A method, system, and apparatus for load transient event response for a charging module

Abstract

The application discloses a load transient event response method, system and device for a charging module, and is used for prolonging the service life of a power device of the charging module. The method comprises the following steps: acquiring an output terminal electrical parameter time domain waveform of the charging module, and predicting a short-time output terminal electrical parameter time domain waveform within a preset time window by using a preset recurrent neural network model; judging whether the amplitude of the short-time output terminal electrical parameter time domain waveform exceeds the amplitude range of a preset mutation waveform; if yes, calculating a dynamic change rate characteristic parameter, and constructing a difference matrix based on the dynamic change rate characteristic parameter; determining a transient type based on the difference matrix, and selecting a corresponding basic response action sequence from a preset response action library; adjusting the duty cycle of a switch tube driving parameter of a power conversion unit in the charging module according to the basic response action sequence, and generating a control signal based on the adjusted output terminal electrical parameter time domain waveform by using a preset control algorithm, so as to perform dynamic convergence through the control signal.

Description

Technical Field

[0001] This application relates to the field of charging technology, and in particular to a method, system and apparatus for responding to load transient events in a charging module. Background Technology

[0002] With the large-scale deployment of scenarios such as fast charging for new energy vehicles, high-density power supply for data centers, and energy replenishment for 5G communication base stations, charging modules need to cope with increasingly complex transient load demands. In practical applications, the load current of data center servers, 5G communication equipment, or electric vehicle fast charging piles often exhibits rapid and significant step or pulse-like changes. These transient events can severely impact the output voltage of the charging module. If the response is not timely or the control is inadequate, it will lead to output voltage overshoot, continuous oscillation, or even instability, which will not only affect the reliable operation of downstream load equipment but may also endanger the safety of power devices.

[0003] In existing technologies, a fixed-parameter compensation strategy is typically used to respond to load transient events in charging modules. Specifically, a voltage sensor at the output terminal monitors the voltage deviation in real time. When the voltage deviation exceeds a threshold, a load transient event is triggered. After determining that a load transient event has been triggered, a fixed compensation scheme is directly invoked for rapid adjustment, thereby suppressing the load transient and ensuring that the output voltage of the charging module does not exceed the safe range.

[0004] However, using this fixed parameter compensation strategy makes it difficult for the charging module to adapt to load transient events of different amplitudes and rates of change, which can easily lead to over-compensation or under-compensation, thereby affecting the power supply reliability of the equipment and the safe operating life of the power devices inside the charging module. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method, system, and apparatus for responding to load transient events in a charging module.

[0006] The technical solution provided in this application is described below:

[0007] The first aspect of this application provides a load transient event response method for a charging module, the load transient event response method comprising:

[0008] When a load transient trigger signal of the charging module is detected, the time-domain waveform of the output electrical parameters of the charging module is obtained, and a preset recurrent neural network model is used to predict the short-term time-domain waveform of the output electrical parameters of the charging module within a preset time window based on the time-domain waveform of the output electrical parameters.

[0009] Determine whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the preset amplitude range of the sudden change waveform;

[0010] If so, then the dynamic rate of change characterization parameter is calculated based on the time-domain waveform of the output electrical parameters, and a difference matrix is ​​constructed based on the dynamic rate of change characterization parameter before the load transient trigger signal is triggered and the dynamic rate of change characterization parameter after the load transient trigger signal is triggered. The dynamic rate of change characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters.

[0011] The transient type is determined based on the difference matrix, and a basic response action sequence corresponding to the transient type is selected from a preset response action library;

[0012] The duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence. A preset control algorithm is used to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

[0013] Optionally, constructing a difference matrix based on the dynamic rate of change characterization parameters before and after the load transient trigger signal includes:

[0014] The dynamic rate of change characterization parameter within a first preset time period before the triggering time of the load transient trigger signal is used as the first parameter;

[0015] The dynamic rate of change characterization parameter within a second preset time after the triggering time of the load transient trigger signal is used as the second parameter;

[0016] Subtract the corresponding data points in the first parameter and the second parameter element by element to obtain the difference sequence;

[0017] The difference sequence is segmented in time domain order and then filled into a preset matrix template row by row to obtain the difference matrix.

[0018] Optionally, after constructing the difference matrix based on the dynamic rate of change characterization parameters before and after the load transient trigger signal, the load transient event response method further includes:

[0019] The difference matrix is ​​subjected to singular value decomposition to obtain several singular values ​​and a singular vector corresponding to each singular value, wherein the singular values ​​are used to represent the contribution of the singular vector;

[0020] The largest singular value and the corresponding singular vector are used as the feature signature of the load transient trigger signal;

[0021] The feature signature is matched with the feature signature in the preset load transient signal feature signature library for similarity.

[0022] Optionally, the transient types include load step increase type, load step decrease type, and load unknown disturbance type;

[0023] Determining the transient type based on the difference matrix includes:

[0024] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is higher than the first preset threshold, then the transient type is determined to be the load step increase type.

[0025] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the first preset threshold and higher than the second preset threshold, then the transient type is determined to be the load step reduction type.

[0026] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the second preset threshold, then the feature signature is stored as a new sample in the preset load transient signal feature signature library, and the transient type is determined to be the load unknown disturbance type.

[0027] Optionally, after determining the transient type as the unknown load disturbance type, the load transient event response method further includes:

[0028] The output terminal electrical parameters time-domain waveforms are obtained from the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time, and the output terminal electrical parameters time-domain waveforms between the first preset duration and the second preset duration are divided into several analysis periods according to the target time dimension;

[0029] Calculate the average slope value of the time-domain waveform of the output electrical parameters within each analysis period;

[0030] Determine whether each of the average change slope values ​​is within the preset step-increase slope range;

[0031] If so, the transient type is updated from the unknown load disturbance type to the load step increase type.

[0032] Optionally, after determining whether each of the average change slope values ​​is within a preset step-increase slope range, the load transient event response method further includes:

[0033] If not, then the pre-compensation control parameters are generated by the preset action generation model based on the time-domain waveform of the output electrical parameters within the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time.

[0034] Based on the pre-compensation control parameters, a basic response action sequence is constructed, and the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence.

[0035] Optionally, the preset recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer;

[0036] The step of using a preset recurrent neural network model to predict the short-time output electrical parameter time-domain waveform of the charging module within a future preset time window based on the output electrical parameter time-domain waveform includes:

[0037] The time-domain waveform of the output electrical parameters is input to the input layer, and the time-domain waveform of the output electrical parameters is distributed to the loop layer through the input layer;

[0038] The time-series feature sequence of the output electrical parameters in the time-domain waveform is extracted by the loop layer, and the time-series feature sequence is transmitted to the fully connected output layer.

[0039] The fully connected output layer maps the time-series feature sequence to the predicted electrical parameter values ​​at each discrete time point within a future preset time window.

[0040] The predicted electrical parameters at each discrete time point are reconstructed in chronological order to obtain the time-domain waveform of the short-time output electrical parameters.

[0041] Optionally, mapping the time-series feature sequence to predicted electrical parameters at discrete time points within a future preset time window via the fully connected output layer includes:

[0042] The temporal feature sequence output by the recurrent layer is standardized to generate a standardized feature vector that matches the dimension of the fully connected output layer.

[0043] The standardized feature vector is mapped from the high-dimensional feature space to the preset multi-dimensional output space by a set of linear transformation units in the fully connected output layer, thereby obtaining the predicted values ​​of electrical parameters at each discrete time point within the preset time window.

