An AI scheduling management system for optimizing SGT MOSFET switching losses

Abstract

This application belongs to the field of scheduling management technology. It discloses an AI scheduling management system for optimizing SGT MOSFET switching losses. The system includes: real-time acquisition of operating characteristic data and condition transition characteristics of multiple parallel charging modules; recalculation of switching loss states to obtain the actual module loss sequence for the real-time scheduling cycle; acquisition of output requests to be executed and type splitting to obtain a subtask distribution requirement set; evaluation of the actual thermal recovery state inside the SGT MOSFET switch and interception of pseudo-used modules to filter and obtain a set of callable modules; construction of candidate scheduling schemes and prediction of switching losses, outputting the global loss evaluation results of the corresponding candidate scheduling schemes; determination of the target scheduling scheme and generation of control instructions for collaborative management; and improvement of the accuracy and interpretability of switching loss scheduling in multi-parallel cluster scenarios.

Description

Technical Field

[0001] This application relates to the field of scheduling management technology, and more specifically, to an AI scheduling management system for optimizing the switching losses of SGT MOSFETs. Background Technology

[0002] SGT MOSFET is a power semiconductor device with extremely low on-resistance and excellent high-frequency characteristics. It is a key component in multi-parallel charging module clusters. The optimized scheduling of its switching losses has a direct impact on reducing heat generation and improving power conversion efficiency. Traditional scheduling management systems commonly used in existing technologies can complete basic module rotation and allocation under ideal conditions with stable load fluctuations. However, for complex parallel charging scenarios with frequent high current jumps, they cannot effectively achieve precise loss control of underlying devices, and there are still many technical bottlenecks.

[0003] In practical multi-parallel charging scheduling scenarios, system load often experiences drastic changes in operating conditions due to external demand. However, traditional scheduling management systems neglect to capture these changes and update loss status at the boundaries of the scheduling window. This can lead to the system continuing to use the old loss sorting labels from the previous cycle after a sudden change in operating conditions, resulting in the incorrect allocation of high-frequency tasks to charging modules with extremely high losses under the new conditions. Furthermore, the output requests of a charging system typically consist of a mixture of stable basic power supply demand and high-frequency fluctuating dynamic compensation demand. Traditional scheduling management systems treat output requests as a uniform allocation of the overall power load, failing to rationally allocate tasks based on their response frequency. This forces modules unsuitable for frequent operations to undertake switching-intensive compensation tasks, resulting in the overall parallel cluster maintaining high actual switching losses while meeting total output requirements. Simultaneously, SGT... After MOSFETs operate at high frequencies, the dissipation of internal heat exhibits a significant lag. Traditional scheduling and management systems often rely solely on external temperature measurements, such as the surface temperature of the heatsink, to determine module recovery. This fails to effectively extrapolate the actual residual thermal stress within the device, potentially leading to the misclassification of modules whose surfaces have cooled but whose internal cores have not yet recovered as usable, thus causing repeated thermal shocks. On the other hand, when multiple parallel modules are switched over, significant hard switching overlap and reverse recovery losses occur. Traditional scheduling and management systems evaluate the overall power of candidate solutions based solely on steady-state conditions, without considering the transient impact costs during the switching transition. Ultimately, the selected optimal solution often experiences severe transient loss spikes during actual handover, affecting the accuracy of overall loss assessment and the reliability of hardware collaborative control and thermal management.

[0004] In view of this, this application proposes an AI scheduling and management system for optimizing SGT MOSFET switching losses to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, this application provides the following technical solution: an SGT MOSFET switching loss optimization AI scheduling management system, comprising: The status refresh module acquires the operating characteristic data of multiple parallel charging modules in real time and identifies the working condition transition characteristics of the real-time scheduling cycle. When the working condition transition characteristics meet the preset refresh conditions, the switching loss status of each charging module is recalculated based on the operating characteristic data, and the loss label across cycles is blocked to obtain the real module loss sequence of the real-time scheduling cycle. The task splitting module obtains the output requests to be executed, splits the output requests to be executed by type, and obtains the basic power supply task and the switch dense compensation sub-task respectively, which are combined into a sub-task distribution demand set. The thermal recovery verification module extracts cooling process information from the operational characteristic data of each charging module after completing historical high-frequency tasks, and evaluates the true thermal recovery state inside the SGT MOSFET switch; it performs pseudo-use module interception based on the true thermal recovery state; and it combines the interception results and the true module loss sequence to screen callable modules and obtain a set of callable modules. The loss assessment module constructs candidate scheduling schemes based on the set of callable modules and the subtask distribution requirement set; for each group of candidate scheduling schemes, it separates the handover transition interval and the stable power supply interval, and combines the operating characteristic data to predict the switching loss of each group of candidate scheduling schemes, and outputs the global loss assessment result of the corresponding candidate scheduling scheme. The collaborative execution module determines the target scheduling scheme based on the global loss assessment results; generates module control instructions that match the target scheduling scheme, and uses these module control instructions to perform collaborative management and control on the corresponding charging modules; the modules are connected to each other via wired and / or wireless means.

[0006] Furthermore, methods for identifying the operational transition characteristics of real-time scheduling cycles include: Extract the first sampled dataset from the end of the previous scheduling cycle and the second sampled dataset from the beginning of the real-time scheduling cycle from the running characteristic data. The sampled dataset includes DC bus voltage, total output current and duty cycle of SGT MOSFET. Calculate the bus voltage step difference, current output abrupt change rate, and duty cycle offset between the second sampling dataset and the first sampling dataset, respectively. If any one of the bus voltage step difference, current output mutation rate, or duty cycle offset exceeds the corresponding preset steady-state tolerance threshold, the corresponding physical quantity type and value will be combined into a working condition transition characteristic.

