An adaptive voltage dynamic stabilization control method for a new energy vehicle converter

Abstract

The application relates to the technical field of electric control of new energy vehicles, and discloses a self-adaptive voltage dynamic stabilization control method for a new energy vehicle converter. The method comprises the following steps: acquiring real-time voltage time sequence data sequences of the new energy vehicle converter in a running period; performing voltage disturbance feature analysis on the real-time voltage time sequence data sequences based on a converter rated voltage interval, and identifying converter voltage disturbance mode information; then, generating an initial voltage regulation instruction sequence according to an embedded voltage control strategy library and voltage stabilization threshold conditions, and in combination with the converter voltage disturbance mode information; performing voltage change trend prediction on the converter voltage disturbance mode information, and obtaining predicted voltage disturbance mode information; performing voltage control compensation optimization on the initial voltage regulation instruction sequence based on the predicted voltage disturbance mode information, and generating an optimized voltage regulation instruction sequence; and performing self-adaptive voltage regulation on the new energy vehicle converter according to the optimized voltage regulation instruction sequence.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electric control of new energy vehicles, and particularly relates to a self-adaptive voltage dynamic stability control method for a new energy vehicle converter. BACKGROUND

[0002] In the operation system of a new energy vehicle, as a core component for electric energy conversion and transmission, the stability of the voltage output of a converter is directly related to the power performance of the vehicle, the energy utilization efficiency, and the safe operation of the electric equipment. With the increasing requirements of new energy vehicles for the cruising range and the power response speed, the converter needs to work continuously under complex and changeable working conditions, such as a large current impact during vehicle start-up and acceleration, load fluctuation during high-speed driving, and energy reverse transmission during the brake energy recovery process. These working condition changes will cause different degrees of disturbance of the input and output voltages of the converter.

[0003] The traditional voltage control method of the new energy vehicle converter is mostly based on a fixed control strategy, that is, a regulation and control action is triggered according to a preset voltage threshold, and lacks the dynamic identification and adaptive ability of the voltage disturbance mode. In the actual operation process, the voltage disturbance of the converter often presents multiple modal characteristics, such as instantaneous pulse disturbance, continuous fluctuation disturbance, and gradual drift disturbance. The influence mechanism of different disturbance modes on voltage stability and the required regulation and control strategy are significantly different. Due to the inability to accurately analyze the disturbance mode information, the traditional control method can only use a unified regulation and control logic to cope with various voltage fluctuations, and is prone to regulation and control lag or over-regulation problems.

[0004] When facing instantaneous pulse disturbance, the traditional control method may fail to suppress voltage mutation in time due to detection delay, thereby causing impact on subsequent power electronic devices. In the continuous fluctuation disturbance scenario, fixed regulation and control parameters are difficult to match the frequency and amplitude of voltage fluctuation, which easily causes voltage oscillation and affects the normal work of the electric equipment. For gradual drift disturbance, the traditional method usually starts regulation and control only after the voltage exceeds the preset threshold, which causes the voltage to be in a non-stable interval for a long time, thereby reducing the energy conversion efficiency. The traditional control method mostly does not have voltage change trend prediction capability, and can only passively respond to the voltage fluctuation that has occurred, and cannot take preventive measures in advance, further increasing the difficulty of voltage stability control. With the development of new energy vehicle technology towards high power density and high integration, the working condition of the converter is more complex, and the limitations of the traditional voltage control method are increasingly prominent, and the traditional voltage control method has been difficult to meet the high-precision requirements of the vehicle for voltage stability. Therefore, a control method capable of realizing dynamic identification, trend prediction and self-adaptive regulation and control of voltage disturbance modes is urgently needed. SUMMARY

[0005] The present application aims to provide a new energy vehicle converter adaptive voltage dynamic stability control method to solve the problems in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides a new energy vehicle converter adaptive voltage dynamic stability control method, which comprises:

[0007] Obtain the real-time voltage time series data sequence of the new energy vehicle converter within the operation cycle;

[0008] Based on the voltage disturbance characteristic analysis of the real-time voltage time series data sequence in the rated voltage interval of the converter, the voltage disturbance mode information of the converter is identified;

[0009] According to the embedded voltage control strategy library and voltage stability threshold condition, the initial voltage regulation instruction sequence is generated combined with the voltage disturbance mode information of the converter;

[0010] The voltage change trend of the voltage disturbance mode information of the converter is predicted to obtain the predicted voltage disturbance mode information;

[0011] Based on the predicted voltage disturbance mode information, the initial voltage regulation instruction sequence is optimized for voltage control compensation to generate an optimized voltage regulation instruction sequence;

[0012] According to the optimized voltage regulation instruction sequence, the new energy vehicle converter is adaptively voltage-regulated.

[0013] Preferably, the step of obtaining the real-time voltage time series data sequence of the new energy vehicle converter within the operation cycle comprises:

[0014] Collecting multi-channel voltage sampling data of the converter DC side and AC side;

[0015] The multi-channel voltage sampling data is processed by time series alignment to form a timestamp-synchronized voltage data set;

[0016] The timestamp-synchronized voltage data set is processed by abnormal data filtering to obtain a purified voltage data sequence;

[0017] The purified voltage data sequence is segmented and reorganized according to the operation cycle to generate a real-time voltage time series data sequence.

[0018] Preferably, the step of voltage disturbance characteristic analysis of the real-time voltage time series data sequence based on the rated voltage interval of the converter comprises:

[0019] Extracting the upper threshold and lower threshold of the rated voltage interval of the converter;

[0020] Calculating the deviation value of each data point in the real-time voltage time series data sequence from the rated voltage interval;

[0021] determining the voltage disturbance intensity level distribution according to the deviation value;

[0022] analyzing the spatiotemporal characteristic pattern of the voltage disturbance intensity level distribution to generate the transformer voltage disturbance modal information.

[0023] Preferably, the step of generating the initial voltage regulation instruction sequence according to the embedded voltage control strategy library and the voltage stability threshold condition comprises:

[0024] traversing the historical voltage regulation strategy records in the voltage control strategy library;

[0025] calculating the matching degree between the transformer voltage disturbance modal information and the disturbance characteristics in the historical voltage regulation strategy records;

[0026] selecting the historical voltage regulation strategy records with a matching degree exceeding a set threshold as candidate strategies;

[0027] verifying the stability of the candidate strategies according to the voltage stability threshold condition;

[0028] composing the initial voltage regulation instruction sequence from the verified candidate strategies.