[0044] Optionally, after driving the power conversion unit to perform dynamic convergence via the control signal, the load transient event response method further includes:

[0045] Collect the actual convergence parameters after convergence is complete;

[0046] The preset recurrent neural network model and the preset response action library are updated collaboratively based on the actual convergence parameters, the transient type, and the basic response action sequence.

[0047] A second aspect of this application provides a load transient event response system for a charging module, the load transient event response system comprising:

[0048] The prediction unit is used to acquire the time-domain waveform of the output electrical parameters of the charging module when a load transient trigger signal of the charging module is detected, and to predict the short-term time-domain waveform of the output electrical parameters of the charging module within a future preset time window based on the time-domain waveform of the output electrical parameters using a preset recurrent neural network model.

[0049] The judgment unit is used to determine whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the amplitude range of the preset sudden change waveform;

[0050] The construction unit is used to calculate the dynamic change rate characterization parameter based on the time-domain waveform of the output electrical parameters if the condition is met, and to construct a difference matrix based on the dynamic change rate characterization parameter before the load transient trigger signal is triggered and the dynamic change rate characterization parameter after the load transient trigger signal is triggered. The dynamic change rate characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters.

[0051] The selection unit is used to determine the transient type based on the difference matrix and select a basic response action sequence corresponding to the transient type from a preset response action library;

[0052] The execution unit is used to adjust the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module according to the pre-compensation action in the basic response action sequence, and to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment using a preset control algorithm, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

[0053] Optionally, the building unit is specifically used for:

[0054] The dynamic rate of change characterization parameter within a first preset time period before the triggering time of the load transient trigger signal is used as the first parameter;

[0055] The dynamic rate of change characterization parameter within a second preset time after the triggering time of the load transient trigger signal is used as the second parameter;

[0056] Subtract the corresponding data points in the first parameter and the second parameter element by element to obtain the difference sequence;

[0057] The difference sequence is segmented in time domain order and then filled into a preset matrix template row by row to obtain the difference matrix.

[0058] Optionally, a matching unit may also be included, specifically for:

[0059] The difference matrix is ​​subjected to singular value decomposition to obtain several singular values ​​and a singular vector corresponding to each singular value, wherein the singular values ​​are used to represent the contribution of the singular vector;

[0060] The largest singular value and the corresponding singular vector are used as the feature signature of the load transient trigger signal;

[0061] The feature signature is matched with the feature signature in the preset load transient signal feature signature library for similarity.

[0062] Optionally, the transient types include load step increase type, load step decrease type, and load unknown disturbance type;

[0063] Determining the transient type based on the difference matrix includes:

[0064] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is higher than the first preset threshold, then the transient type is determined to be the load step increase type.

[0065] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the first preset threshold and higher than the second preset threshold, then the transient type is determined to be the load step reduction type.

[0066] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the second preset threshold, then the feature signature is stored as a new sample in the preset load transient signal feature signature library, and the transient type is determined to be the load unknown disturbance type.

[0067] Optionally, a judgment unit may also be included, specifically used for:

[0068] The output terminal electrical parameters time-domain waveforms are obtained from the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time, and the output terminal electrical parameters time-domain waveforms between the first preset duration and the second preset duration are divided into several analysis periods according to the target time dimension;

[0069] Calculate the average slope value of the time-domain waveform of the output electrical parameters within each analysis period;

[0070] Determine whether each of the average change slope values ​​is within the preset step-increase slope range;

[0071] If so, the transient type is updated from the unknown load disturbance type to the load step increase type.

[0072] Optionally, a generation unit may also be included, specifically for:

[0073] If not, then the pre-compensation control parameters are generated by the preset action generation model based on the time-domain waveform of the output electrical parameters within the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time.

[0074] Based on the pre-compensation control parameters, a basic response action sequence is constructed, and the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence.

[0075] Optionally, the preset recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer;

[0076] The step of using a preset recurrent neural network model to predict the short-time output electrical parameter time-domain waveform of the charging module within a future preset time window based on the output electrical parameter time-domain waveform includes:

[0077] The time-domain waveform of the output electrical parameters is input to the input layer, and the time-domain waveform of the output electrical parameters is distributed to the loop layer through the input layer;

[0078] The time-series feature sequence of the output electrical parameters in the time-domain waveform is extracted by the loop layer, and the time-series feature sequence is transmitted to the fully connected output layer.

[0079] The fully connected output layer maps the time-series feature sequence to the predicted electrical parameter values ​​at each discrete time point within a future preset time window.

[0080] The predicted electrical parameters at each discrete time point are reconstructed in chronological order to obtain the time-domain waveform of the short-time output electrical parameters.

[0081] Optionally, the prediction unit is specifically used for:

[0082] The temporal feature sequence output by the recurrent layer is standardized to generate a standardized feature vector that matches the dimension of the fully connected output layer.

[0083] The standardized feature vector is mapped from the high-dimensional feature space to the preset multi-dimensional output space by a set of linear transformation units in the fully connected output layer, thereby obtaining the predicted values ​​of electrical parameters at each discrete time point within the preset time window.

[0084] Optionally, an update unit is also included, specifically for:

[0085] Collect the actual convergence parameters after convergence is complete;

[0086] The preset recurrent neural network model and the preset response action library are updated collaboratively based on the actual convergence parameters, the transient type, and the basic response action sequence.

[0087] A third aspect of this application provides a load transient event response device for a charging module, the load transient event response device comprising:

[0088] Processor, memory, input / output units, and bus;

[0089] The processor is connected to the memory, the input / output unit, and the bus;

[0090] The memory stores a program that the processor invokes to execute the first aspect and any optional load transient event response method within the first aspect.

[0091] A fourth aspect of this application provides a computer-readable storage medium storing a program that, when executed on a computer, performs the first aspect and any optional load transient event response method of the first aspect.

[0092] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0093] 1. Real-time acquisition of the time-domain waveform of the output electrical parameters of the charging module provides the raw data basis for subsequent prediction and analysis. At the same time, the time-domain waveform of the output electrical parameters within a preset time window is predicted by a preset recurrent neural network model, so as to realize the early perception of the transient development trend of the load and avoid the passive situation of responding only after the transient occurs in the existing technology.

[0094] 2. By judging whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the preset amplitude range of the sudden change waveform, high-risk transient events that need intervention can be accurately screened out, avoiding over-adjustment of normal fluctuations, improving the pertinence and efficiency of the response, and reducing unnecessary parameter disturbances;

[0095] 3. When it is determined that the amplitude range is exceeded, a difference matrix is ​​constructed based on the dynamic change rate characterization parameters before and after the transient trigger. At the same time, the corresponding basic response action sequence is matched from the preset response action library, which solves the defects of not being able to distinguish the transient type and blind compensation, and ensures that the compensation strategy is accurately adapted to the transient characteristics.