[0007] Furthermore, when the operating condition transition characteristics meet the preset refresh conditions, the switching loss state of each charging module is recalculated based on the operating characteristic data. Methods to prevent the reuse of loss labels across cycles include: The difference between the calculated value of the working condition transition characteristic and the corresponding preset steady-state tolerance threshold is used to obtain the excess limit. If the excess limit is higher than the preset recalculation trigger threshold, a tag refresh instruction is generated and the initial loss tag of the corresponding charging module in the previous scheduling cycle is cleared. After receiving the tag refresh command, extract the real-time switching frequency, switching transient voltage and peak shutdown current of each charging module from the real-time scheduling cycle's operational characteristic data; Obtain the preset on-resistance of the SGT MOSFET switch and the recovery charge of the body diode. Combine the real-time switching frequency, switching transient voltage and peak turn-off current to calculate the estimated instantaneous on-resistance loss and instantaneous switching loss of the corresponding charging module. Then sum them up to obtain the recalculated value of the total module loss. Charging modules whose total recalculated loss value is lower than the preset safe loss limit are filtered out and sorted in ascending order based on the size of the total recalculated loss value to obtain the true module loss sequence.

[0008] Furthermore, methods for splitting the type of the output request to be executed include: Identify the target output power command in the output request to be executed, and extract the low-frequency power component and high-frequency power component of the target output power command within a preset time window; The number of steady-state modules is obtained by calculating the ratio of the low-frequency power component to the rated power of the corresponding charging module. The average power allocation value of the low-frequency power component in each steady-state module is calculated. The average power allocation value is encapsulated with the constraint condition of reducing the duty cycle adjustment frequency to generate the basic power supply task. Frequency domain analysis is performed on the high-frequency power components to output the fluctuation frequency characteristics corresponding to the dynamic compensation requirements; the amplitude of the high-frequency power components is bound to the fluctuation frequency characteristics, and driving conditions to improve the switching response speed are added to generate a switch-intensive compensation subtask. The power demand nodes corresponding to the basic power supply subtask and the frequency response nodes corresponding to the switch-dense compensation subtask are time-aligned and combined to obtain the subtask distribution demand set.

[0009] Furthermore, methods for evaluating the true thermal recovery state inside the SGT MOSFET switch include: Locate the shutdown time of each charging module after it last completed a historical high-frequency task. At the same time, extract the ambient temperature and heat sink surface temperature from the shutdown time to the current time from the operation feature data, and plot them as a temperature decay curve to obtain cooling process information. Calculate the real-time temperature difference between the heat sink surface temperature and the ambient temperature at the current moment in the temperature decay curve, as well as the temperature drop gradient of the heat sink surface at the current moment. Combine this with the preset chip thermal resistance parameters to estimate the internal junction temperature at the current moment. The residual thermal coefficient is obtained by statistically analyzing the duration of continuous high-frequency switching and the average commutation frequency of the corresponding charging module during the most recent execution of a historical high-frequency task, and then multiplying them. The time interval from the shutdown moment to the current moment is calculated as the cooling interval. The ratio of the residual heat coefficient to the cooling interval is calculated to obtain the residual heat factor. The residual heat factor is multiplied by the preset heat mapping coefficient to obtain the residual heat compensation temperature. The real-time core junction temperature is obtained by summing the residual heat compensation temperature and the estimated internal junction temperature, and this real-time core junction temperature is taken as the true thermal recovery state.

[0010] Furthermore, pseudo-usage module interception is performed based on the actual thermal recovery state; the methods for filtering callable modules by combining the interception results and the actual module loss sequence include: The thermal deviation is obtained by calculating the temperature difference between the heat sink surface temperature of each charging module at the current moment and the real-time core junction temperature corresponding to the thermal recovery state. If the surface temperature of the heat sink is lower than the preset safe temperature threshold at the current moment, and the corresponding thermal deviation is higher than the preset internal thermal safety threshold, then the corresponding charging module will be identified as a fake usage module, and a scheduling lock interception command will be generated to execute on the fake usage module. The remaining charging modules, except for the pseudo-use modules, are marked as available candidate modules, and the sorting number of each available candidate module is identified by mapping it to the real module loss sequence. All available candidate modules are sorted in ascending order of their serial numbers. Available candidate modules belonging to the first category of preset number ranges are assigned to the switch intensive compensation subtask and used as a high-priority calling subset. Available candidate modules belonging to the second category of preset number ranges are assigned to the basic power supply task and used as a low-priority calling subset. Standby rotation instructions are sent to the remaining available candidate modules and assigned to the forced rotation subset. Integrate the high-priority call subset and the low-priority call subset into a set of callable modules.

[0011] Furthermore, the methods for constructing candidate scheduling schemes based on the set of callable modules and the set of subtask distribution requirements include: Extract the power demand node sequence corresponding to the basic power supply task and the frequency response node sequence corresponding to the switch-dense compensation subtask. Traverse the high-priority call subset and combine its modules into a high-frequency response array; traverse the low-priority call subset and combine its modules into a basic conduction array; cross-match the high-frequency response array and the basic conduction array to generate a preliminary scheduling array group; The system compiles a list of currently running online modules and compares it with the time sequence of each array in the initial scheduling array group. If a module is marked as being called at any time but is not in the list of online modules, then that time is marked as the entry node for the corresponding module. If a module is marked as being exited at any time but is in the list of online modules, then that time is marked as the exit node for the corresponding module. If several charging modules simultaneously trigger the cut-in node or cut-out node within the same preset very short time window, it is determined that the corresponding array has a conflict, and the array is removed from the initial scheduling array group; the output of each remaining array group is used as a candidate scheduling scheme.