[0029] Preferably, the step of predicting the voltage change trend of the transformer voltage disturbance modal information comprises:

[0030] establishing a prediction model for the voltage disturbance characteristic time series;

[0031] inputting the time series characteristic data in the transformer voltage disturbance modal information;

[0032] outputting the voltage disturbance intensity prediction values at future time points through the prediction model;

[0033] integrating the voltage disturbance intensity prediction values to form the predicted voltage disturbance modal information.

[0034] Preferably, the step of performing voltage control compensation optimization on the initial voltage regulation instruction sequence based on the predicted voltage disturbance modal information comprises:

[0035] performing correlation analysis between the predicted voltage disturbance modal information and the initial voltage regulation instruction sequence;

[0036] identifying the instruction segments in the initial voltage regulation instruction sequence that need to be compensated and adjusted;

[0037] generating compensation control instructions according to the predicted voltage disturbance modal information;

[0038] inserting the compensation control instructions into the corresponding positions of the initial voltage regulation instruction sequence;

[0039] The method comprises the following steps of: performing conflict detection and resolution on the instruction sequence after inserting the compensation control instruction, and generating an optimized voltage regulation instruction sequence.

[0040] Preferably, the step of performing adaptive voltage regulation on the new energy automobile converter according to the optimized voltage regulation instruction sequence comprises the following steps of:

[0041] Analyzing each control instruction in the optimized voltage regulation instruction sequence;

[0042] Converting the control instruction into a driving signal of the converter power device;

[0043] Performing voltage regulation on the converter according to the timing of the driving signal;

[0044] Monitoring actual voltage response data in the voltage regulation process;

[0045] Feeding the actual voltage response data back to the voltage disturbance feature analysis step.

[0046] Preferably, the method further comprises the following steps of:

[0047] Calculating a converter operation stability index based on the actual voltage response data;

[0048] Triggering an update process of the voltage control strategy library when the operation stability index is lower than a preset standard;

[0049] Generating a new voltage regulation strategy record according to the latest operation data;

[0050] Adding the new voltage regulation strategy record to the voltage control strategy library.

[0051] Preferably, the step of triggering the update process of the voltage control strategy library comprises the following steps of:

[0052] Comparing the difference degree of the current operation stability index and the historical operation stability index;

[0053] Starting the strategy library update program when the difference degree exceeds an update threshold;

[0054] Screening a strategy record with poor effect in the voltage control strategy library, and replacing the strategy record with poor effect with the newly generated voltage regulation strategy record;

[0055] Re-calibrating the voltage stability threshold condition after the update is completed.

[0056] Preferably, the method further comprises the following steps of:

[0057] Collecting environmental parameter data around the converter in real time, analyzing the influence weight of the environmental parameter data on voltage stability, and dynamically adjusting the voltage stability threshold condition according to the influence weight.

[0058] Compared with the prior art, the application has the beneficial effects that:

[0059] The adaptive voltage dynamic stability control method of the new energy vehicle converter can provide comprehensive and continuous data basis for subsequent voltage disturbance analysis by acquiring real-time voltage time sequence data sequence in the operation cycle of the converter, can completely capture the dynamic process of voltage change, and avoids the problem of missing disturbance information caused by incomplete data sampling. Based on the voltage disturbance characteristic analysis of the real-time voltage time sequence data sequence in the rated voltage interval of the converter, the voltage disturbance modal information of the converter can be accurately identified, different types of voltage disturbances such as instantaneous pulse, continuous fluctuation and gradual drift can be distinguished, the subsequent regulation strategy can be matched with the disturbance characteristics, the traditional control method of one-size-fits-all regulation mode for various disturbances is changed, and the adaptability of the control strategy to the actual disturbance scene is improved.

[0060] According to the embedded voltage control strategy library and the voltage stability threshold condition, the initial voltage regulation instruction sequence is generated in combination with the identified voltage disturbance modal information of the converter, the diversified control strategy resources are fully utilized, the adaptive basic regulation logic is called for different disturbance modalities, the initial regulation instruction is ensured to have certain rationality and effectiveness, the time loss caused by building the regulation instruction from zero is avoided, and the initial regulation instruction is ensured to meet the basic voltage stability requirement, thereby laying a good foundation for subsequent optimization regulation. The voltage change trend prediction is performed on the voltage disturbance modal information of the converter, the predicted voltage disturbance modal information is obtained, the voltage control is changed from passive response to active prediction, the future change direction and possible disturbance trend of the voltage can be perceived in advance, the limitation of the traditional control method that can only cope with the occurred voltage fluctuation is broken, the basis for formulating regulation measures in advance is provided, the voltage disturbance is effectively avoided from further expanding, and the duration of the voltage unstable state is reduced.

[0061] Based on the predicted voltage disturbance modal information, the initial voltage regulation instruction sequence is subjected to voltage control compensation optimization to generate an optimized voltage regulation instruction sequence, which can correct and improve the initial regulation instruction according to the future voltage change trend, and make up for the possible deficiencies of the initial regulation instruction based on the current disturbance information. For example, when it is predicted that the voltage will have a continuous rising trend, the regulation parameters can be adjusted in advance through compensation optimization to avoid the voltage exceeding the stable interval; if it is predicted that the disturbance will gradually weaken, the regulation strength can be appropriately reduced to prevent new voltage fluctuation caused by excessive regulation, so that the regulation instruction is more accurate and gentle, and the fine level of voltage regulation is improved.

