capture the complex boundaries between fault characteristics and normal characteristics; Model training: The preprocessed dataset is divided into a training set and a validation set in a ratio of 7:3. A grid search combined with cross-validation is used to optimize the penalty and kernel parameters of the SVM, selecting the parameter combination that achieves the highest accuracy on the validation set. The optimized parameters are then used to train the SVM model based on the training set data to construct the classification boundaries for the normal operation characteristics of the equipment. The selected modal components are used as the input feature vectors of the SVM model. Result output: The SVM model uses the input feature vector to calculate the decision function of the classification hyperplane and outputs a binary result. 1 indicates that the equipment's operating characteristics are normal, and 0 indicates an abnormality, providing a key basis for subsequent fault diagnosis. Extract electromagnetic disturbance features that can characterize electromagnetic compatibility faults from the preprocessed results, use the isolation forest algorithm to make anomaly judgments, and output anomaly confidence levels. It should be noted that based on the preprocessed results, time-domain and frequency-domain feature extraction is performed. The extracted features are screened using the Pearson correlation coefficient method, redundant features with large correlation coefficients are eliminated, and valuable features are retained and marked as electromagnetic disturbance features to form the input feature set of the isolation forest. Considering the sparsity of electromagnetic disturbances, the number of trees can be set to a certain number (e.g., 100), and the sample size of each tree can be a certain proportion of the total samples (e.g., 20%) to enhance the capture of low-probability anomalies. The maximum depth of the tree is limited to the log2 value of the feature dimension to avoid overfitting the subtle fluctuations of normal samples. In response to the time-varying characteristics of electromagnetic disturbances (e.g., the transient interference duration during the discharge phase is short), time decay is introduced. Due to the sample timing difference, the isolated path length of historical samples is weighted, giving a higher weight to recent abnormal features to improve sensitivity to sudden faults. The anomaly score is calculated by the ratio of the average path length of the sample in all trees to the average path length of normal samples. The score range of the anomaly score is mapped to the interval [0, 1] to output the anomaly confidence level. Summarize the judgment results. Based on the pre-configured rule engine, perform a weighted fusion of the SVM binarization results and the isolation forest anomaly confidence level to obtain the failure rate of the current time series. Compared with the failure rate of the previous time series, the difference between the current time series failure rate and the previous time series failure rate is calculated to analyze the failure change trend. At the same time, the failure rate and failure change trend are marked as monitoring feature vectors; Fault output: Multiple time series are randomly selected as target variables. The failure rate and failure trend are substituted into a pre-established logistic regression model to calculate the operating probability of the corresponding operating state at each time node. This includes the first probability of normal state, the second probability of excessive conducted interference fault, the third probability of excessive radiated interference fault, the fourth probability of ground loop resonance state, and the fifth probability of imbalance state. The calculation formula of the logistic regression model is: ; Where d(y=1|D) represents a certain operation probability obtained by monitoring feature vector D corresponding to different circuit states in a certain time interval. Monitoring feature vector D includes failure rate and failure change trend. y represents the target variable, and Z represents the fault assessment index. The steps for obtaining fault assessment indicators include: The fault change trend is abstracted into a fault change curve. Through cluster analysis, the fault diffusion rate is obtained. Based on the fault diffusion rate, several state cycles are obtained, including the initial, middle and final stages. Assume that there are N0 points on the fault change curve, set several time window scales, and calculate the slope of the vertical coordinates of N0 points within multiple time window scales. For the slope sequence of each scale w, calculate the difference between the slopes of adjacent windows, and obtain the curve change rate by calculating the difference between the slopes of any two points in N0. Combine the time series monitoring data set (such as voltage ripple CV, phase deviation Δθ, magnetic field radiation spectrum E_rad) with the charging state as the feature space. Based on the corresponding curve change rate, use DBSCAN clustering to identify the curve trend. For example, normal operating points should be clustered as core points, showing a steady-state oscillation trend. Fault operating points (such as excessive conducted interference and excessive radiated interference) form independent clusters when uncompensated and show