[0096] 4. The power conversion unit switching transistor drive parameters are adjusted through pre-compensation action, and then the time-domain waveform of the adjusted output electrical parameters is combined with the control signal generated by the preset control algorithm to drive dynamic convergence, forming a dual guarantee of pre-compensation and dynamic correction. This avoids the rigid limitations of fixed parameter adjustment, effectively smooths out transient impacts, and ensures the power supply reliability of the charging module and the service life of the power devices. Attached Figure Description

[0097] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0098] Figure 1 A schematic flowchart of an embodiment of the load transient event response method provided in this application;

[0099] Figure 2 A flowchart illustrating one implementation of step S101 in the load transient event response method provided in this application;

[0100] Figure 3 A flowchart illustrating one implementation of step S103 in the load transient event response method provided in this application;

[0101] Figure 4 A flowchart illustrating an optional embodiment of the load transient event response method provided in this application after step S103;

[0102] Figure 5 A flowchart illustrating one implementation of step S104 in the load transient event response method provided in this application;

[0103] Figure 6 A flowchart illustrating an optional embodiment of the load transient event response method provided in this application after step S1043;

[0104] Figure 7 A schematic diagram of an embodiment of the load transient event response system provided in this application;

[0105] Figure 8 A schematic diagram of an embodiment of the load transient event response device provided in this application. Detailed Implementation

[0106] In this embodiment, the execution entity of the load transient event response method for the charging module is not limited to a specific type of charging control device, data processing terminal, or power regulation system. The load transient event response method can be executed by any hardware, software, or a combination of hardware and software capable of monitoring load transient signals, acquiring and processing electrical parameters in the time domain, calculating recurrent neural network models, and controlling power conversion units. Examples include a dedicated local controller for the charging module, an industrial-grade embedded computing unit, a site-level intelligent charging operation and maintenance server, an edge computing node, and a distributed charging management platform with AI prediction and control functions.

[0107] The load transient event response method for the charging module in this embodiment can also be implemented through the embedded control program built into the charging module, the software module of the charging station operation and maintenance system, or through the execution unit that controls the charging system via an industrial communication interface. Regardless of whether the specific execution entity is a single local processing and control device directly connected to the charging module, multiple parallel cooperating functional units, or a scalable distributed load transient response management system, this method can be operated according to the step sequence and logic in the embodiments below.

[0108] Based on this, the load transient event response method for charging modules in this application is particularly suitable for scenarios with frequent load fluctuations, severe transient impacts, and extremely high requirements for charging stability and safety, including new energy vehicle supercharging stations, industrial electric engineering machinery charging stations, charging modules for distributed energy storage systems, and park charging clusters with multiple devices sharing power supply.

[0109] To enable those skilled in the art to clearly understand the core technical terms involved in the load transient event response method for charging modules in this application, the key terms are defined as follows:

[0110] Load Transient Trigger Signal: This refers to the trigger signal used to identify the occurrence of transient load changes in the charging module. It is generated after real-time detection of the sudden change trend or sudden change threshold of the electrical parameters at the output terminal of the charging module, and is mainly used as the trigger condition for starting the response method of this application.

[0111] Output Electrical Parameter Time-Domain Waveform: This refers to waveform data formed with time as the horizontal axis and the instantaneous values ​​of the output electrical parameters of the charging module as the vertical axis. It can intuitively reflect the dynamic change process of electrical parameters over time, including steady-state fluctuations, abrupt changes, and decay characteristics.

[0112] Preset Amplitude Range of Mutant Waveform: This refers to the amplitude boundary range of the normal fluctuation of the time-domain waveform of the output electrical parameters of the charging module. If it exceeds this range, it is determined that there is a load transient change that requires intervention. It is mainly used as a benchmark to judge whether there is a serious transient risk in the short-term predicted waveform and to screen out high-risk transient events that need further processing.

[0113] Pre-compensation control parameters: These are control parameters used to compensate for transient load fluctuations in advance. They are generated by a preset action generation model for unknown transient disturbances and can be directly used to construct pre-compensation actions in the basic response action sequence.

[0114] Please see Figure 1 This application first provides an embodiment of a load transient event response method for a charging module, the embodiment including:

[0115] S101. When the load transient trigger signal of the charging module is detected, the output electrical parameters time-domain waveform of the charging module is obtained, and a preset recurrent neural network model is used to predict the short-term output electrical parameters time-domain waveform of the charging module within a preset time window based on the output electrical parameters time-domain waveform.

[0116] In this embodiment, the operating status of the charging module will be continuously monitored. When a load transient trigger signal is detected, such as a sudden increase in load current or fluctuation in output voltage, which are signals that indicate a sudden change in load status, the time-domain waveform of the output electrical parameters will be immediately acquired to obtain the time-domain waveform data of the electrical parameters of the charging module output changing over time.

[0117] Simultaneously, a preset recurrent neural network model is invoked, and the time-domain waveform of the collected output electrical parameters is used as the model input. Based on the time-series characteristics of the output electrical parameters time-domain waveform collected in real time and through historical training, the preset recurrent neural network model predicts the short-term output electrical parameters time-domain waveform of the charging module within a preset time window in the future. The preset time window can be 10ms, 20ms, or 30ms, and the specific setting depends on the actual application scenario. No limitation is made here.

[0118] See Figure 2 The following is an implementation of step S101, in which the pre-defined recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer. This implementation includes:

[0119] S1011. Input the time-domain waveform of the output electrical parameters to the input layer, and distribute the time-domain waveform of the output electrical parameters to the loop layer through the input layer;

[0120] In this embodiment, the hierarchical structure of the pre-defined recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer. Each layer has different signal transmission logic. The input layer is responsible for data reception and format adaptation. Through standardized distribution logic, it ensures that the original waveform data can be accurately and completely transmitted to the recurrent layer.

[0121] When a load transient trigger signal is detected and the time-domain waveform of the output electrical parameters is acquired, the time-domain waveform of the output electrical parameters is transmitted as an input signal to the input layer of the preset recurrent neural network model. After receiving the signal, the input layer performs preliminary format adaptation processing on the time-domain waveform of the output electrical parameters according to the distribution rules built into the preset recurrent neural network model to make it conform to the input requirements of the recurrent layer. Then, the adapted time-domain waveform of the output electrical parameters is completely distributed to the recurrent layer.

[0122] S1012. Extract the timing feature sequence from the time-domain waveform of the output electrical parameters through the loop layer, and transmit the timing feature sequence to the fully connected output layer;

[0123] The recurrent layer is a module in a recurrent neural network that processes time-series data. It can use its own memory characteristics to mine the temporal correlations in the data and transform the original waveform data into a more representative temporal feature sequence, thus realizing the transformation from raw data to feature data.

[0124] After receiving the time-domain waveforms of the output electrical parameters distributed by the input layer, the recurrent layer analyzes the changing patterns of the waveform data time-by-time, capturing the correlation between output electrical parameters at adjacent time points. For example, it extracts features such as the rise rate, fall rate, and amplitude variation trend of the voltage waveform at different time points. Simultaneously, it integrates these scattered local features into an ordered time-series feature sequence. After completing the extraction of the time-series feature sequence, the recurrent layer transmits the complete sequence to the fully connected output layer.

[0125] S1013. Standardize the temporal feature sequence output by the recurrent layer to generate a standardized feature vector that matches the dimension of the fully connected output layer.

[0126] Because of the inconsistency in feature scales in time-series feature sequences, in order to ensure that the fully connected output layer treats each feature more fairly, avoids a feature of excessive magnitude dominating the prediction result, and ensures the accuracy of the prediction, it is necessary to first obtain the time-series feature sequence output by the recurrent layer and analyze the data distribution characteristics of the time-series feature sequence, including the value range and dispersion of the feature values.