[0062] According to the adaptive voltage regulation instruction sequence, the new energy vehicle converter is self-adaptively voltage-regulated, the dynamic and flexible control of the converter voltage can be realized, the voltage can be kept in the stable interval under different operating conditions, the occurrence of unstable phenomena such as voltage mutation, oscillation and drift is reduced, the converter and subsequent power electronic devices are protected from voltage impact, and the service life of the devices is prolonged. At the same time, stable voltage output can ensure that the electrical equipment is always in the best working condition, improve the energy conversion efficiency, reduce unnecessary energy loss, and indirectly improve the endurance performance of the vehicle. The whole control process does not need manual intervention, and the whole process from data acquisition, disturbance identification, trend prediction to instruction generation and execution can be completed automatically, which is suitable for the complex and changeable operating conditions of new energy vehicles, enhances the automation degree and robustness of the converter voltage control, and provides protection for the safe and efficient operation of the vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The working principle diagram of the adaptive voltage dynamic stability control method of the new energy vehicle converter is described.

[0064] Figure 2 The method flow chart for obtaining the real-time voltage time sequence data sequence of the new energy vehicle converter is described.

[0065] Figure 3 The method flow chart for generating the initial regulation instruction sequence according to the voltage control strategy library and the stable threshold is described. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0067] Please refer to Figure 1The application provides a new energy vehicle converter adaptive voltage dynamic stability control method, which comprises obtaining real-time voltage time series data sequence of the new energy vehicle converter in a running cycle, which contains voltage change information in the running process of the converter. Based on the rated voltage interval of the converter, the real-time voltage time series data sequence is analyzed for voltage disturbance characteristics. The analysis process involves calculating the deviation of voltage data from the rated interval, thereby identifying the converter voltage disturbance mode information, which represents the strength and mode of voltage disturbance. According to the embedded voltage control strategy library and voltage stability threshold condition, combined with the identified converter voltage disturbance mode information, an initial voltage regulation instruction sequence is generated. The initial sequence is generated based on historical strategy matching and stability verification. The voltage change trend of the converter voltage disturbance mode information is predicted, and the future voltage disturbance state is output by using the prediction model to obtain the predicted voltage disturbance mode information. Based on the predicted voltage disturbance mode information, the initial voltage regulation instruction sequence is optimized for voltage control compensation. The compensation process is realized through correlation analysis and instruction adjustment to generate an optimized voltage regulation instruction sequence. According to the optimized voltage regulation instruction sequence, the new energy vehicle converter is adaptively regulated. The adjustment action is completed through the driving signal control power device, while the actual voltage response is monitored and fed back to the analysis step to form a closed loop control.

[0068] Example 1: see Figure 2, the high-precision differential voltage sensor is used to collect the signal at a sampling frequency of megahertz level, and each channel is equipped with an independent signal conditioning circuit to amplify and filter the original signal for preprocessing, so as to eliminate the measurement error caused by high-frequency switching noise. The sampled data is converted into digital quantity by an isolated ADC module, transmitted to the signal processing unit through a parallel data bus for subsequent operation, and the hardware time stamp counter in the processing unit marks the accurate collection time for each sampling point. When the sampling data of different channels has clock drift, the interpolation algorithm is used to resample the time axis of the non-synchronous data points, so that all channel data are unified to a common time reference. The timing alignment process pays special attention to the synchronization of phase-sensitive disturbance signals, such as the start and end time of voltage sag or surge, which need to be accurately matched across channels, and the timestamp-synchronized voltage data set formed thereby provides a space-time consistent input basis for subsequent analysis. The data set is stored in the memory in a ring buffer structure, supporting real-time updating and historical data backtracking in a sliding window manner. The timestamp-synchronized voltage data set needs to be filtered to improve data quality, and the filtering algorithm uses an outlier detection mechanism based on dynamic threshold. The algorithm first calculates the statistical characteristics of the voltage data in the sliding window, including the mean and standard deviation, and then marks the data points deviating from the mean by more than three times the standard deviation as outliers. For the detected outliers, instead of simply removing them, context information is used to identify them, distinguishing between real voltage disturbance events and false anomalies caused by measurement system failures, and only the data points confirmed as false anomalies are removed from the sequence.

[0069] Segmenting and reorganizing the purified voltage data sequence according to the operating period requires clearly defining the operating period boundary of the converter, which usually corresponds to the fundamental period or integer multiple of the switching period of the converter control system. The reorganization algorithm divides the data segments according to the voltage zero-crossing points or specific synchronization signals, and each data segment contains an integer number of complete periods to maintain waveform continuity. The overlapping window technique is used in the segmentation process to ensure that transient events are not fragmented, and appropriate overlap is set between adjacent data segments to capture events across the period boundary. In this way, the real-time voltage time series data sequence generated not only preserves the waveform details but also has period comparability. The sequence is organized in a first-in-first-out queue structure, with the latest data segment dynamically replacing the oldest data segment to achieve rolling update. Voltage disturbance feature analysis based on the real-time voltage time series data sequence first requires extracting the upper and lower threshold values of the converter rated voltage range. These threshold values are obtained from the converter technical specification and can be dynamically adjusted according to the operating state. The rated voltage range is defined as the allowable normal working voltage range, the upper threshold corresponds to the maximum safe operating voltage, and the lower threshold corresponds to the minimum stable working voltage. The threshold setting considers temperature compensation and aging factors to improve adaptability.

[0070] The analysis system periodically checks the threshold accuracy, automatically adjusts the threshold boundary when significant changes in operating conditions are detected, and ensures that the rated voltage interval always reflects the actual withstand capability of the converter. The deviation value of each data point in the real-time voltage time series data sequence from the rated voltage interval is calculated using a relative deviation calculation method. The deviation is defined as the ratio of the difference between the measured voltage and the nearest threshold boundary to the threshold interval. The calculation process distinguishes between two cases of deviation, i.e., positive deviation when the voltage exceeds the upper threshold, negative deviation when the voltage is below the lower threshold, and zero deviation when the voltage is within the rated interval. The deviation value is continuously calculated to form a deviation time series, which quantifies the degree and direction of voltage deviation from the rated state, providing a quantitative basis for disturbance intensity classification. The deviation calculation introduces smoothing processing to avoid numerical jitter caused by noise, and uses a first-order low-pass filter to eliminate the influence of high-frequency fluctuations on evaluation. According to the deviation value, a scientific classification standard is established to determine the voltage disturbance intensity level distribution, which maps the absolute value of the deviation to discrete intensity levels. The classification standard adopts a nonlinear division strategy, with a higher classification density in the small deviation region to distinguish slight disturbances, and a lower classification density in the large deviation region to focus on significant events. Each intensity level corresponds to a different deviation range, and the level classification considers the converter's tolerance and operating requirements, such as setting stricter level boundaries in the critical region. The voltage disturbance intensity level distribution is recorded in the form of a time series, which not only contains the level value but also the duration information, forming a complete disturbance intensity spatiotemporal pattern.