an upward trend. Ground resonance operating points are identified as outliers and show a downward trend. Further identify the change rate of steady-state oscillation trend, upward trend, and downward trend. At the same time, mark the change rate as the fault diffusion rate. In this embodiment, the state cycle is defined mainly for the change rate of fault operating points and ground resonance operating points: If the rate of change is a sudden increase or decrease, the curve appears as a broken line and is defined as the initial stage; If the rate of change is increasing or decreasing, the curve is parabolic and is defined as the medium term. If the rate of change is in a slowly increasing or slowly decreasing state, the curve is in a parabolic form and is defined as the terminal stage; Based on the state cycle, the initial weight is defined; the initial weight in the early stage is smaller than the initial weight in the middle stage, which is smaller than the initial weight in the final stage; the fault assessment model is defined by minimizing the failure rate and minimizing the failure propagation speed: ; In the formula, min(Z) represents the fault assessment model, min(gz) represents the minimization of the failure rate, min(ks) represents the minimization of the fault diffusion speed, b and c represent weights, and b and c are greater than 0 and b+c=1; The particle swarm algorithm is used to update the weights and optimize the fault assessment model. The fitness value of the individual is calculated through the fault assessment model. The weighted combination of the individual with the highest fitness value in the population is selected. The current failure rate and fault diffusion speed are weighted and summed to generate the fault assessment index. It should be noted that the process of establishing the fault assessment model also includes converting the parameters involved in the algorithm model into dimensionless form to facilitate subsequent analysis and calculation. The steps of using the particle swarm algorithm to update the weights include: Randomly generate an initial population, which consists of m0 particles and mark any particle as i. During initialization, randomly generate a certain number of individuals, that is, the corresponding weight combination in this embodiment, to ensure that all weight values ​​are within the appropriate range and the range is between [0, 1], and the sum of the weight values ​​is 1; Based on individual fitness, individuals with high fitness are selected as parents. Roulette wheel selection strategy is used to select particles from the population and put them into the mating pool in turn until the number of particles in the mating pool reaches m0. New individuals are obtained by performing selection, crossover, and mutation operations on the population and added to the population to update the population composition. If the individual fitness is the highest, the iteration is terminated, and the current failure rate and failure diffusion speed are weighted and summed to generate the failure assessment index: ; Where Z represents the fault assessment index, b* and c* are the updated weights; If the fault type is excessive conducted interference, the dynamic filtering strategy is triggered, including: Characteristic identification: Extract the characteristics of conducted interference, which are characterized by the center frequency of the interference spectrum; Specifically, it includes: identifying the original signal (such as voltage signal or current signal) that exceeds the conducted interference standard in the current working cycle, removing the DC component through baseline correction, using Hanning window to perform windowing processing on the signal to reduce spectrum leakage, performing FFT transformation on the signal, converting the time domain signal into a frequency domain spectrum, and obtaining a frequency-amplitude distribution curve; through peak recognition, determining the frequency point with the highest amplitude in the spectrum curve, that is, the main interference frequency, and recording the frequency value f p and the corresponding amplitude A p , with the main peak as the center, expand to both sides to the frequency point where the amplitude drops by a certain amount (for example: 3dB), and determine the interference bandwidth; if the interference spectrum is unimodal and symmetrical, the center frequency f c Equal to the frequency value f p ; If there are multiple adjacent peaks in the spectrum (such as harmonic interference), use the energy weighting method: Where, f i is the peak frequency, A i represents the corresponding amplitude, and n represents the number of significant peaks; Switching analysis: Based on the characteristics of conducted interference, switch the currently operating LC filter bank topology; The LC filter group includes a third-order π-type filter circuit and a second-order band-stop filter circuit, and the switching logic is: If the center frequency of the interference spectrum is ≤ the preset frequency threshold (e.g., 500kHz), the third-order π-type filter circuit is activated: the third-order π-type filter circuit consists of two-stage capacitors (C1, C2) and an intermediate inductor (L); If the center frequency of the interference spectrum is greater than the preset frequency threshold (e.g. 500kHz), the second-order band-stop filter circuit is activated: the second-order band-stop filter circuit adopts an LC parallel resonant structure; Reverse compensation current injection: Common-mode noise is offset by injecting reverse compensation current to achieve conducted interference suppression: Adopting an adaptive least mean square (LMS) algorithm, a reverse compensation current is generated in real time based on the detected common-mode current. The compensation current is amplified by the power amplifier and injected into the circuit to offset the common-mode noise. The residual current after compensation is monitored in real time, and the amplitude and phase of the compensation current are dynamically adjusted to ensure the best suppression effect. If the fault type is excessive radiation interference, the shielding and reconstruction strategy is triggered, including: Radiation source positioning: extracting radiation interference characteristics, which are characterized by radiation spatial distribution and intensity; Specifically, it includes: identifying the original signal (such as the magnetic field radiation signal) that identifies excessive radiation interference in the current working cycle, locating the point of maximum electromagnetic intensity, that is, the core position of the radiation source; with the peak point as the center, expanding outward to the area where the intensity drops to several peaks (such as -3dB), fitting the equivalent spatial range of the radiation source (usually an ellipsoid), recording the major axis, minor axis length and spatial pointing angle to determine the spatial distribution of the radiation; calculating the intensity attenuation curves at different azimuth angles (0° to 360°) and pitch angles (-90° to 90°), determining the main radiation direction (the direction with the slowest intensity attenuation), and determining the radiation intensity; Shielding layer adjustment: Based on the radiation interference characteristics, the opening and closing degree of the adjustable shielding layer is controlled; for example, for areas with higher radiation intensity, the coverage area and number of shielding layers are increased; for areas with weaker radiation, the shielding layers are appropriately reduced to reduce cost and weight; Grounding optimization adjustment: Based on the characteristics of radiation interference, the grounding method of the shield layer is dynamically switched; for example, when the radiation interference frequency is lower than 100MHz, a single-point grounding method is used to reduce the grounding impedance; when the radiation interference frequency is higher than 100MHz, a multi-point grounding method is switched to reduce ground loop interference; If the fault type is ground loop resonance, the impedance tuning strategy is triggered, including: Impedance location: Extract the ground loop characteristics, which are characterized by the resonant frequency and impedance amplitude of the loop; Specifically, it includes: identifying the original signal (such as voltage or current signal) that identifies ground loop resonance during the current working cycle, converting it into a frequency domain spectrum (with the horizontal axis being frequency and the vertical axis being amplitude) through Fourier transform (FFT) to identify the resonant frequency; and synchronously measuring the voltage or current spectrum; Adjustment analysis: Based on the ground loop characteristics, the proportional, integral, and differential terms are dynamically adjusted through the PID control algorithm until the resonance phenomenon is eliminated; It should be noted that the system has built-in target resonant frequency and target impedance amplitude. The resonant frequency offset is calculated by the difference between the resonant frequency of the loop and the target resonant frequency, and the impedance amplitude offset is calculated by the difference between the impedance amplitude of the loop and the target impedance amplitude. For any offset, a gradient range of ±wc% is set. If the offset is within the gradient range of ±wc%, no processing is performed. If the offset exceeds the gradient range of ±wc%, the PID control algorithm is used to respond to the proportional term to adjust the resonant frequency offset and the impedance amplitude offset. The proportional term is used to control the adjustment speed and resonance convergence, the integral term is used to eliminate steady-state residual deviation, and the differential term is used to suppress oscillation and smooth the adjustment. Through the fusion of multi-dimensional electromagnetic features, early identification of fault types (conduction / radiation / resonance) can be achieved, avoiding the misjudgment and omission of traditional threshold alarms; by matching the optimal suppression rules for different fault types, such as dynamic filtering, shielding reconstruction, impedance tuning, etc., precise management can be achieved, and the efficiency and effect of fault handling can be improved, making it more targeted and effective; it can effectively suppress various electromagnetic interference faults, ensure the normal operation of high-voltage systems and communication systems of new energy vehicles, reduce the impact of electromagnetic interference on electronic equipment in the vehicle, improve the stability and reliability of the vehicle's electrical system, and ensure vehicle driving safety.