[0012] Furthermore, for each group of candidate scheduling schemes, the handover transition interval and stable power supply interval are separated, and the switching loss prediction method for each group of candidate scheduling schemes is based on operational characteristic data, including: Obtain the preset dead time of the SGT MOSFET switch, extract the entry node and exit node in each candidate scheduling scheme, and extend the time according to the preset dead time with each node as the center to obtain the handover transition interval, while taking the remaining interval as the stable power supply interval. Within a stable power supply range, the expected operating frequency and expected effective current in the operational characteristic data corresponding to the candidate scheduling schemes are statistically analyzed, and the steady-state conduction loss is calculated in combination with the preset conduction internal resistance. At the same time, the steady-state switching loss is calculated by multiplying the unit switching energy loss corresponding to the expected effective current with the expected operating frequency. The steady-state conduction loss and steady-state switching loss are summed to obtain the estimated steady-state switching loss for the stable power supply range. Within the handover transition interval, the voltage fluctuation amplitude and surge current peak of the corresponding charging module at the moment of switching in or out are extracted, and the instantaneous energy loss estimate is obtained by integration according to the preset switching time; at the same time, the product of the voltage fluctuation amplitude and the recovery charge of the body diode is calculated to obtain the recovery energy loss estimate; the instantaneous energy loss estimate and the recovery energy loss estimate are summed to obtain the switching impact loss estimate of the handover transition interval. The weighted sum of the estimated switching impact losses of all handover transition sections and the estimated steady-state switching losses of all stable power supply sections in the same candidate scheduling scheme is used to obtain the global loss assessment result of the candidate scheduling scheme.

[0013] Furthermore, methods for determining the target scheduling scheme based on the global loss assessment results include: All candidate scheduling schemes are sorted in ascending order according to the numerical value of the global loss assessment results. The candidate scheduling scheme at the top is selected, and the corresponding set of callable modules is checked to see if it is online. If it is online, the candidate scheduling scheme is determined as the target scheduling scheme; otherwise, the next candidate scheduling scheme is checked until the corresponding set of callable modules is determined to be online.

[0014] Furthermore, generating module control instructions that match the target scheduling scheme, and using these module control instructions to perform coordinated management and control of the corresponding charging modules includes the following methods: For the switching-in and switching-out times of each charging module in the target scheduling scheme, generate timing pulse commands to execute grid connection or grid disconnection actions; Based on the basic power supply task or switch-intensive compensation subtask allocated by the target scheduling scheme, drive configuration instructions are generated to adjust the switching resistance and dead time of the SGT MOSFET.

[0015] The technical effects and advantages of the SGT MOSFET switching loss optimization AI scheduling management system proposed in this application are as follows: The system recalculates switching loss states based on operating condition transition characteristics, while preventing the reuse of loss labels across cycles. This improves upon existing technologies where the continued use of old labels after sudden changes in operating conditions leads to incorrect module scheduling and localized overheating. Specifically, output requests are split into basic power supply and switching-intensive compensation subtasks for targeted allocation. This avoids the problem of traditional methods performing overall allocation, which results in unsuitable modules for high-frequency operations being assigned dynamic compensation tasks, thus drastically amplifying switching losses. The system combines cooling process information to calculate the true thermal recovery state and identifies and intercepts pseudo-usable modules, ensuring that modules with surface cooling but residual internal thermal stress are not misjudged as healthy, preventing repeated residual heat shocks. Segmented loss prediction is performed using separate handover transition intervals and stable power supply intervals, solving the problem of conventional methods omitting power consumption at the moment of switching in and out, leading to distorted evaluation results of the scheduling scheme. This improves the accuracy and interpretability of switching loss scheduling in multi-path parallel cluster scenarios, while ensuring that the final output collaborative control commands have effective energy-saving effects and hardware operational reliability in real-world scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an AI scheduling management system for optimizing the switching losses of an SGT MOSFET according to this application; Figure 2 This is a schematic diagram of the working condition transition feature identification process in this application; Figure 3 A schematic diagram illustrating the construction process of the candidate scheduling schemes in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Example

[0018] Please see Figure 1 As shown in this embodiment, an SGT MOSFET switching loss optimization AI scheduling management system includes: The status refresh module acquires the operating characteristic data of multiple parallel charging modules in real time and identifies the working condition transition characteristics of the real-time scheduling cycle. When the working condition transition characteristics meet the preset refresh conditions, the switching loss status of each charging module is recalculated based on the operating characteristic data, and the loss label across cycles is blocked to obtain the real module loss sequence of the real-time scheduling cycle. The task splitting module obtains the output requests to be executed, splits the output requests to be executed by type, and obtains the basic power supply task and the switch dense compensation sub-task respectively, which are combined into a sub-task distribution demand set. The thermal recovery verification module extracts cooling process information from the operational characteristic data of each charging module after completing historical high-frequency tasks, and evaluates the true thermal recovery state inside the SGT MOSFET switch; it performs pseudo-use module interception based on the true thermal recovery state; and it combines the interception results and the true module loss sequence to screen callable modules and obtain a set of callable modules. The loss assessment module constructs candidate scheduling schemes based on the set of callable modules and the subtask distribution requirement set; for each group of candidate scheduling schemes, it separates the handover transition interval and the stable power supply interval, and combines the operating characteristic data to predict the switching loss of each group of candidate scheduling schemes, and outputs the global loss assessment result of the corresponding candidate scheduling scheme. The collaborative execution module determines the target scheduling scheme based on the global loss assessment results; generates module control instructions that match the target scheduling scheme, and uses these module control instructions to perform collaborative management and control on the corresponding charging modules; the modules are connected to each other via wired and / or wireless means.

[0019] SGT MOSFET refers to shielded gate trench power MOSFET. Due to its extremely low on-resistance and excellent high-frequency switching characteristics, it is widely used in various power conversion circuits. In this embodiment, the application scenario of multi-parallel charging module is selected (such as high-power DC charging pile or energy storage converter cluster).

[0020] In this embodiment, high-precision sensors deployed inside each parallel charging module collect underlying physical and control signals in real time and synchronously according to a preset high-frequency sampling period (e.g., millisecond level) to obtain operational characteristic data and upload it in real time. The operational characteristic data includes electrical state data such as DC bus voltage, output current of each module, and peak switching transient voltage and turn-off current when a jump occurs; control command data such as the real-time operating frequency and PWM duty cycle of the SGT MOSFET; and environmental and thermodynamic data such as the ambient temperature inside the cabinet and the surface temperature of the heat sink of each module.