[0071] Analyzing the spatiotemporal characteristic patterns of voltage disturbance intensity level distribution involves multi-dimensional pattern recognition. Time characteristic analysis focuses on disturbance duration, frequency, and periodicity, while spatial characteristic analysis examines the propagation path and correlation of disturbances among multiple channels. The pattern recognition algorithm extracts characteristic parameters such as the starting time, duration, and amplitude variation trend of disturbance events, and classifies similar patterns through clustering analysis. The spatiotemporal characteristic patterns are finally integrated into converter voltage disturbance modal information, which is encapsulated in a structured data object and contains complete descriptions such as disturbance type identification, intensity distribution pattern, and spatiotemporal characteristic vector. When the disturbance modal information is output to the downstream processing module, it is accompanied by a confidence index reflecting the reliability level of the characteristic analysis. The hardware platform selects a microprocessor that meets the functional safety requirements of automotive level, and the software algorithm is optimized to meet real-time constraints. Multiple checkpoints are set during data stream processing to verify the rationality of intermediate results, and once an anomaly is found, the data re-sampling or flow backtracking mechanism is immediately activated. The design of the implementation fully considers the complexity of the new energy vehicle operating environment, including temperature fluctuations, mechanical vibrations, and electromagnetic interference. The voltage disturbance characteristic analysis results are finally encapsulated in a standard communication protocol for subsequent voltage control strategy generation module calls.

[0072] Example 2: see Figure 3, the voltage control strategy library is stored in the form of a relational database in non-volatile memory, and each historical voltage regulation strategy record contains complete context information, such as the disturbance feature vector, the control instruction sequence taken, the voltage response curve after execution, and the effect evaluation index. The traversal process uses an index-based optimization query algorithm. First, according to the current converter operating mode (such as Boost mode or Buck mode), preliminary screening is performed to narrow the search range of the strategy record. The traversal mechanism is designed in a bidirectional scanning mode, searching both forward for similar historical working conditions and backward for continuous control logic to ensure the completeness of the strategy matching. The database index is established according to multiple dimensions such as disturbance type, timestamp, and control target, supporting fast query under composite conditions. The hit frequency and recent use effect of each strategy are recorded during the traversal process. When matching the converter voltage disturbance modal information with the disturbance features in the historical voltage regulation strategy record, a feature mapping relationship needs to be established. The current modal information is parsed into a feature vector form, including disturbance amplitude, duration, rate of change, and other key attributes. The disturbance features in the historical record are also stored in a standardized vector format. The matching degree calculation uses a weighted Euclidean distance algorithm, assigning weight coefficients according to the influence of different features on the control effect, for example, the weight of voltage deviation amplitude is higher than that of duration. The calculation process introduces a fuzzy matching mechanism, allowing feature values to match within a certain tolerance range, avoiding matching failure due to minor fluctuations. The matching degree result is normalized to convert it into a similarity score between 0 and 1, and a matching report is generated to explain the contribution of each feature item.

[0073] Selecting historical voltage regulation strategy records with matching degrees exceeding a set threshold as candidate strategies is a dynamic optimization process. The set threshold is not a fixed value but is adaptively adjusted based on the system's operating state. When the converter is in a critical stable state, the threshold is lowered to expand the range of candidate strategies; under stable operating conditions, the threshold is raised to ensure strategy quality. A tournament selection algorithm is used for candidate strategy selection, choosing the strategies with the highest comprehensive scores from the matching degree records. The novelty of the strategies is also considered, prioritizing records that have been recently validated. The candidate strategy set needs to meet diversity requirements, avoiding all strategies originating from the same historical scenario, and ensuring the ability to cope with different disturbances. The system generates metadata for each candidate strategy, recording its matching degree score, applicable condition confidence level, and other information. Stability verification of candidate strategies based on voltage stability threshold conditions requires the construction of a virtual execution environment. The voltage stability threshold conditions define boundary parameters such as the allowable voltage fluctuation range and maximum rate of change for the converter. The verification process simulates the application effect of candidate strategies using a digital twin model. The simulation model receives the control command sequence of the candidate strategies, predicts the voltage response over several control cycles based on the current converter state, and focuses on monitoring indicators such as voltage overshoot, settling time, and steady-state error. The validation algorithm employs the Monte Carlo method, considering uncertainties such as load fluctuations and temperature changes to evaluate the robustness of the strategy. Validated strategies must satisfy all stability threshold conditions and retain a certain safety margin. Strategies that fail validation are marked, and the reasons for failure are analyzed for optimization of the strategy library.

[0074] The validated candidate strategies, forming the initial voltage regulation command sequence, require temporal integration. The command segments of each candidate strategy are arranged and combined according to the timeline to eliminate temporal overlap and logical conflicts between commands. The integration algorithm employs command interpolation technology to smooth command transitions at different strategy points, ensuring the continuity of the control sequence. After the initial voltage regulation command sequence is generated, priority markers are added, key commands are marked as non-skippable, and auxiliary commands have flexible execution intervals. The sequence format adopts a standard time-command-value pair structure, with accompanying execution condition descriptions and backoff schemes, providing complete input for subsequent compensation optimization. A time series prediction model based on an attention mechanism is used to establish the voltage disturbance feature time series prediction model. The model input layer is designed as a multivariate time window, receiving temporal feature data such as voltage disturbance intensity levels and trends from multiple consecutive sampling times. The network structure includes an encoder-decoder framework; the encoder encodes historical sequences into feature representations, and the decoder recursively predicts future states based on these representations. The attention mechanism enables the model to focus on key time points, especially being more sensitive to feature extraction at moments of sudden disturbance changes. Model parameters are adversarially trained using historical normal operation data and fault case data to enhance their generalization ability. The training process employs a rolling time window validation to prevent overfitting.