[0127] Subsequently, a standardization method is used to uniformly calibrate each feature value in the time-series feature sequence to eliminate the weight imbalance problem caused by excessive differences in the value ranges of different features. For example, voltage change rate features and current change rate features of different orders of magnitude are calibrated to the same numerical range. At the same time, a set of standardized feature vectors with fixed dimensions is generated. The dimensions of these standardized feature vectors perfectly match the input dimensions of the fully connected output layer, ensuring that they can be successfully input into the fully connected output layer for subsequent processing.

[0128] S1014. The standardized feature vector is mapped from the high-dimensional feature space to the preset multi-dimensional output space through a set of linear transformation units in the fully connected output layer, thereby obtaining the predicted values ​​of electrical parameters at each discrete time point within the preset time window.

[0129] The linear transformation unit in the fully connected output layer has fixed weight parameters and built-in mapping rules from the high-dimensional feature space to the multi-dimensional output space. In this way, the mapping transformation from the feature space to the prediction space can be realized through linear transformation. By using the weight parameters, the temporal features in the extracted standardized feature vector are associated with the future electrical parameter values, thereby realizing accurate prediction of electrical parameters at future discrete time points.

[0130] When the standardized feature vector is input into the fully connected output layer, the linear transformation unit in the fully connected output layer performs linear operations on the standardized feature vector, mapping the high-dimensional standardized feature vector to a multi-dimensional output space. Each dimension of this multi-dimensional output space corresponds to a discrete time point within a preset future time window. Through mapping operations, the predicted electrical parameters corresponding to each discrete time point can be obtained. For example, every 1 ms within the next 10 ms corresponds to a voltage prediction value, and finally, discrete electrical parameter prediction values ​​are obtained.

[0131] S1015. Reconstruct the predicted electrical parameters at each discrete time point in chronological order to obtain the time-domain waveform of the short-time output electrical parameters.

[0132] In order to restore the discrete electrical parameter predictions to continuous time-domain waveforms and intuitively present the evolution of future electrical parameters, it is necessary to first clarify the specific discrete time coordinates corresponding to each electrical parameter prediction value. For example, prediction value 1 corresponds to the time 1ms in the future, and prediction value 2 corresponds to the time 2ms in the future.

[0133] Subsequently, according to the order of discrete time coordinates, all discrete electrical parameter prediction values ​​are sorted and organized to ensure that the order of electrical parameter prediction values ​​is consistent with the actual time elapsed order. Based on the sorted electrical parameter prediction values, according to the actual reconstruction rule requirements, the electrical parameter prediction values ​​are connected to form a continuous short-time output electrical parameter time domain waveform. This short-time output electrical parameter time domain waveform fully presents the change trend of electrical parameters within the future preset time window.

[0134] S102. Determine whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the preset amplitude range of the sudden change waveform;

[0135] After predicting the short-time output electrical parameter time-domain waveform, the amplitude of the short-time output electrical parameter time-domain waveform is extracted time by time. The extracted amplitude at each time is compared with the amplitude range of the preset sudden change waveform to determine whether there is an amplitude point in the short-time output electrical parameter time-domain waveform that exceeds the amplitude range of the preset sudden change waveform.

[0136] Furthermore, the amplitude range of the preset sudden change waveform is a range of amplitude of the sudden change waveform that is pre-set based on the rated operating parameters of the charging module, the load safety operation requirements, and other indicators. This range clarifies the limit of electrical parameter fluctuations that the charging module can stably withstand. For example, the amplitude range of the output voltage sudden change is set to ±5% of the rated voltage, and the amplitude range of the output current sudden change is set to ±10% of the rated current.

[0137] If the amplitude of the short-time output electrical parameter time-domain waveform exceeds the preset amplitude range of the sudden change waveform, then step S103 is executed; if the amplitude of the short-time output electrical parameter time-domain waveform does not exceed the preset amplitude range of the sudden change waveform, then the existing operating state is maintained, and there is no need to start the response process.

[0138] S103. Calculate the dynamic change rate characterization parameter based on the time-domain waveform of the output electrical parameters, and construct a difference matrix based on the dynamic change rate characterization parameter before the load transient trigger signal and the dynamic change rate characterization parameter after the load transient trigger signal. The dynamic change rate characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters.

[0139] When it is determined that the amplitude of the short-term output electrical parameter time-domain waveform exceeds the amplitude range of the preset sudden change waveform, based on the collected output electrical parameter time-domain waveform of the charging module, the dynamic change rate characterization parameter representing the instantaneous dynamic evolution trend of the output electrical parameter time-domain waveform is obtained by calculating the amount and rate of change of the electrical parameter at adjacent time moments. The dynamic change rate characterization parameter can accurately reflect the fluctuation intensity and change direction of the output electrical parameter over time.

[0140] Subsequently, dynamic rate of change parameters for a period of time before the transient trigger signal and dynamic rate of change parameters for the same period of time after the trigger were selected to construct a difference matrix. Significant differences exist in the dynamic rate of change parameters before and after the transient trigger. By constructing the difference matrix, the impact of transient triggering on the trend of electrical parameter changes can be comprehensively reflected, providing data support for subsequent identification of transient types.

[0141] See Figure 3 The following is an implementation of step S103, which includes:

[0142] S1031, The dynamic change rate characterization parameter within a first preset time period before the triggering time of the load transient trigger signal is used as the first parameter;

[0143] In this embodiment, the first preset duration is determined by combining the actual working conditions such as the response speed of the charging module and the evolution cycle of transient events. This first preset duration can fully cover the stable evolution state of electrical parameters before transient triggering, ensuring that the collected parameters can reflect the normal dynamic trend before triggering.

[0144] Subsequently, the trigger time of the load transient trigger signal is determined. Using this trigger time as a time reference point, a first preset time range is traced back. For example, if the first preset time range is 5ms, then all dynamic change rate characterization parameters within 5ms before the trigger time are extracted from the time reference point and defined as the first parameter in this sequence of dynamic change rate characterization parameters. By locking parameters for a specific time before the trigger, the baseline dynamic characteristics before the transient event occur are obtained, providing a basis for subsequent comparisons with parameters after the trigger, ensuring the relevance and effectiveness of the comparison.

[0145] S1032, The dynamic change rate characterization parameter within the second preset time after the triggering time of the load transient trigger signal is used as the second parameter;

[0146] In this embodiment, the triggering time is used as the time reference point, and the time range of the second preset duration is traced back. For example, if the second preset duration is 5ms, then all dynamic change rate characterization parameters within 5ms after the triggering time are intercepted by tracing back from the time reference point, and the dynamic change rate characterization parameter sequence is defined as the second parameter.

[0147] It should be noted that in practical applications, the first and second preset durations are generally kept consistent. However, if the transient evolution speed is fast, adjustments can be made according to actual needs while ensuring comparability in the time dimension.

[0148] S1033. Subtract the corresponding data points in the first parameter and the second parameter element by element to obtain the difference sequence;

[0149] This embodiment quantifies the difference in the dynamic change rate of electrical parameters before and after triggering by subtracting elements one by one. Each difference data point can intuitively reflect the degree of influence of the trigger on the dynamic change trend at the corresponding moment, thus realizing the visualization of the difference characteristics.

[0150] Specifically, data alignment is performed on the first and second parameters to ensure a one-to-one correspondence between the data points in the two parameter sequences in the time domain. That is, the parameter at time n in the first parameter and the parameter at time n in the second parameter correspond to the dynamic change rate at the same relative time point before and after the trigger, respectively. After the correspondence is established, each set of data points corresponding to the first and second parameters is selected sequentially according to the time domain order, and element-wise subtraction is performed. Then, all the results obtained from the element-wise subtraction are arranged in the original time domain order to form an ordered difference sequence.