[0021] Methods for identifying the operational transition characteristics of real-time scheduling cycles include: Extract the first sampled dataset from the end of the previous scheduling cycle and the second sampled dataset from the beginning of the real-time scheduling cycle from the running characteristic data. The sampled dataset includes DC bus voltage, total output current and duty cycle of SGT MOSFET.

[0022] The first sampling dataset represents the electrical operating status of the parallel charging module at the end of the previous scheduling cycle, and the second sampling dataset represents the electrical operating status of the parallel charging module at the beginning of the current real-time scheduling cycle. The DC bus voltage reflects the change in the current DC voltage, the total output current reflects the change in the total load borne by the parallel charging module, and the duty cycle reflects the driving adjustment magnitude of the SGT MOSFET performed by the underlying control module to maintain output stability. By extracting the sampling datasets at the boundary between the two scheduling cycles, the observation range of the operating condition change is limited to the boundary of the scheduling cycle, providing a comparison benchmark for subsequent identification of whether cross-cycle load transitions have occurred.

[0023] Calculate the bus voltage step difference, current output abrupt change rate, and duty cycle offset between the second sampling dataset and the first sampling dataset, respectively.

[0024] The bus voltage step difference refers to the difference between the voltage at the beginning of the current scheduling cycle and the voltage at the end of the previous scheduling cycle, used to indicate whether a sudden voltage transition has occurred; the current output mutation rate refers to the rate of change of the total output current between the boundaries of two adjacent scheduling cycles, used to indicate the degree of change in load demand; the duty cycle offset refers to the change in the duty cycle of the SGT MOSFET drive at the boundary between two consecutive scheduling cycles, used to reflect whether the control loop has undergone significant adjustment after the operating conditions change.

[0025] If any one of the bus voltage step difference, current output mutation rate, or duty cycle offset exceeds the corresponding preset steady-state tolerance threshold, the corresponding physical quantity type and value will be combined into a working condition transition characteristic.

[0026] Preset steady-state tolerance thresholds are set for the bus voltage step difference, current output mutation rate, and duty cycle offset. The specific values ​​are based on the historical sampling fluctuation range of the parallel charging module under stable operation, used to distinguish between normal fluctuations and abnormal mutations. If any physical quantity exceeds its corresponding steady-state tolerance threshold, it indicates an electrical disturbance at the boundary of the corresponding scheduling cycle. In this case, the type label of the exceeding physical quantity and its value are combined to form a condition transition characteristic, for example... Figure 2 The diagram shows the process of identifying the transition features under different operating conditions.

[0027] When the operating condition transition characteristics meet the preset refresh conditions, the switching loss state of each charging module is recalculated based on the operating characteristic data. The methods to prevent the use of loss labels across cycles include: The difference between the calculated value of the operating condition transition characteristic and the corresponding preset steady-state tolerance threshold is used to obtain the excess limit. If the excess limit is higher than the preset recalculation trigger threshold, a tag refresh instruction is generated and the initial loss tag of the corresponding charging module in the previous scheduling cycle is cleared.

[0028] The "exceeding the limit" indicates the degree of deviation of the current operating condition transition from the normal situation; the preset recalculation trigger threshold is set based on the critical value in historical samples where changes in operating conditions cause the module loss ranking to be rearranged, and is used to determine whether the current transition is serious enough to cause the original scheduling ranking to fail; the initial loss tag refers to the status identifier generated after evaluation in the previous scheduling cycle, used to mark the real-time loss ranking priority of each charging module; the tag refresh instruction refers to the forced reset signal issued to the scheduling controller.

[0029] When the excess exceeds the preset recalculation trigger threshold, it indicates that the original sorting of conditions under the old operating conditions has become invalid. Therefore, a tag refresh command is generated to forcibly clear the initial loss tag bound to each module, preventing the system from continuing to use the old priority tag under the new operating conditions.

[0030] After receiving the tag refresh command, extract the real-time switching frequency, switching transient voltage, and peak shutdown current of each charging module from the real-time scheduling cycle's operational characteristic data.

[0031] After the old tags have been cleared by the tag refresh command, the real-time switching frequency, switching transient voltage, and peak shutdown current of each charging module in the corresponding real-time scheduling cycle are extracted again. The real-time switching frequency is used to determine the frequency of switching losses. The switching transient voltage reflects the potential difference across the corresponding component at the moment of a single turn-on or turn-off. The peak shutdown current reflects the maximum inrush current at the moment of disconnection.

[0032] Obtain the preset on-resistance of the SGT MOSFET switch and the recovery charge of the body diode. Combine the real-time switching frequency, switching transient voltage and peak turn-off current to calculate the estimated instantaneous conduction loss and instantaneous switching loss of the corresponding charging module. Then sum them up to obtain the recalculated value of the total module loss.

[0033] The preset on-resistance is an inherent physical parameter representing the conduction characteristics of the SGT MOSFET switch, obtained by consulting relevant theoretical knowledge of the device; the body diode recovery charge represents the amount of charge released during reverse recovery, and the specific value is obtained by consulting the configuration parameters of the corresponding SGT MOSFET switch; the instantaneous conduction loss estimate is the consumption amount calculated by applying Joule's law to the current and on-resistance; for the instantaneous switching loss estimate, it can be obtained by mathematically calculating the extracted switching transient voltage, peak turn-off current, body diode recovery charge, and real-time switching frequency by substituting them into a known switching frequency product formula, based on the hard-switching overlap loss theory.