[0075] The time series feature data in the input transformer voltage disturbance modal information needs to be processed by feature engineering. The time series features with predictive value are extracted from the modal information, including the moving average, standard deviation, skewness, and other statistical characteristics of the disturbance intensity, as well as the frequency domain features obtained by fast Fourier transform. The feature data is standardized to eliminate the dimension effect while preserving the time dependence. The input interface supports variable-length sequence processing, and the mirror expansion method is used to supplement when the historical data length is insufficient, and importance sampling compression is used when it is too long. The feature data is detected for abnormal values and missing values before being input, ensuring that the prediction model receives complete and valid information. The prediction model outputs the predicted values of the voltage disturbance intensity at multiple future time points using a multi-step prediction strategy. The model output layer is designed as a multi-head structure, with each head responsible for predicting the disturbance intensity at different time scales. The near-term prediction time interval is smaller to ensure accuracy, while the long-term prediction time interval is larger to reduce uncertainty. The prediction values are output in the form of probability distribution, providing the most likely prediction value and confidence interval to reflect the uncertainty of the prediction. The model integrates the results of multiple prediction algorithms and obtains the final prediction value through weighted averaging. The weights are dynamically adjusted based on the performance of each algorithm on the validation set. The prediction result is accompanied by a reliability indicator, which triggers the online learning mechanism of the model when the prediction confidence is below a threshold. The predicted voltage disturbance modal information needs to be reconstructed to form a complete modal description. The predicted values at different time points are combined into a time series, and the disturbance statistical characteristics during the prediction period are calculated based on this. The modal information reconstruction considers the prediction uncertainty and generates modal descriptions for three scenarios: optimistic, pessimistic, and most likely. The predicted voltage disturbance modal information inherits the structured format of the current modal, with the addition of a prediction reliability label and a time effectiveness indicator.

[0076] In embodiment 3, the predicted modal information is parsed into a sequence of feature vectors, including dynamic attributes such as disturbance intensity, trend, and duration, while the initial instruction sequence is mapped into a sequence of control action vectors, reflecting the amplitude and timing of voltage adjustment. The correlation degree calculation introduces a sliding window comparison mechanism to evaluate the correspondence between the predicted event and the control instruction within each time window. The correlation degree value is calculated through a custom similarity function defined as:

[0077]

[0078] where: represents the correlation degree score at time point t, with a value range of [-1, 1], and a value closer to 1 indicates a stronger correlation; is the total number of feature dimensions, corresponding to the number of attributes of the disturbance modal and control instruction; is the weight coefficient of the kth feature, which is pre-set according to the importance of the feature in affecting the control effect; is the standardized value of the predicted voltage disturbance modal information on the kth feature; is the mapping value of the initial voltage regulation instruction sequence on the k-th feature; function is the feature-specific similarity calculation, which adopts the product form for continuous features and the matching indicator function for discrete features. This formula is specifically designed for the correlation analysis of this embodiment, and the symbol design is independent to avoid duplication with other embodiments.

[0079] Identifying the instruction segments in the initial voltage regulation instruction sequence that require compensation adjustment depends on the correlation analysis results. When the correlation score is below the set threshold, the instructions in the corresponding time period are marked as segments to be compensated. The identification algorithm combines the confidence information of the prediction mode, prioritizes the instruction segments corresponding to high-confidence predictions, and avoids unnecessary adjustments based on uncertain predictions. The instruction segment division takes into account the response delay of the control system, ensuring that the compensation window covers the entire life cycle of the disturbance event. The identification result is output as a list of instruction segments, each annotated with start time, end time, and compensation priority. Generating compensation control instructions based on predicted voltage disturbance mode information is a reverse design process, and the goal of the compensation instructions is to offset the negative impact of the predicted disturbance on voltage stability. The generation algorithm is based on the disturbance compensation principle, calculating the required voltage adjustment to counteract the predicted disturbance. Instruction parameters include voltage reference value correction, control gain adjustment, and timing offset. The compensation instruction generation considers the dynamic response characteristics of the converter and introduces feedforward control logic to act in advance. The instruction format remains consistent with the initial instruction sequence to ensure seamless integration. The generation process performs sensitivity analysis to evaluate the robustness of the compensation instructions to uncertainty and generates an alternative instruction set to handle multiple scenarios.

[0080] Inserting compensation control instructions into the corresponding positions of the initial voltage regulation instruction sequence requires precise time alignment. The insertion point is determined based on the timestamp of the predicted disturbance and the system delay. The insertion operation uses an instruction overlap resolution strategy. When new instructions conflict with existing instructions in time, instruction merging or replacement is performed based on priority. The insertion algorithm maintains the timing consistency of the instruction sequence, avoids introducing time jumps, and preserves the logical dependency between instructions. The post-insertion sequence undergoes integrity checks to ensure that there is only one valid instruction at each time point. Conflict detection and resolution processing of the instruction sequence after inserting compensation control instructions is a key step in the optimization process. Conflict detection scans the instruction sequence for time overlaps, logical contradictions, and control target conflicts. The detection algorithm uses constraint satisfaction theory to define mutual exclusion relationships and timing constraints between instructions. Conflict resolution uses a negotiation mechanism to resolve conflicts by adjusting instruction parameters or timing. The resolution process introduces a rollback strategy when conflicts cannot be completely eliminated, selecting conservative control actions. The final optimized voltage regulation instruction sequence is verified to ensure that it meets the stability requirements in typical scenarios.

[0081] The parsing of each control instruction in the optimized voltage regulation instruction sequence involves instruction decoding and parameter extraction. The parser identifies the instruction type, such as voltage setpoint adjustment, PID parameter modification, or protection action triggering. The parsing process verifies the correctness of the instruction syntax and semantics, filters illegal instructions, and converts instruction parameters into internal representation formats for subsequent processing. The parsing module supports stream processing of instruction sequences, enabling low-latency parsing. Converting control instructions into drive signals for power devices in the converter requires hardware-specific mapping. The conversion process generates corresponding PWM waveforms or digital drive signals based on the characteristics of power devices such as IGBT or MOSFET. The conversion algorithm considers dead-time, switching frequency, and drive capability limitations, optimizing signal shape to reduce switching loss and electromagnetic interference. The drive signal generation integrates protection functions, monitors device status in real-time, and prevents overdrive. The signal output is buffered to ensure timing accuracy.