[0151] S1034. Divide the difference sequence into segments according to the time domain order, and fill them into the preset matrix template row by row to obtain the difference matrix.

[0152] This embodiment transforms a one-dimensional difference sequence into a two-dimensional matrix structure through matrix processing, making the temporal distribution characteristics and overall variation patterns of the difference data clearer, while also adapting to the requirements of subsequent feature extraction and data format.

[0153] Specifically, a fixed-dimensional matrix template is preset according to actual needs, such as a 3x5 matrix template. The specific dimensions can be determined based on the actual data volume and recognition requirements. Then, following the temporal order, the difference sequence is divided into several segments matching the number of rows in the matrix template. Each segment contains several consecutive difference data points. For example, a difference sequence containing 15 data points is divided into 3 segments of 5 data points each, following the temporal order. Simultaneously with segmentation, the differences from each segment are sequentially filled into the preset matrix template to obtain the difference matrix.

[0154] See Figure 4 Following step S103, an optional implementation method is also provided, which includes:

[0155] S301. Perform singular value decomposition on the difference matrix to obtain several singular values ​​and a singular vector corresponding to each singular value, where the singular values ​​are used to represent the contribution of the singular vector.

[0156] After constructing the difference matrix, singular value decomposition is performed on the difference matrix. Singular value decomposition has the ability to extract the core features of the difference matrix and can extract the main feature components and secondary redundant information from the complex difference matrix.

[0157] During the decomposition process, the difference matrix is ​​broken down into several singular values ​​and a singular vector corresponding to each singular value. Each singular value corresponds to the degree of contribution of its singular vector to the overall features of the difference matrix. The larger the value of the singular value, the more information of the difference matrix is ​​contained in the corresponding singular vector.

[0158] S302. Use the largest singular value and the corresponding singular vector as the feature signature of the load transient trigger signal;

[0159] Since the singular vector corresponding to the largest singular value is the feature component with the highest contribution in the difference matrix, it can most accurately reflect the core difference in the trend of electrical parameter changes before and after transient triggering. Therefore, it is necessary to sort all the singular values ​​obtained by decomposition first, arrange them in descending order of value, and select the singular value with the largest value.

[0160] Subsequently, the singular vector corresponding to the largest singular value is located, and the largest singular value and its corresponding singular vector are combined to form a set of feature data. This set of feature data is then defined as the feature signature of the load transient trigger signal. This feature signature serves as a unique identifier to distinguish different transient types and centrally carries the core transient feature information in the difference matrix.

[0161] S303. Perform similarity matching between the feature signature and the feature signature in the preset load transient signal feature signature library.

[0162] In this embodiment, a preset load transient signal signature library stores signatures corresponding to various known types of load transient events, with each signature associated with a specific transient type. The signatures for different types of transient events are unique. The signature of the current load transient trigger signal is compared with each standard signature in the signature library to calculate similarity, resulting in a similarity value between the current signature and each standard signature. This similarity value quantifies the degree of agreement between the current signature and the standard signatures.

[0163] S104. Determine the transient type based on the difference matrix, and select the basic response action sequence corresponding to the transient type from the preset response action library;

[0164] Since different types of transient events have unique characteristics of the difference matrix, the transient type can be accurately identified through feature matching. Therefore, in this embodiment, it is necessary to first determine the specific type of the current transient signal by using the characteristics of the difference matrix.

[0165] While determining the transient type, the corresponding basic response action sequence is selected from the preset response action library according to the determined transient type. The preset response action library has a pre-configured basic response action sequence that matches each transient type. For example, when the transient type is "load surge type", the basic response action sequence corresponding to "load surge type" in the preset response action library is "increase the duty cycle of the power conversion unit switch".

[0166] See Figure 5The following is an implementation of step S104, in which the transient types include load step increase type, load step decrease type, and load unknown disturbance type. The implementation includes:

[0167] S1041. If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is higher than the first preset threshold, then the transient type is determined to be the load step increase type.

[0168] In this embodiment, the characteristic of the load step increase type can be a sudden increase in the rate of change of current or a slight decrease in the rate of change of voltage. The corresponding feature signature has a high similarity with the feature signature in the preset load transient signal feature signature library. Other types of interference can be eliminated by high threshold screening to ensure the uniqueness and accuracy of the judgment result.

[0169] Furthermore, the first preset threshold is set based on the historical matching data of the feature signatures corresponding to "load step increase type" in the preset load transient signal feature signature library, for example, 90%, mainly to ensure that only cases with high feature matching are judged as "load step increase type".

[0170] In practical applications, the maximum similarity between the current feature signature and the feature signature in the preset load transient signal feature signature library is compared with a first preset threshold. If the maximum similarity is higher than the first preset threshold, the current load transient type is directly determined to be the load step increase type. For example, if the maximum similarity is 95% and the first preset threshold is 90%, and 95% > 90%, then the current transient type is the load step increase type.

[0171] S1042. If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the first preset threshold and higher than the second preset threshold, then the transient type is determined to be the load step reduction type.

[0172] In this embodiment, the characteristic of the load step reduction type can be that the load suddenly decreases, causing the current to drop sharply. The similarity range between the corresponding feature signature and the feature signature in the preset load transient signal feature signature library has a clear boundary with other types.

[0173] Furthermore, the second preset threshold is also set based on the historical matching data of the feature signature corresponding to "load step increase type" in the preset load transient signal feature signature library, and the preset second preset threshold is usually lower than the first preset threshold, mainly to ensure that only when the similarity is in this range will it be judged as "load step decrease type".

[0174] In practical applications, the maximum similarity between the current feature signature and the feature signatures in the preset load transient signal feature signature library is compared with the first preset threshold and the second preset threshold. If the maximum similarity is lower than the first preset threshold but higher than the second preset threshold, the current load transient type is directly determined to be the load step reduction type. For example, if the maximum similarity is 85%, the first preset threshold is 90%, the second preset threshold is 70%, and 70% < 85% < 90%, then the current transient type is the load step reduction type.

[0175] S1043. If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the second preset threshold, then the feature signature is stored as a new sample in the preset load transient signal feature signature library, and the transient type is determined to be load unknown disturbance type.

[0176] In this embodiment, the unknown load disturbance type is a similarity level below the second preset threshold, which indicates that the current transient has unique dynamic characteristics that have not been recorded and belongs to the unknown disturbance.

[0177] In practical applications, the maximum similarity between the current feature signature and the feature signatures in the preset load transient signal feature signature library is compared with a second preset threshold. If the maximum similarity is lower than the second preset threshold, the feature signature is stored as a new sample in the preset load transient signal feature signature library, and the current load transient type is determined to be the load unknown disturbance type. For example, if the maximum similarity is 60% and the second preset threshold is 70%, and 70% > 60%, then the current transient type is the load unknown disturbance type.