[0034] The instantaneous conduction loss estimate is used to quantify the steady-state thermal stress level of the charging module in the current conduction state, while the instantaneous switching loss estimate is used to quantify the dynamic thermal shock level of the charging module caused by high-frequency operation and reverse recovery. The sum of the two is used to obtain the total loss recalculation value of the module, which can truly reflect the overall switching load level faced by the charging module under the condition transition.

[0035] Charging modules whose total recalculated loss value is lower than the preset safe loss limit are filtered out and sorted in ascending order based on the size of the total recalculated loss value to obtain the true module loss sequence.

[0036] The preset safe loss limit is a threshold set based on the rated heat dissipation capacity of the charging module, for example, it is set to 85% of the heat dissipation limit power; dangerous modules that are close to the thermal runaway boundary after recalculation are eliminated by comparing the threshold; the real module loss sequence refers to the sorting result generated by arranging the remaining healthy modules from low to high according to the magnitude of the loss cost.

[0037] Methods for splitting the type of output requests to be executed include: Identify the target output power command in the output request to be executed, and extract the low-frequency power component and high-frequency power component of the target output power command within a preset time window.

[0038] In this embodiment, the output request to be executed refers to the overall power demand signal of the current charging task, which is intelligently issued by the AI ​​of the upstream management terminal or the control terminal on the user side based on the specific working conditions. This signal includes the target power value that the parallel charging module cluster needs to continuously output in the next period of time and the response time limit requirement. The target output power instruction refers to the power reference value sequence extracted from the output request to be executed. The preset time window refers to the time interval set for the execution of component extraction. The specific length is set based on the load response status of the current scheduling cycle. For example, it can be the duration of one or more scheduling cycles to ensure that the window contains enough data to distinguish between low-frequency trends and high-frequency disturbances.

[0039] Low-frequency power components refer to the basic power components in the target output power command that change slowly and are approximately stable within a preset time window, representing the basic energy supply demand of the overall system. High-frequency power components refer to the rapidly fluctuating components superimposed on the basic requirements within the same time window, representing the dynamic disturbance demand that the overall system needs to compensate in real time. By analyzing historical scheduling samples using AI, the basis for dividing the two components is obtained, enabling classification correction and reducing misallocation due to overlapping boundaries.

[0040] The number of steady-state modules is obtained by calculating the ratio of the low-frequency power component to the rated power of the corresponding charging module, and the average power distribution value of the low-frequency power component in each steady-state module is calculated.

[0041] Rated power refers to the maximum continuous safe output power value specified in the design specifications of each charging module; steady-state module quantity refers to the minimum number of charging modules that theoretically need to be fully operational to carry all low-frequency power components; for example, when the low-frequency power component is 90kW and the rated power of a single module is 30kW, the corresponding steady-state module quantity is 3; when the low-frequency power component is 95kW, in order to ensure power supply redundancy, the corresponding steady-state module quantity is 4; average power distribution value refers to the theoretical power value of distributing the low-frequency power component evenly to each module undertaking the task of low-frequency component after determining the number of steady-state modules.

[0042] The average power allocation value is encapsulated with the constraint of reducing the duty cycle adjustment frequency to generate the basic power supply task.

[0043] In this embodiment, the reduced duty cycle adjustment frequency constraint is a control constraint attached to the power allocation command. Based on this constraint, the controller of the charging module receiving the task actively reduces the response bandwidth of its control terminal or increases dead-zone filtering, so that it ignores small external high-frequency disturbances when outputting the average power allocation value, and tries to keep the duty cycle constant. The basic power supply task is obtained by combining the average power allocation value and the constraint into a complete data structure. This type of task is specifically used to carry the steady-state conduction mode, requiring the module to maintain a continuous and stable power output without frequently changing the switching state.

[0044] Frequency domain analysis is performed on the high-frequency power components to output the fluctuation frequency characteristics corresponding to the dynamic compensation requirements.

[0045] In this embodiment, frequency domain analysis is performed by combining AI's intelligent analysis with existing theories such as the Fast Fourier Transform, transforming high-frequency components from the time domain to the frequency domain.

[0046] The corresponding dynamic compensation requirement refers to the fact that due to factors such as grid ripple, battery transient power fluctuations, rapid load changes, and uneven internal current distribution, the charging module must compensate for a portion of rapidly changing power demand in real time in addition to providing power in a steady state. This is a rapid compensation action that the entire system must perform to maintain a stable total output for this high-frequency fluctuation.

[0047] Fluctuation frequency characteristics refer to the main disturbance frequency components and their distribution obtained after frequency domain analysis. They are used to reflect the rate of change of dynamic compensation requirements and the frequency range in which they are distributed. For example, if the main energy in the spectrum is concentrated around 100Hz to 300Hz, it indicates that the overall system mainly faces lower frequency compensation requirements. If the main components are concentrated above 1kHz, it indicates that a faster switching response is required.

[0048] The amplitude of the high-frequency power component is bound to the fluctuation frequency characteristics, and driving conditions to improve the switching response speed are added to generate a switch-dense compensation subtask.

[0049] In this embodiment, the driving condition for improving switching response speed refers to a control constraint added to this type of task. It is mainly used to require the charging module undertaking this task to adjust the duty cycle and shorten the switching response lag more quickly than the basic power supply task. By combining and encapsulating the amplitude of the high-frequency power component, the fluctuation frequency characteristics, and the driving condition for improving switching response speed to form a switch-dense compensation subtask, the high-frequency compensation behavior that is most likely to cause high switching losses is independently separated from the total task.

[0050] The power demand nodes corresponding to the basic power supply subtask and the frequency response nodes corresponding to the switch-dense compensation subtask are time-aligned and combined to obtain the subtask distribution demand set.

[0051] The power demand node refers to the power demand sampling point corresponding to each sampling moment of the basic power supply task, which is used to indicate how much power the module required to provide basic power supply needs to output at a certain moment; the frequency response node refers to the compensation response requirement point corresponding to each sampling moment of the switching dense compensation subtask, which is used to indicate how fast the switching rhythm of the module required to provide high-frequency compensation at a certain moment is to respond to the compensation requirement; timing alignment refers to mapping these two types of nodes to the same time coordinate system.