[0082] Voltage regulation of the converter by executing drive signals in a timely manner is triggered by the microcontroller timer. The execution mechanism uses interrupt-driven or DMA transmission to achieve high-precision timing control. The execution process considers control system delays and introduces an early trigger compensation mechanism to ensure that instructions take effect at the correct time point. Regulation actions monitor real-time voltage feedback and dynamically adjust the execution timing to address timing jitter. The execution log records the actual execution timestamp of each instruction for post-analysis. Actual voltage response data during voltage regulation is monitored through high-precision ADC channels, with monitoring points covering key nodes of the converter such as DC bus voltage and AC output voltage. Data collection is synchronized with the control cycle, recording voltage waveforms, ripples, and transient responses in real-time. The monitoring data is filtered to eliminate noise effects and extract effective voltage features. The monitoring module integrates fault detection functions to trigger emergency processing procedures when voltage anomalies occur. Actual voltage response data is fed back to the voltage disturbance feature analysis step to form a closed-loop control. The feedback path is designed as a low-latency communication link to ensure data timeliness. Feedback data is compared with predicted modal information to calculate prediction errors for system calibration. The feedback mechanism supports adaptive learning to continuously improve control performance. Closed-loop control maintains voltage stability by periodic adjustments, adapting to changes in converter operating conditions.

[0083] Embodiment 4: The implementation of embodiment 4 involves calculating the converter operation stability index based on actual voltage response data and triggering the update process of the voltage control strategy library when the index is below the preset standard, which realizes the self-optimization of the control system in a closed-loop form. The calculation of the operation stability index relies on the actual voltage response data fed back from the voltage regulation link, which contains the dynamic behavior record of the converter after applying the optimized voltage control instruction sequence. The index calculation uses a multi-dimensional evaluation method, considering parameters such as voltage deviation, fluctuation frequency, and recovery time. The stability index calculation algorithm is integrated in the firmware of the controller, and the evaluation is performed every 100 milliseconds. The index value is quantized as a score between 0 and 100, with a higher score indicating better stability. The calculation process introduces a weighted average mechanism, with recent data having a higher weight to reflect the current state of the system. The index output is accompanied by a timestamp and a working condition label for historical tracing. When the operation stability index is below the preset standard, the system automatically triggers the update process of the voltage control strategy library. The preset standard is dynamically set according to the type of the converter and the application scenario. The trigger mechanism has a delay confirmation function to avoid false triggering due to transient fluctuations. Before starting the update process, the system checks the resource availability and safety conditions to ensure that the update process does not affect real-time control tasks. The update threshold is set with a hysteresis interval to prevent frequent switching at the critical point. The update command is passed to the management module through an event flag.

[0084] The difference between the current running stability index and the historical running stability index is the key step of the update decision, and the difference calculation uses the sliding window comparison method. The historical index data is stored in a circular buffer, retaining the record of the last 1000 sampling points, and the difference is obtained by calculating the deviation of the current index from the historical mean. For example, the system defines the difference as the absolute difference between the current value and the index value at the same time in the previous period, and then divides by the historical range for normalization. When the difference exceeds 0.15, it is considered that the system performance has changed significantly. The comparison algorithm considers the influence of day and night patterns and load cycles, and introduces a seasonal adjustment factor to avoid false positives due to normal fluctuations. The difference result is compared with the update threshold, and the threshold is dynamically adjusted according to the running experience, initially set to 0.2, and gradually optimized with system learning. When the difference exceeds the update threshold, the strategy library update program starts immediately, and the update program runs in the background task with lower priority than real-time control tasks. The update process first creates a snapshot of the strategy library to prevent interruptions during updating from affecting system operation, and then suspends the strategy query service to ensure data consistency. The update program calls the strategy evaluation module to score all records in the current strategy library based on recent execution success rate and stability contribution. The program logs the update trigger reason and time for maintenance personnel analysis, and the system switches to basic control mode during the update process to ensure minimum functional safety. The multi-criteria filtering algorithm is used to filter the strategy records with poor performance in the voltage control strategy library, and the filtering criteria include strategy usage frequency, recent performance score, and applicable scene matching degree. Strategies with poor performance are defined as scores lower than 60% of the average score or three consecutive execution failures, and the screening process introduces a human confirmation link, and important strategy deletion requires high-level authority. For example, a voltage boost strategy in the strategy library for sudden acceleration scenarios is marked for replacement due to severe voltage overshoot after recent execution, and the screening algorithm uses fuzzy logic to evaluate the potential value of the strategy to avoid deleting rare scene strategies. The screening results generate a list of strategies to be replaced, sorted by urgency, with high-risk strategies being processed first.

[0085] Replacing the ineffective strategy record with the newly generated voltage regulation strategy involves strategy generation and verification, and the new strategy is derived from the current operational data. The strategy generator analyzes recent successful control cases, extracts disturbance features and control instruction sequences, and generates candidate strategies. The new strategy is verified through simulation and short-term real vehicle testing to ensure its effectiveness and safety, and the replacement operation uses a transaction processing mechanism, either all successful or rolled back. For example, the system detects a new type of voltage oscillation pattern, generates a suppression strategy based on current control data, replaces the outdated linear control strategy in the library, and updates the strategy version number after replacement. The strategy library update supports both incremental and full update modes, selected according to the update size. After the update is complete, the voltage stability threshold conditions are recalibrated, taking into account the characteristics of the new strategy library and changes in the operating environment. Threshold calibration is based on statistical learning, analyzing the distribution of recent stability indicators, recalculating the upper and lower limits of the threshold, and introducing robust statistics to avoid the influence of outliers. For example, the new strategy library focuses more on dynamic response, and the allowed fluctuation range in the voltage stability threshold conditions is appropriately relaxed, but the change rate threshold is tightened. The calibrated threshold passes all historical data backtest verification. The calibration results are written to the configuration file, and the old settings are backed up for recovery, and the system records the threshold change history to support auditing. Generating new voltage regulation strategy records based on the latest operational data is a data-driven process, and the new strategy record contains complete context information. Strategy generation uses machine learning algorithms to learn the mapping relationship from successful control sequences, and the generated strategy is formatted to meet library standards. New strategies are attached with metadata such as creation time, applicable conditions and effect prediction, and strategy testing is conducted in a virtual environment to evaluate its generalization ability. The generation process considers strategy diversity to avoid homogenization of strategies in the library, and new strategies preferentially cover the current strategy blank area. Referring to Table 1, a data segment of an operational stability index calculation example is shown to illustrate the relationship between index changes and update triggers.