[0020] S3: Suppression evaluation: After executing any suppression rule, the original signal of the pulse withstand voltage generator is collected again and input into the pre-built risk assessment model to generate the interference suppression effectiveness index; It should be noted that the risk assessment model is built based on the LSTM model architecture, including but not limited to the input layer, LSTM hidden layer, Dropout layer, fully connected layer, and output layer: the input layer is set to 3 nodes, the LSTM hidden layer is set to 128 units, and the activation function is tanh; the Dropout layer is set to a dropout rate of 0.2; the fully connected layer is set to 3 linear nodes, the activation function is ReLU, and the output layer is 1 linear node; Compare the interference suppression effectiveness index with the preset effectiveness threshold: If the interference suppression effectiveness index is lower than the effectiveness threshold, a strategy compensation instruction is generated and fed back to step S2 to iteratively adjust the parameters; Generate strategy compensation instructions and feed them back to step S2 to iteratively adjust parameters, including: Obtain the difference between the interference suppression effectiveness index and the effectiveness threshold, and obtain strategy compensation instructions based on the difference, including fine-tuning compensation instructions, adjusting compensation instructions, and reconstructing compensation instructions: Compare the difference with the difference interval: If the difference is less than the difference interval, it is judged as a slight deficiency and a slight compensation instruction is issued: if the interference suppression effect is good but there is still a slight gap, the suppression rule parameters are adjusted linearly by a small amount (for example, the adjustment range is usually 5%-10% of the current parameter value); for example, if it is a shielding layer adjustment, if there is a slight deficiency, only the opening degree is slightly adjusted (for example, from 30° to 33°); If the difference is within the difference range, it is determined to be moderately insufficient and a moderate compensation instruction is issued: the interference suppression effect is average and the parameters need to be adjusted significantly. For example, in ground loop resonance suppression, the integral term of the PID algorithm is activated (for example, increase it by 20%) and the proportional term is fine-tuned (for example, increase it by 15%) to accelerate the elimination of steady-state deviation. If the difference exceeds the difference range, it is judged as seriously insufficient and a heavy compensation instruction is issued: the interference suppression effect is significantly reduced, or the system experiences abnormal conditions such as continuous oscillation and component overheating. For example, in radiation interference suppression, the adjustment is upgraded from a single shielding layer to a composite strategy of shielding layer and absorbing material, and the grounding method is redesigned (such as from single-point grounding to multi-point grounding). In addition, after heavy compensation, a hardware self-test is required (such as testing the capacitance and inductance withstand voltage values, and the mechanical strength of the shielding layer) to avoid secondary failures caused by drastic adjustments. Pass the policy compensation instruction to the parameter adjustment link corresponding to the suppression rule, reconfigure the parameters and execute it cyclically until the efficiency threshold is met; If the interference suppression effectiveness index exceeds the effectiveness threshold, the corresponding parameters of the current suppression rule are output as the optimal solution; In this embodiment, the process of presetting the effectiveness threshold is as follows: collecting the interference suppression effectiveness index of the pulse withstand voltage generator in past operations to understand the range of the interference suppression effectiveness index under effective suppression (i.e., when the parameters corresponding to the current suppression rule are the optimal solution); performing statistical analysis on the collected data to determine the average value and standard deviation of the suppression effectiveness index under effective suppression; setting a standard threshold based on the statistical results, and setting the standard threshold to the average value of the effective suppression effectiveness index plus a certain multiple of the standard deviation (e.g., 2 times or 3 times the standard deviation); The setting of the difference interval is similar to this and will not be explained in detail here. The minimum interval is set to the mean of the difference minus a certain multiple of the standard deviation (such as 2 or 3 times the standard deviation), and the maximum interval is set to the mean of the difference plus a certain multiple of the standard deviation (such as 2 or 3 times the standard deviation). It should be noted that the priority is: heavy compensation instructions > moderate compensation instructions > light compensation instructions, ensuring that strong intervention measures are executed first in emergency situations. Through this hierarchical compensation mechanism, the system can dynamically select the optimal adjustment strategy based on the quantitative difference in interference suppression effect, achieving a balance between rapid response and precise control, ensuring fault suppression efficiency and stability in complex electromagnetic environments. After executing the suppression rules, an interference suppression effectiveness index is generated through the risk assessment model and compared with the preset threshold. If the requirements are not met, feedback is given to iteratively adjust the parameters. This closed-loop feedback mechanism can continuously optimize the suppression strategy, so that the system can automatically adjust according to actual conditions, find the optimal solution, continuously improve the electromagnetic interference suppression effect, enhance the stability and reliability of the system, and deal with electromagnetic interference problems in a timely manner. It can reduce the damage of electromagnetic interference to electronic components, reduce the speed of aging and damage of components due to long-term interference, extend the service life of new energy vehicle-related equipment and components, and reduce maintenance costs.