[0052] Methods for evaluating the true thermal recovery state inside an SGT MOSFET switch include: Locate the shutdown time of each charging module after it last completed a historical high-frequency task. At the same time, extract the ambient temperature and radiator surface temperature from the shutdown time to the current time from the operation feature data, and plot them as a temperature decay curve to obtain cooling process information.

[0053] The historical high-frequency task refers to the most recent switch-intensive compensation subtask or other work task involving high-frequency switching that the charging module was assigned to execute; the shutdown time refers to the specific time point at which the charging module exits operation after completing the historical high-frequency task; the ambient temperature refers to the air temperature inside the cabinet where the charging module is located; the radiator surface temperature refers to the temperature value measured on the radiator of the charging module; the temperature decay curve refers to the data sequence of temperature changes over time generated by arranging the radiator surface temperature and ambient temperature in chronological order from the shutdown time to the current time, and this curve is used as information on the cooling process.

[0054] The system calculates the real-time temperature difference between the heatsink surface temperature and the ambient temperature at the current moment in the temperature decay curve, as well as the temperature gradient of the heatsink surface at the current moment. Combined with the preset chip thermal resistance parameters, it estimates the internal junction temperature at the current moment.

[0055] The real-time temperature difference refers to the difference between the surface temperature of the heat sink and the ambient temperature at the current moment, reflecting how much residual heat the heat sink still retains relative to the surrounding environment. The descent gradient refers to the rate at which the surface temperature of the heat sink decreases at the current moment, calculated using derivatives, reflecting the current state of the heat dissipation process. The preset chip thermal resistance parameter refers to the thermal resistance between the junction of the SGT MOSFET chip and the surface of the heat sink, indicating the ease with which heat is conducted from the inside of the chip to the outside. This thermal resistance value can be obtained from the setting specifications of the SGT MOSFET switch, for example... The estimated internal junction temperature refers to the core temperature inside the SGT MOSFET chip obtained by back-calculating based on these externally measurable data. In this embodiment, those skilled in the art can calculate the appropriate level of the chip's internal junction temperature under the current external heat dissipation conditions based on existing thermal resistance network analysis theories and common mathematical and physical knowledge, according to the real-time temperature difference, the decreasing gradient of the heat sink surface temperature, and the chip's thermal resistance parameters, combined with execution experience.

[0056] The residual thermal coefficient is obtained by statistically analyzing the duration of continuous high-frequency switching and the average commutation frequency of the corresponding charging module during the most recent execution of a historical high-frequency task, and then multiplying them.

[0057] The continuous high-frequency switching duration refers to the total duration of the charging module from the start of high-frequency operation to the end of operation in the most recent historical high-frequency task. The average commutation frequency refers to the average density of commutation actions caused by current transfer in the SGT MOSFET during the duration of the task, i.e., how many commutations occurred per unit time. Commutation actions can trigger reverse recovery of the body diode and other internal transient thermal pulses. Therefore, the residual thermal coefficient is obtained by multiplying the continuous high-frequency switching duration and the average commutation frequency. This value is a dimensionless evaluation value used to comprehensively represent the total amount of thermal shock injected into the charging module before it exits the high-frequency task. The larger the coefficient, the deeper the residual thermal stress accumulated inside the charging module, and the less suitable it is to be immediately put into a high-frequency task again.

[0058] The time interval from the shutdown moment to the current moment is calculated as the cooling interval, and the ratio of the residual heat coefficient to the cooling interval is calculated to obtain the residual heat factor.

[0059] The residual heat factor is obtained by calculating the ratio of the residual thermal coefficient to the cooling interval. It represents the ratio of the total thermal stress accumulated before exiting the high-frequency task to the time that has been spent dissipating heat. The larger the value, the more the corresponding charging module has undergone a cooling period, but due to the large amount of thermal stress accumulated before, there is still a heavy residual heat load inside. Conversely, the smaller the value, the more the internal heat has been dissipated. This factor is a quantitative value that changes dynamically over time and can reflect the real-time residual heat severity of the corresponding charging module at the current moment.

[0060] The residual heat compensation temperature is obtained by multiplying the residual heat factor with the preset heat mapping coefficient.

[0061] In this embodiment, the preset thermal mapping coefficient is a conversion factor used to convert the dimensionless residual heat factor into a compensation value with real temperature physical meaning. The specific value is set based on the thermal capacitance characteristics of the SGT MOSFET device and the mapping relationship between the residual heat factor and temperature obtained from the cooling process in multiple historical high-frequency tasks. The residual heat compensation temperature refers to the temperature correction amount obtained after multiplication and conversion by the preset thermal mapping coefficient, which is used to represent the equivalent temperature offset corresponding to the internal residual thermal stress caused by historical high-frequency tasks.

[0062] The real-time core junction temperature is obtained by summing the residual heat compensation temperature and the estimated internal junction temperature, and this real-time core junction temperature is taken as the true thermal recovery state.

[0063] The real-time core junction temperature is obtained by superimposing the estimated internal junction temperature with the residual heat compensation temperature. This value is used to more accurately reflect the temperature value of the core thermal state of the SGT MOSFET chip. For example, if the conventionally calculated estimated internal junction temperature is 70℃ and the residual heat compensation temperature is 12℃, then the real-time core junction temperature is 82℃, indicating that the actual temperature inside the module is higher than the temperature corresponding to the data observed externally.

[0064] The method of intercepting pseudo-usage modules based on the actual thermal recovery state, and filtering callable modules by combining the interception results and the actual module loss sequence, includes: The thermal deviation is obtained by calculating the temperature difference between the heat sink surface temperature of each charging module at the current moment and the real-time core junction temperature corresponding to the thermal recovery state.