[0178] See Figure 6 In step S1043, after the transient type is determined to be an unknown load disturbance, an optional implementation method is also provided, which includes:

[0179] S501. Obtain the time-domain waveform of the output electrical parameters within a first preset time period before the triggering time of the load transient trigger signal to a second preset time period after the triggering time, and divide the time-domain waveform of the output electrical parameters between the first preset time period and the second preset time period into several analysis periods according to the target time dimension;

[0180] After determining the transient type as an unknown load disturbance, the preset duration before and after the transient trigger ensures that the acquired waveform data can fully reflect the evolution process of the unknown disturbance. Furthermore, time-domain segmentation allows for a more detailed analysis of the waveform change characteristics at different time periods, providing more refined data support for subsequent judgments.

[0181] The time range from the first preset duration before the load transient trigger signal to the second preset duration after the trigger moment can completely cover the change process of the main electrical parameters before and after the transient trigger. From the collected time-domain waveforms of the output electrical parameters of the charging module, the output electrical parameter time-domain waveforms within this time range are extracted. Based on the target time dimension, the extracted output electrical parameter time-domain waveforms within this time range are evenly divided into several continuous and non-overlapping analysis periods. For example, if the target time dimension is in 1ms units, and the output electrical parameter time-domain waveform within this time range is 10ms, then 10ms is divided into 10 analysis periods of 1ms each.

[0182] S502. Calculate the average slope value of the time-domain waveform of the output electrical parameters in each analysis period.

[0183] The average slope value can reflect the overall trend of the output electrical parameter time domain waveform within a specific time period. For example, the average speed and direction of the output electrical parameter time domain waveform change with time within the corresponding time period, and different types of transient events correspond to different slope value distribution patterns.

[0184] For each defined analysis period, all data points within that period are extracted, and the corresponding time coordinate and electrical parameter value for each data point are determined. Then, based on the difference in electrical parameter values ​​between the first and last data points within each period, and the time interval between the two data points, the overall trend of electrical parameter variation within that period is calculated, yielding the average slope value. This process of calculating the average slope value for all analysis periods is repeated sequentially, forming a sequence of average slope values ​​corresponding one-to-one with each analysis period.

[0185] S503. Determine whether each average change slope value is within the preset step-increase slope range.

[0186] After calculating the average slope value for each analysis period, the average slope value for each analysis period is compared one by one with the preset step-increase slope range to determine whether the average slope values ​​for all analysis periods are within the preset step-increase slope range. The preset step-increase slope range mainly defines the range that the average slope value should meet when the load increases stepwise. For example, the preset step-increase slope range can be set to the range of 0.5V / ms to 2V / ms. In this case, the average slope value of analysis period A is 1V / ms. 1V / ms falls within the range of 0.5V / ms to 2V / ms, which means that the average slope value of analysis period A is within the preset step-increase slope range.

[0187] If the average slope of any analysis period is not within the preset step-increase slope range, then proceed to step S505; if the average slope of each analysis period is within the preset step-increase slope range, then proceed to step S504.

[0188] S504. Update the transient type from unknown load disturbance type to load step increase type.

[0189] When the average slope value for all analysis periods falls within the preset step-increase slope range, it indicates that the waveform characteristics of the unknown load disturbance type match those of the step-increase load type, and the previous determination of the unknown load disturbance type was due to insufficient feature matching. Subsequently, the original "unknown load disturbance type" was updated to "step-increase load type," and the corresponding feature records were updated simultaneously and incorporated into the feature set of the step-increase load type, providing richer samples for the identification of similar events in the future. In this way, type updating can correct previous misjudgments and ensure the accuracy of transient type identification.

[0190] S505. Based on the time-domain waveform of the output electrical parameters within the first preset duration before the triggering time of the load transient triggering signal to the second preset duration after the triggering time, the pre-compensated control parameters are generated by the preset action generation model.

[0191] When it is determined that the average slope of any analysis period is not within the preset step-increase slope range, it indicates that the unknown disturbance does not belong to the load step-increase type. In this case, the time-domain waveform of the output electrical parameters within the first preset duration before the load transient trigger signal to the second preset duration after the trigger time needs to be used as input to the preset action generation model. Based on the input time-domain waveform of the output electrical parameters within the first preset duration before the trigger time to the second preset duration after the trigger time, the preset action generation model analyzes the intensity, rate of change, and duration of the unknown disturbance. According to the compensation strategy built into the preset action generation model and the operating constraints of the charging module, it generates pre-compensation control parameters adapted to the type of unknown load disturbance.

[0192] S506. Construct a basic response action sequence based on the pre-compensation control parameters, and perform the step of adjusting the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module according to the pre-compensation action in the basic response action sequence.

[0193] Based on the pre-compensated control parameters generated by the preset action generation model, a basic response action sequence is constructed. This sequence can accurately adapt to the characteristics of unknown disturbances, quickly suppress disturbances by adjusting the duty cycle, and achieve stable convergence by combining closed-loop control.

[0194] In practical applications, based on the pre-compensation control parameters and the characteristics and response logic of the power conversion unit of the charging module, a basic response action sequence is constructed. This basic response action sequence clarifies the execution order, execution duration, and control parameter adjustment range of the pre-compensation actions. Subsequently, based on the constructed basic response action sequence, the duty cycle of the switching transistor drive parameters in the power conversion unit of the charging module is adjusted. The steps for adjusting the duty cycle are similar to those in step S105 above and will not be repeated here.

[0195] S105. Adjust the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module according to the pre-compensation action in the basic response action sequence, and use a preset control algorithm to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

[0196] This embodiment quickly suppresses transient trends through pre-compensation actions, and then continuously optimizes the control signal based on actual waveform feedback using a closed-loop control algorithm to achieve dynamic correction, ensuring that the charging module quickly returns to a stable operating state.

[0197] Specifically, after selecting the corresponding basic response action sequence, the duty cycle of the switching transistor drive parameters in the power conversion unit of the charging module is adjusted according to the extracted pre-compensation actions in the basic response action sequence. For example, the output power of the power conversion unit is changed by increasing or decreasing the duty cycle, thereby achieving initial suppression of transient trends. Subsequently, the time-domain waveform of the output electrical parameters of the charging module after adjusting the duty cycle is acquired in real time, and the time-domain waveform of the output electrical parameters is compared with the target steady-state waveform to calculate the deviation between the two. Based on the deviation data, an optimized control signal is generated through a preset control algorithm, and this control signal is input to the power conversion unit to drive the switching transistor to work according to the new control logic, so that the output electrical parameters of the charging module gradually and dynamically converge to the target steady-state value. The preset control algorithm here can be a PID control algorithm or a model control algorithm.

[0198] In this embodiment, the time-domain waveform of the output electrical parameters of the charging module is acquired in real time, providing a raw data foundation for subsequent prediction and analysis. Simultaneously, a preset recurrent neural network model predicts the short-term output electrical parameter time-domain waveform within a preset time window, enabling early perception of load transient trends and avoiding the passive situation of responding only after a transient occurs, as is common in existing technologies. By determining whether the amplitude of the short-term output electrical parameter time-domain waveform exceeds the amplitude range of a preset abrupt change waveform, high-risk transient events requiring intervention can be accurately identified, avoiding over-adjustment of normal fluctuations, improving the targeting and efficiency of the response, and reducing unnecessary parameter disturbances. When it is determined... When the amplitude exceeds the range, a difference matrix is ​​constructed based on the dynamic rate of change characterization parameters before and after the transient trigger. At the same time, the corresponding basic response action sequence is matched from the preset response action library, which solves the defects of not being able to distinguish the transient type and blind compensation, and ensures that the compensation strategy is accurately adapted to the transient characteristics. The power conversion unit switching transistor drive parameters are adjusted by pre-compensation action, and then the control signal is generated by the preset control algorithm to drive dynamic convergence in combination with the time domain waveform of the adjusted output electrical parameters. This forms a dual guarantee of pre-compensation and dynamic correction, avoiding the rigid limitations of fixed parameter adjustment, effectively smoothing transient impacts, and ensuring the power supply reliability of the charging module and the service life of power devices.