[0086] Table 1: Operational stability index record

[0087]

[0088] The table data is derived from actual operation records, the voltage deviation represents the average deviation of the actual voltage from the rated value, the fluctuation times refer to the number of threshold fluctuation events per unit time, the recovery time represents the time spent by the voltage from disturbance to stable interval, and the stability index is calculated by weighting these parameters. When the index is continuously lower than the preset standard of 75 points, the system triggers the difference calculation, for example, at the moment of 1024600 milliseconds, the index drops to 65 points, which is 0.28 different from the average level of 90 points in the historical uniform working condition, exceeding the update threshold of 0.2, thereby starting the strategy library update. The execution of the update process depends on the system resource management, and the update task is allocated to a dedicated memory pool and computing resources to avoid conflicts with real-time tasks. After the strategy library is updated, the system performs integration testing to verify the performance of the new strategy in typical scenarios, and after the test is passed, the update state flag is set, and the system resumes normal query service. The entire implementation is embedded in the management module of the converter controller, and a state machine model is used to manage the state transition of the update process, from the idle state to the verification state through verification, backup, and update. The implementation design considers fault tolerance, and when an exception occurs during the update process, it rolls back to the last stable version, and records the fault code to assist in diagnosis. The system generates an update report periodically, summarizes the strategy change and effect evaluation, and supports long-term performance optimization.

[0089] In example 5, real-time acquisition of environmental parameter data around the converter relies on a multi-type sensor network distributed on the converter shell and heat dissipation structure. A PT1000 platinum resistance temperature sensor is attached to the interface between the IGBT module substrate and the heat sink. A humidity sensor measures the dew point change inside the case based on the capacitance principle. A vibration accelerometer is installed at the power inductor magnetic core to monitor mechanical stress. The acquisition system synchronously acquires environmental data at a sampling frequency of 100 Hz. Each parameter channel is equipped with hardware filtering to eliminate high-frequency interference. The data is transmitted to the main controller through the CAN bus, and the timestamp accuracy reaches the microsecond level. For example, when an electric vehicle drives through a tunnel, the external temperature rises sharply, causing the temperature inside the case to rise from 45°C to 68°C in 3 minutes. At this time, the temperature sensor records the gradient change of the power module junction temperature, the humidity sensor detects that the condensation risk level is rising, and the vibration sensor captures the frequency characteristics of the structural deformation caused by thermal expansion.

[0090] The influence weight of environmental parameters on voltage stability is analyzed by establishing a causal relationship model between parameters and voltage fluctuations. The feature importance evaluation based on mutual information entropy is used as the analysis method. The joint probability distribution of environmental parameters and voltage deviation is constructed, and the initial weight is determined by calculating the information entropy ratio of each environmental parameter and the voltage stability index. Time delay compensation is introduced in the weight analysis to consider the transmission delay from environmental changes to voltage response. The influence of high temperature environment on electrolytic capacitor ESR usually appears after a delay of 5-8 minutes. The sliding time window mechanism is used in the analysis process, and the weight coefficient is updated every 30 seconds. Recent environmental mutation events will be given higher weight distribution. The weight result is stored in the form of a vector for the threshold adjustment module to call. The fuzzy reasoning mechanism is used to adjust the voltage stability threshold condition according to the influence weight. The threshold adjuster receives the environmental weight vector and the current threshold baseline as input. The adjustment algorithm sets an influence curve for each environmental parameter. For example, when the temperature weight exceeds 0.7, the voltage fluctuation threshold is relaxed by 15%-20% according to the S-shaped curve, and when the humidity weight dominates, the insulation-related threshold constraint is strengthened by about 6mA. The dynamic adjustment process preserves the safety margin to ensure that threshold changes do not trigger false protection. For example, in a high temperature and high humidity environment, the allowed fluctuation range of the DC bus voltage is adjusted from the standard value of ±5% to ±7%, but the leakage current protection threshold is tightened from 10mA to 6mA. The adjustment result is subjected to a rationality check to prevent adjacent period threshold jumps from exceeding the set limit. If the check fails, the historical optimal threshold parameter is used.

[0091] The synchronization of environmental parameter data collection is ensured by hardware trigger signals, and each sensor uses a unified clock source to reduce time deviation. The collected data is subjected to physical dimension normalization processing, temperature is converted to relative temperature scale, humidity is represented by relative saturation degree, and vibration data is integrated to displacement order. The acquisition module has a self-diagnosis function, which automatically switches to the estimation mode when the sensor fails, and supplements the missing parameters based on historical data interpolation. The data quality index is calculated in real time, and the data reconstruction algorithm is triggered when the signal-to-noise ratio is below the set value to ensure that the analysis module obtains reliable input. The influence weight analysis introduces a cross-validation mechanism, and the weight result is verified for correlation with the real-time voltage stability index. The analysis module maintains a parameter contribution matrix to record the weight distribution of each environmental factor under different working conditions, such as finding that the vibration parameter weight increases to 0.6 under high-speed working conditions, while it is only 0.2 under low-speed working conditions. The weight analysis result is accompanied by a confidence evaluation, and low-confidence weights need to be verified multiple times before being adopted. The analysis algorithm has learning ability.