[0065] The thermal deviation is used to represent the temperature difference between the external observable temperature and the actual internal thermal state of the same charging module. The larger the value, the greater the difference between the external and internal temperatures of the module.

[0066] If the surface temperature of the radiator is lower than the preset safe temperature threshold at the current moment, and the corresponding thermal deviation is higher than the preset internal thermal safety threshold, then the corresponding charging module will be identified as a fake usage module, and a scheduling lock interception command will be generated to execute on the fake usage module.

[0067] In this embodiment, the preset safe temperature threshold refers to the upper limit of the external temperature at which the dispatching terminal determines that the external temperature measurement conditions of the charging module are in a state where it can be put back into operation. It is set based on the temperature level safety specifications of the heat sink, for example, it can be set to 55℃ or 60℃. The preset internal thermal safety threshold refers to the maximum tolerance limit for the charging module to have an external cold and internal heat deviation. It is set based on the statistical analysis of the critical difference in the charging module's performance degradation caused by internal overheating in historical records, for example, it can be set to 15℃ or 20℃.

[0068] When the surface temperature of a module has dropped below the safe temperature threshold, but the thermal deviation is still higher than the internal thermal safety threshold, it indicates that the charging module is in a pseudo-recovery state where the surface has cooled down but the internal thermal stress has not dissipated. Therefore, it is determined to be a pseudo-use module. The scheduling lock interception command refers to the forced blocking signal issued by the control terminal to the scheduling controller. It is used to remove the pseudo-use module from the callable resources in the current scheduling cycle to prevent it from being reassigned to any task and to avoid putting the module that has not yet completed internal thermal recovery back into high-frequency task operation.

[0069] The remaining charging modules, excluding the pseudo-use modules, are marked as available candidate modules, and their sorting numbers are mapped to the real module loss sequence to identify each available candidate module.

[0070] Available candidate modules refer to charging modules that have not been locked after being intercepted by pseudo-use modules. They represent charging modules that have passed the hot recovery test within the current scheduling cycle and are qualified to accept tasks again. The mapping is achieved by matching the identifiers of available candidate modules with the actual module loss sequence. The sorting number refers to the position number of each available candidate module in the actual module loss sequence. The earlier the number, the lower the overall switching loss and the better the condition of the charging module under the current operating conditions.

[0071] All available candidate modules are sorted in ascending order of their serial numbers. Available candidate modules belonging to the first category of preset number ranges are assigned to the switch-intensive compensation subtask and used as a high-priority calling subset. Available candidate modules belonging to the second category of preset number ranges are assigned to the basic power supply task and used as a low-priority calling subset.

[0072] One type of preset numbering interval refers to the range of numbers ranked first in the sorting sequence, corresponding to the charging module with the lowest overall loss, which is most suitable for undertaking the high-frequency switching intensive compensation subtask. The second type of preset numbering interval refers to the range of numbers following the first type of preset numbering interval, corresponding to the charging module with a higher overall loss, which is suitable for undertaking the basic power supply task that does not require frequent switching. The specific division of numbering intervals can be determined based on the number of steady-state modules and the number of response modules required by the subtask in the current scheduling cycle. For example, if there are a total of 8 available candidate modules, the switching intensive compensation subtask requires 2 modules, and the basic power supply task requires 4 modules, then the module ranked 2nd is assigned to the high-priority calling subset, and the module ranked 4th is assigned to the low-priority calling subset.

[0073] Send standby rotation instructions to the remaining available candidate modules and assign them to the forced rotation subset.

[0074] The remaining available candidate modules refer to those whose sorting numbers fall outside the preset numbering range of Category 1 and Category 2. These are modules that, although they passed the pseudo-use module interception check, have a relatively low loss ranking in the current scheduling cycle, and their losses are much greater than other modules, making them unsuitable for participating in this round of scheduling tasks. The standby rotation instruction is a control signal issued to these remaining modules, causing them to enter a low-power standby state and cease performing any tasks. These charging modules are then grouped into the forced rotation subset. This ensures that these modules have time for continuous heat dissipation and recovery, providing an opportunity for them to re-participate in the sorting process in subsequent scheduling cycles. Simultaneously, it avoids the situation where forcing the use of modules with state deviations leads to an overall high loss in the scheduling scheme.

[0075] Integrate the high-priority call subset and the low-priority call subset into a set of callable modules.

[0076] The methods for constructing candidate scheduling schemes based on the set of callable modules and the set of subtask distribution requirements include: Extract the power demand node sequence corresponding to the basic power supply task and the frequency response node sequence corresponding to the switch-dense compensation subtask.

[0077] The power demand node sequence refers to the ordered set of data points of the power required by the basic power supply task at each time moment, obtained from the requirements of a certain task; the frequency response node sequence refers to the ordered set of data points of the compensation amplitude and response frequency requirements of the switching dense compensation subtask at each time moment, obtained from the requirements of the same task.

[0078] Iterate through the high-priority call subset and combine its modules into a high-frequency response array; iterate through the low-priority call subset and combine its modules into a basic conduction array.

[0079] The high-priority call subset uses different permutations and combinations of charging modules to form a high-frequency response array, which is used to undertake the switch-dense compensation subtask; the low-priority call subset uses various permutations and combinations of charging modules to form a basic conduction array, which is used to undertake the basic power supply task.

[0080] The high-frequency response array and the basic conduction array are cross-matched to generate a preliminary scheduling array group.

[0081] In this embodiment, cross-matching refers to pairing each combination in the high-frequency response array with each combination in the basic conduction array to generate a preliminary scheduling array group, forming several complete scheduling schemes. For example, if there are 3 combinations in the high-frequency response array and 4 combinations in the basic conduction array, then after cross-matching, the preliminary scheduling array group will contain 12 scheduling schemes.

[0082] The system compiles a list of online modules currently in operation and compares their timing with that of each array in the initial scheduling array group.