[0199] In an optional embodiment, after the power conversion unit is driven to perform dynamic convergence by the control signal, the actual convergence parameters after convergence can be collected, and the preset recurrent neural network model and the preset response action library can be updated collaboratively based on the actual convergence parameters, transient type and basic response action sequence after the collection is completed.

[0200] After the power conversion unit is driven by the control signal to complete dynamic convergence, the actual convergence parameters of the output terminal of the charging module are collected. These actual convergence parameters include the stable output voltage, current, convergence time, and fluctuation amplitude.

[0201] After data collection, the actual convergence parameters are correlated and integrated with the transient type and corresponding basic response action sequence to form a complete set of updated sample data. Based on this updated sample data, the preset recurrent neural network model is optimized and adjusted. This is mainly done by inputting the waveform features corresponding to the actual convergence parameters back into the preset recurrent neural network model, fine-tuning the weight parameters of the preset recurrent neural network model, and improving the prediction accuracy of the preset recurrent neural network model for similar transient events. Simultaneously, the response effect of the basic response action sequence is evaluated in conjunction with the actual convergence parameters. Based on the response effect, the basic response action sequence corresponding to this transient type is optimized according to the actual convergence parameters and updated to the preset response action library. This collaborative approach between the preset recurrent neural network model and the preset response action library enhances their adaptability and gradually improves the prediction and response capabilities for transient events.

[0202] The following provides a detailed description of the load transient event response system for charging modules provided in this application. Please refer to [link / reference]. Figure 7 , Figure 7 An embodiment of the load transient event response system provided in this application includes:

[0203] The prediction unit 701 is used to acquire the time-domain waveform of the output electrical parameters of the charging module when the load transient trigger signal of the charging module is detected, and to predict the short-term output electrical parameter time-domain waveform of the charging module within a future preset time window based on the time-domain waveform of the output electrical parameters using a preset recurrent neural network model.

[0204] The first judgment unit 702 is used to judge whether the amplitude of the short-time output terminal electrical parameter time domain waveform exceeds the preset amplitude range of the sudden change waveform;

[0205] The construction unit 703 is used to calculate the dynamic change rate characterization parameter based on the time-domain waveform of the output electrical parameters if the condition is met, and to construct a difference matrix based on the dynamic change rate characterization parameter before the load transient trigger signal is triggered and the dynamic change rate characterization parameter after the load transient trigger signal is triggered. The dynamic change rate characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters.

[0206] Unit 704 is selected to determine the transient type based on the difference matrix and select the basic response action sequence corresponding to the transient type from the preset response action library;

[0207] The execution unit 705 is used to adjust the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module according to the pre-compensation action in the basic response action sequence, and to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment using a preset control algorithm, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

[0208] Optionally, building unit 703 is specifically used for:

[0209] The dynamic change rate characterization parameter within a first preset time period before the triggering time of the load transient trigger signal is used as the first parameter;

[0210] The dynamic rate of change within a second preset time after the triggering time of the load transient trigger signal is used as the second parameter;

[0211] Subtract the corresponding data points in the first and second parameters element by element to obtain the difference sequence;

[0212] The difference sequence is segmented in time domain order and then filled into a preset matrix template row by row to obtain the difference matrix.

[0213] Optionally, a matching unit 706 is also included, specifically for:

[0214] Singular value decomposition is performed on the difference matrix to obtain several singular values ​​and a singular vector corresponding to each singular value, where the singular values ​​are used to represent the contribution of the singular vector;

[0215] The largest singular value and its corresponding singular vector are used as the feature signature of the load transient trigger signal;

[0216] Perform similarity matching between the feature signature and the feature signature in the preset load transient signal feature signature library.

[0217] Optionally, transient types include load step increase type, load step decrease type, and load unknown disturbance type;

[0218] The transient type is determined based on the difference matrix, including:

[0219] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is higher than the first preset threshold, then the transient type is determined to be load step increase type.

[0220] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the first preset threshold and higher than the second preset threshold, then the transient type is determined to be load step reduction type.

[0221] If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the second preset threshold, the feature signature will be stored as a new sample in the preset load transient signal feature signature library, and the transient type will be determined as load unknown disturbance type.

[0222] Optionally, a second judgment unit 707 is also included, specifically used for:

[0223] Acquire the time-domain waveform of the output electrical parameters within a period from the first preset time before the triggering time of the load transient trigger signal to the second preset time after the triggering time, and divide the time-domain waveform of the output electrical parameters between the first preset time and the second preset time into several analysis periods according to the target time dimension;

[0224] Calculate the average slope of the time-domain waveform of the output electrical parameters within each analysis period;

[0225] Determine whether each average change slope value is within the preset step-increase slope range;

[0226] If so, update the transient type from unknown load disturbance type to load step increase type.

[0227] Optionally, a generation unit 708 is also included, specifically for:

[0228] If not, then the pre-compensation control parameters are generated by the preset action generation model based on the time-domain waveform of the output electrical parameters within the first preset time before the triggering time of the load transient trigger signal to the second preset time after the triggering time.

[0229] The basic response action sequence is constructed based on the pre-compensation control parameters, and the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence.

[0230] Optionally, the default recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer;

[0231] A pre-defined recurrent neural network model is used to predict the short-time output electrical parameter time-domain waveform of the charging module within a future pre-defined time window based on the output electrical parameter time-domain waveform, including:

[0232] The time-domain waveform of the output electrical parameters is input to the input layer, and the time-domain waveform of the output electrical parameters is distributed to the loop layer through the input layer;

[0233] The timing feature sequence of the output electrical parameters in the time domain waveform is extracted by the loop layer and then transmitted to the fully connected output layer.

[0234] The time-series feature sequence is mapped to the predicted electrical parameter values ​​at each discrete time point within a preset time window by a fully connected output layer.

[0235] The predicted electrical parameters at each discrete time point are reconstructed in chronological order to obtain the time-domain waveform of the short-time output electrical parameters.

[0236] Optionally, the prediction unit 701 is specifically used for:

[0237] The temporal feature sequence output by the recurrent layer is standardized to generate a standardized feature vector that matches the dimension of the fully connected output layer;

[0238] By using a set of linear transformation units in the fully connected output layer, the standardized feature vector is mapped from the high-dimensional feature space to the preset multi-dimensional output space, thereby obtaining the predicted values ​​of electrical parameters at each discrete time point within the preset time window.

[0239] Optionally, it also includes update unit 709, specifically used for:

[0240] Collect the actual convergence parameters after convergence is complete;

[0241] The preset recurrent neural network model and the preset response action library are updated collaboratively based on the actual convergence parameters, transient types, and basic response action sequences.

[0242] For details on the implementation method, please refer to [link / reference]. Figures 1-6Examples will not be described in detail here.

[0243] This application also provides a load transient event response device for a charging module; please refer to [link to relevant documentation]. Figure 8 , Figure 8 One embodiment of the load transient event response device provided in this application includes:

[0244] Processor 801, memory 802, input / output unit 803, bus 804;

[0245] The processor 801 is connected to the memory 802, the input / output unit 803, and the bus 804;

[0246] The memory 802 stores a program, and the processor 801 calls the program to execute any of the above-mentioned load transient event response methods.