[0092] The dynamic adjustment of the voltage stability threshold condition adopts a hierarchical adjustment strategy, and the adjustment range is selected according to the degree of environmental change. The threshold condition includes two layers of hard threshold and soft threshold, the hard threshold corresponds to the safety boundary that is not allowed to be exceeded, and the soft threshold is used for early warning and can be dynamically adjusted. Simulation verification is performed before the adjustment instruction is issued, the system behavior under the new threshold is predicted, and the adjustment process records the operation log including the pre-adjustment value, the post-adjustment value and the adjustment basis. The threshold condition management adopts version control, supports quick rollback to the historical version, and the adjustment effect is evaluated through the closed-loop evaluation of the subsequent voltage stability index. The implementation process considers the special operating environment of new energy vehicles, for example, when the vehicle drives from a dry environment to a high-humidity coastal area, the system detects that the humidity continues to exceed 85% RH within a 10-minute period, and the humidity weight increases from 0.3 to 0.8. At this time, the insulation resistance threshold in the voltage stability threshold condition is dynamically adjusted from 500Ω / V to 800Ω / V, and the allowed voltage sag duration is extended from 200ms to 350ms. This process is realized through the closed loop of environmental perception-weight analysis-threshold adjustment, ensuring that the converter remains reliable in harsh environments. The environmental parameter acquisition module adopts a redundant design, and key parameters such as temperature are equipped with main and backup sensors, and data validity is confirmed through a majority voting mechanism. The sensor network supports online calibration, and the calibration period is adjusted adaptively according to the environmental change rate, and the collected data is marked with a quality flag to identify the data reliability. The weight analysis module is deployed in the secure core and runs inside the memory protection unit, preventing abnormal modification. The threshold adjustment action requires security authentication, and key threshold adjustment requires a multi-check process, and the adjustment command is transmitted using rolling code encryption.

[0093] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An adaptive voltage dynamic stabilization control method for a new energy vehicle converter, characterized in that, The method comprises: acquiring real-time voltage time series data sequence of new energy vehicle converter in operation cycle; based on the rated voltage interval of the converter, the real-time voltage time series data sequence is analyzed for voltage disturbance characteristics, and the voltage disturbance mode information of the converter is identified; According to the embedded voltage control strategy library and the voltage stability threshold condition, the initial voltage regulation instruction sequence is generated combined with the voltage disturbance mode information of the converter; The voltage change trend of the voltage disturbance mode information of the converter is predicted to obtain the predicted voltage disturbance mode information; Based on the predicted voltage disturbance mode information, the initial voltage regulation instruction sequence is optimized for voltage control compensation to generate an optimized voltage regulation instruction sequence; According to the optimized voltage regulation instruction sequence, the new energy vehicle converter is adaptively voltage regulated; The step of analyzing the real-time voltage time series data sequence based on the rated voltage interval of the converter includes: extracting the upper limit threshold and the lower limit threshold of the rated voltage interval of the converter; Calculate the deviation value of each data point in the real-time voltage time series data sequence from the rated voltage interval; Determine the voltage disturbance intensity level distribution according to the deviation value; Analyze the time and space characteristics of the voltage disturbance intensity level distribution to generate the voltage disturbance mode information of the converter; The step of optimizing the initial voltage regulation instruction sequence based on the predicted voltage disturbance mode information includes: Correlation analysis of the predicted voltage disturbance mode information and the initial voltage regulation instruction sequence; Identify the instruction segment that needs to be compensated and adjusted in the initial voltage regulation instruction sequence; Generate compensation control instructions according to the predicted voltage disturbance mode information; Insert the compensation control instructions into the corresponding position of the initial voltage regulation instruction sequence; The instruction sequence after inserting the compensation control instructions is subjected to conflict detection and resolution processing to generate an optimized voltage regulation instruction sequence.

2. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 1, characterized in that, The step of acquiring real-time voltage time series data sequence of new energy vehicle converter in operation cycle includes: Collecting multi-channel voltage sampling data of the converter DC side and AC side; Time series alignment processing is performed on the multi-channel voltage sampling data to form a timestamp-synchronized voltage data set; The timestamp-synchronized voltage data set is subjected to abnormal data filtering processing to obtain a purified voltage data sequence; The purified voltage data sequence is segmented and reorganized according to the operation cycle to generate a real-time voltage time series data sequence.

3. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 1, characterized in that, The step of generating an initial voltage regulation instruction sequence according to an embedded voltage control strategy library and a voltage stability threshold condition includes: Traverse the historical voltage regulation strategy records in the voltage control strategy library; Calculate the matching degree of the voltage disturbance mode information of the converter and the disturbance characteristics in the historical voltage regulation strategy records; Select the historical voltage regulation strategy records with a matching degree exceeding a set threshold as candidate strategies; Verify the stability of the candidate strategies according to the voltage stability threshold condition; The verified candidate strategies form the initial voltage regulation instruction sequence.

4. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 3, characterized in that, The step of predicting the voltage change trend of the voltage disturbance mode information of the converter includes: Establish a prediction model for the voltage disturbance characteristic time series; Input the time series feature data in the voltage disturbance mode information of the converter; The prediction model outputs voltage disturbance intensity prediction values at future time points; The voltage disturbance intensity prediction values are integrated to form predicted voltage disturbance mode information.

5. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 1, characterized in that, The steps of adaptive voltage regulation of the new energy vehicle converter according to the optimized voltage regulation instruction sequence include: Analyzing each control instruction in the optimized voltage regulation instruction sequence; Converting the control instruction into a driving signal of the converter power device; Performing voltage regulation on the converter according to the timing of the driving signal; Monitoring the actual voltage response data in the voltage regulation process; Feeding back the actual voltage response data to the voltage disturbance feature analysis step.

6. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 5, characterized in that, The method further includes: Calculating the converter operation stability index based on the actual voltage response data; When the operation stability index is lower than the preset standard, triggering the update process of the voltage control strategy library; Generating a new voltage regulation strategy record according to the latest operation data; Adding the new voltage regulation strategy record to the voltage control strategy library.

7. The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 6, characterized in that, The step of triggering the update process of the voltage control strategy library includes: Comparing the difference between the current operation stability index and the historical operation stability index; When the difference exceeds the update threshold, starting the strategy library update program; Screening the strategy records with poor effects in the voltage control strategy library, and replacing the strategy records with poor effects with the newly generated voltage regulation strategy record; After the update is completed, recalibrating the voltage stability threshold condition. 8.The adaptive voltage dynamic stabilization control method of a new energy vehicle converter according to claim 7, characterized in that, The method further includes: Real-time acquisition of converter surrounding environment parameter data, analysis of the influence weight of the environment parameter data on voltage stability, and dynamic adjustment of the voltage stability threshold condition according to the influence weight.

Images