[0021] In the application, the several formulas involved are all calculated by taking their numerical values ​​after removing the dimensions, and the formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The formula is set by technical personnel in this field according to actual conditions.

[0022] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment based on actual needs.

[0024] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for predicting electromagnetic compatibility faults of a pulse withstand voltage generator, characterized in that: The method comprises: S1: Real-time acquisition of raw signals from the pulse withstand voltage generator in different circuit states, extraction of voltage ripple, current phase, and magnetic field radiation spectrum, and generation of a time-series monitoring data set; wherein the circuit states include charging, resonance, and discharging; S2: Based on the time series monitoring data set, extract electromagnetic disturbance features through in-depth analysis, build a rule engine, output fault probability and fault type; based on the fault type, trigger several suppression rules; S3: After executing any of the suppression rules, the original signal of the pulse withstand voltage generator is collected again and input into a pre-built risk assessment model to generate an interference suppression efficiency index; the interference suppression efficiency index is compared with a preset efficiency threshold: if the interference suppression efficiency index is lower than the efficiency threshold, a strategy compensation instruction is generated and fed back to step S2 to iteratively adjust the parameters; if the interference suppression efficiency index exceeds the efficiency threshold, the parameters corresponding to the current suppression rule are output as the optimal solution.

2. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 1, characterized in that: Based on the time series monitoring data set, the electromagnetic disturbance characteristics are extracted through in-depth analysis, and a rule engine is built to output the fault probability and fault type, including: Data processing: preprocessing of time series monitoring data sets, including denoising, normalization, and modal decomposition; Feature analysis: Based on the preprocessed results, a support vector machine is used to determine if the system is qualified. Electromagnetic disturbance features are extracted from the preprocessed results, and anomaly determination is made using the isolation forest algorithm. The results are summarized and, based on a preconfigured rule engine, the failure rate of the current time series is obtained. Compared to the failure rate of the previous time series, the failure trend is analyzed to obtain the failure trend. At the same time, the failure rate and failure change trend are marked as monitoring feature vectors; Fault output: Multiple time series are randomly selected as target variables, and the failure rate and failure change trend are substituted into the pre-established logistic regression model to calculate the operating probability of the corresponding operating state at each time node, including the first probability of normal state, the second probability of excessive conducted interference fault, the third probability of excessive radiated interference fault, the fourth probability of ground loop resonance state, and the fifth probability of imbalance state.

3. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 2, characterized in that: During the fault output process, fault assessment indicators are also generated, including: The fault change trend is abstracted into a fault change curve. Through cluster analysis, the fault diffusion rate is obtained, and several state cycles are obtained based on the fault diffusion rate. Based on the state cycle, the initial weight is defined; the fault assessment model is defined by minimizing the failure rate and minimizing the fault diffusion speed. The particle swarm algorithm is used to update the weight and optimize the fault assessment model. The fitness value corresponding to the individual is calculated through the fault assessment model, and the weight combination of the individual with the highest fitness value in the population is selected. The current failure rate and fault diffusion speed are weighted and summed to generate the fault assessment index.

4. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 1, characterized in that: The fault types include excessive conducted interference, excessive radiated interference, and ground loop resonance. Based on different fault types, several suppression rules are triggered, including: If the fault type is excessive conducted interference, the dynamic filtering strategy is triggered; If the fault type is excessive radiation interference, the shielding and reconstruction strategy is triggered; If the fault type is ground loop resonance, the impedance tuning strategy is triggered.

5. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 4, characterized in that: If the fault type is that the conducted interference exceeds the standard, the dynamic filtering strategy is triggered, including: Characteristic identification: Extract the characteristics of conducted interference, which are characterized by the center frequency of the interference spectrum; Switching analysis: Based on the characteristics of conducted interference, the topology of the currently running LC filter group is switched. The LC filter group includes a third-order π-type filter circuit and a second-order band-stop filter circuit. The switching logic is: If the center frequency of the interference spectrum is less than or equal to the preset frequency threshold, the third-order π-type filter circuit is activated; If the center frequency of the interference spectrum is greater than the preset frequency threshold, the second-order band-stop filter circuit is activated; Reverse compensation current injection: Common-mode noise is offset by injecting reverse compensation current to achieve conducted interference suppression.

6. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 4, characterized in that: If the fault type is that the radiation interference exceeds the limit, the shielding reconstruction strategy is triggered, including: Radiation source positioning: extracting radiation interference characteristics, which are characterized by radiation spatial distribution and intensity; Shielding layer adjustment: Based on the radiation interference characteristics, the opening and closing degree of the adjustable shielding layer is controlled; Grounding optimization: Dynamically switch the grounding method of the shielding layer based on the characteristics of radiation interference.

7. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 4, characterized in that: If the fault type is ground loop resonance, the impedance tuning strategy is triggered, including: Impedance location: Extract the ground loop characteristics, which are characterized by the resonant frequency and impedance amplitude of the loop; Adjustment analysis: Based on the ground loop characteristics, the proportional term, integral term, and differential term are dynamically adjusted through the PID control algorithm until the resonance phenomenon is eliminated.

8. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 1, characterized in that: The risk assessment model is established based on the LSTM model architecture.

9. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 1, characterized in that: If the interference suppression effectiveness index is lower than the effectiveness threshold, a strategy compensation instruction is generated and fed back to step S2 to iteratively adjust the parameters, including: Obtaining a difference between an interference suppression effectiveness index and an effectiveness threshold, and obtaining a strategy compensation instruction based on the difference; The strategy compensation instruction is passed to the parameter adjustment link corresponding to the suppression rule, the parameters are reconfigured and executed cyclically until the efficiency threshold is met.

10. The electromagnetic compatibility fault prediction method for a pulse withstand voltage generator according to claim 1, characterized in that: The state cycle includes an early stage, a middle stage and a late stage; based on the state cycle, the initial weight is defined, including: the initial weight in the early stage is smaller than the initial weight in the middle stage, which is smaller than the initial weight in the late stage.

Images