[0247] This application also relates to a computer-readable storage medium storing a program, characterized in that, when the program is run on a computer, it causes the computer to execute any of the above-mentioned load transient event response methods.

[0248] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0249] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0250] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0251] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0252] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for load transient event response for a charging module, the method comprising: include: When a load transient trigger signal of the charging module is detected, the time-domain waveform of the output electrical parameters of the charging module is obtained, and a preset recurrent neural network model is used to predict the short-term time-domain waveform of the output electrical parameters of the charging module within a preset time window based on the time-domain waveform of the output electrical parameters. Determine whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the preset amplitude range of the sudden change waveform; If so, then the dynamic rate of change characterization parameter is calculated based on the time-domain waveform of the output electrical parameters, and a difference matrix is ​​constructed based on the dynamic rate of change characterization parameter before the load transient trigger signal is triggered and the dynamic rate of change characterization parameter after the load transient trigger signal is triggered. The dynamic rate of change characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters. The transient type is determined based on the difference matrix, and a basic response action sequence corresponding to the transient type is selected from a preset response action library; The duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence. A preset control algorithm is used to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

2. The load transient event response method according to claim 1, characterized in that, The construction of a difference matrix based on the dynamic rate of change characterization parameters before and after the load transient trigger signal includes: The dynamic rate of change characterization parameter within a first preset time period before the triggering time of the load transient trigger signal is used as the first parameter; The dynamic rate of change characterization parameter within a second preset time after the triggering time of the load transient trigger signal is used as the second parameter; Subtract the corresponding data points in the first parameter and the second parameter element by element to obtain the difference sequence; The difference sequence is segmented in time domain order and then filled into a preset matrix template row by row to obtain the difference matrix.

3. The load transient event response method according to claim 2, characterized in that, After constructing the difference matrix based on the dynamic rate of change characterization parameters before and after the load transient trigger signal, the load transient event response method further includes: The difference matrix is ​​subjected to singular value decomposition to obtain several singular values ​​and a singular vector corresponding to each singular value, wherein the singular values ​​are used to represent the contribution of the singular vector; The largest singular value and the corresponding singular vector are used as the feature signature of the load transient trigger signal; The feature signature is matched with the feature signature in the preset load transient signal feature signature library for similarity.

4. The load transient event response method according to claim 3, characterized in that, The transient types include load step increase type, load step decrease type, and load unknown disturbance type; Determining the transient type based on the difference matrix includes: If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is higher than the first preset threshold, then the transient type is determined to be the load step increase type. If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the first preset threshold and higher than the second preset threshold, then the transient type is determined to be the load step reduction type. If the maximum similarity between the feature signature and the feature signature in the preset load transient signal feature signature library is lower than the second preset threshold, then the feature signature is stored as a new sample in the preset load transient signal feature signature library, and the transient type is determined to be the load unknown disturbance type.

5. The load transient event response method according to claim 4, characterized in that, After determining the transient type as the unknown load disturbance type, the load transient event response method further includes: The output terminal electrical parameters time-domain waveforms are obtained from the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time, and the output terminal electrical parameters time-domain waveforms between the first preset duration and the second preset duration are divided into several analysis periods according to the target time dimension; Calculate the average slope value of the time-domain waveform of the output electrical parameters within each analysis period; Determine whether each of the average change slope values ​​is within the preset step-increase slope range; If so, the transient type is updated from the unknown load disturbance type to the load step increase type.

6. The load transient event response method according to claim 5, characterized in that, After determining whether each of the average change slope values ​​is within a preset step-increase slope range, the load transient event response method further includes: If not, then the pre-compensation control parameters are generated by the preset action generation model based on the time-domain waveform of the output electrical parameters within the first preset duration before the triggering time of the load transient trigger signal to the second preset duration after the triggering time. Based on the pre-compensation control parameters, a basic response action sequence is constructed, and the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module is adjusted according to the pre-compensation action in the basic response action sequence.

7. The load transient event response method according to any one of claims 1 to 6, characterized in that, The preset recurrent neural network model includes an input layer, a recurrent layer, and a fully connected output layer; The step of using a preset recurrent neural network model to predict the short-time output electrical parameter time-domain waveform of the charging module within a future preset time window based on the output electrical parameter time-domain waveform includes: The time-domain waveform of the output electrical parameters is input to the input layer, and the time-domain waveform of the output electrical parameters is distributed to the loop layer through the input layer; The time-series feature sequence of the output electrical parameters in the time-domain waveform is extracted by the loop layer, and the time-series feature sequence is transmitted to the fully connected output layer. The fully connected output layer maps the time-series feature sequence to the predicted electrical parameter values ​​at each discrete time point within a future preset time window. The predicted electrical parameters at each discrete time point are reconstructed in chronological order to obtain the time-domain waveform of the short-time output electrical parameters.

8. The load transient event response method according to claim 7, characterized in that, The step of mapping the time-series feature sequence to predicted electrical parameters at discrete time points within a future preset time window through the fully connected output layer includes: The temporal feature sequence output by the recurrent layer is standardized to generate a standardized feature vector that matches the dimension of the fully connected output layer. The standardized feature vector is mapped from the high-dimensional feature space to the preset multi-dimensional output space by a set of linear transformation units in the fully connected output layer, thereby obtaining the predicted values ​​of electrical parameters at each discrete time point within the preset time window.

9. The load transient event response method according to any one of claims 1 to 6, characterized in that, After the power conversion unit is driven to perform dynamic convergence via the control signal, the load transient event response method further includes: Collect the actual convergence parameters after convergence is complete; The preset recurrent neural network model and the preset response action library are updated collaboratively based on the actual convergence parameters, the transient type, and the basic response action sequence.

10. A load transient event response system for a charging module, characterized in that, include: The prediction unit is used to acquire the time-domain waveform of the output electrical parameters of the charging module when a load transient trigger signal of the charging module is detected, and to predict the short-term time-domain waveform of the output electrical parameters of the charging module within a future preset time window based on the time-domain waveform of the output electrical parameters using a preset recurrent neural network model. The judgment unit is used to determine whether the amplitude of the short-time output terminal electrical parameter time-domain waveform exceeds the amplitude range of the preset sudden change waveform; The construction unit is used to calculate the dynamic change rate characterization parameter based on the time-domain waveform of the output electrical parameters if the condition is met, and to construct a difference matrix based on the dynamic change rate characterization parameter before the load transient trigger signal is triggered and the dynamic change rate characterization parameter after the load transient trigger signal is triggered. The dynamic change rate characterization parameter is used to represent the instantaneous dynamic evolution trend of the time-domain waveform of the output electrical parameters. The selection unit is used to determine the transient type based on the difference matrix and select a basic response action sequence corresponding to the transient type from a preset response action library; The execution unit is used to adjust the duty cycle of the switching transistor drive parameters of the power conversion unit in the charging module according to the pre-compensation action in the basic response action sequence, and to generate a control signal based on the time-domain waveform of the output electrical parameters of the charging module after adjustment using a preset control algorithm, so as to drive the power conversion unit to perform dynamic convergence through the control signal.

11. A load transient event response device for a charging module, characterized in that, include: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor invokes to execute the load transient event response method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the load transient event response method as described in any one of claims 1 to 9.

Images