[0083] The online module list refers to the set of charging modules that are actually undertaking the output power task within the current scheduling cycle. The current actual state represented by the online module list is compared with the corresponding planned future module state sequence in the preliminary scheduling array group on a time-by-time basis. This is used to search for which modules need to be added to or removed from operation at which times for each scheduling scheme relative to the current operating state, thereby accurately locating the entry and exit points of the corresponding charging modules.

[0084] If a module is marked as being called at any given time but is not in the list of online modules, then that time will be marked as the entry point for the corresponding module.

[0085] The "call status" refers to the state at which a certain charging module in a certain array is considered to be in operation at a certain moment; the "switch-in node" refers to the critical point at which the module switches from never running to being in operation, representing a module startup and grid connection action, and accurately locating the moment when the new module begins to take over the load.

[0086] If a module is marked as exited at any time but is in the list of online modules, then that time will be marked as the cut-out node for the corresponding module.

[0087] The exit state refers to the state in which a module in a certain array is considered to be in a state that should stop running at a certain moment; the cut-out node refers to the critical point at which the module switches from being currently running to being shut down, representing a module shutdown and network disconnection action.

[0088] If several charging modules simultaneously trigger the cut-in node or cut-out node within the same preset very short time window, it is determined that the corresponding array has a conflict, and the array is removed from the initial scheduling array group.

[0089] In this embodiment, the same preset very short time window refers to a microsecond-level time window used to detect concurrent switching conflicts. The specific time length is set based on the hardware switching response speed of the parallel charging module, for example, it can be set to 50 microseconds. A conflict occurs when multiple modules simultaneously trigger the switching-in or switching-out action within this very short time window. Such concurrent switching will cause a large-scale redistribution of the bus current, causing the SGT MOSFET switch to withstand transient voltage and current surges far exceeding the steady-state level. The array with the conflict is removed from the initial scheduling array group, eliminating the undesirable scheme with the risk of concurrent switching.

[0090] The output of each remaining array group is used as a candidate scheduling scheme.

[0091] The remaining arrays refer to the arrays remaining after removing conflicting arrays from the initial scheduling array group. These arrays do not have the risk of concurrent switching. Each scheduling scheme includes clearly defined module division of labor information, such as... Figure 3 The flowchart of the candidate scheduling scheme construction process is shown to obtain the candidate scheduling scheme.

[0092] For each group of candidate scheduling schemes, the handover transition interval and stable power supply interval are separated. The methods for predicting switching losses for each group of candidate scheduling schemes based on operational characteristic data include: Obtain the preset dead time of the SGT MOSFET switch, extract the entry and exit nodes in each candidate scheduling scheme, and extend the time according to the preset dead time with each node as the center to obtain the handover transition interval, while taking the remaining interval as the stable power supply interval.

[0093] In this embodiment, the preset dead time refers to the blank time interval forcibly inserted by the drive circuit of the SGT MOSFET to prevent shoot-through short circuits when switching between the upper and lower transistors. It is obtained by querying the drive specifications or parameter configuration of the device, for example, 1 microsecond. By taking the moment when the module actually switches in or out as the center, the preset dead time length is extended to both sides to obtain the handover transition interval corresponding to each switch in or out. This interval refers to the extremely short time period during which the current and voltage change drastically when the module switches from one state to another. The stable power supply interval refers to the interval remaining on the time axis of the corresponding candidate scheduling scheme after removing all handover transition intervals, indicating the stage when the corresponding charging module is in normal steady-state operation.

[0094] Within a stable power supply range, the expected operating frequency and expected effective current in the operational characteristic data corresponding to the candidate scheduling schemes are statistically analyzed, and the steady-state conduction loss is calculated in combination with the preset conduction resistance.

[0095] The expected operating frequency and expected effective current are the switching frequency and effective output current values ​​of the corresponding charging module within the stable power supply range, as planned by the candidate scheduling scheme; the formula for calculating steady-state conduction loss is: ;in This represents the steady-state conduction loss of the corresponding candidate scheduling scheme; Indicates the expected effective current; This indicates the preset on-resistance of the SGT MOSFET switch; for example, when the expected effective current is... The preset on-resistance is At that time, the steady-state conduction loss is Steady-state conduction loss is used to represent the basic heat dissipation level of the corresponding charging module when it is in a continuous conduction state in the candidate scheduling scheme.

[0096] Meanwhile, the steady-state switching loss is calculated by multiplying the unit switching energy loss corresponding to the expected effective current with the expected operating frequency.

[0097] The energy loss per unit number of switching cycles refers to the basic energy consumed by the SGT MOSFET switch to complete one turn-on or turn-off switching under specific current conditions. This can be obtained by querying the historical switching energy changes of the device. The steady-state switching loss is obtained by multiplying the energy loss per unit number of switching cycles by the expected operating frequency. For example, if the energy loss per unit number of switching cycles is 10 microjoules and the expected operating frequency is 100 kHz, the steady-state switching loss is about 1 watt. This steady-state switching loss is used to reflect the dynamic heat dissipation generated by the switching of the corresponding charging module during steady-state operation.

[0098] The steady-state switching loss is estimated by summing the steady-state conduction loss and the steady-state switching loss for the stable power supply range.

[0099] The steady-state switching loss estimate is obtained by summing the loss values ​​of both steady-state conduction loss and steady-state switching loss, which both belong to the steady-state power supply range. This value is used to quantify the overall steady-state thermal cost of the charging module in the steady-state power supply range.

[0100] Within the handover transition interval, the voltage fluctuation amplitude and surge current peak value of the corresponding charging module at the moment of switching in or out are extracted, and the instantaneous energy loss estimate is obtained by integral calculation according to the preset switching time.

[0101] The voltage surge amplitude refers to the drastic change in voltage across the corresponding charging module at the moment of switching in or out; the surge current peak value refers to the maximum inrush current flowing through the charging module at the moment of switching; the preset switching time refers to the time required for the SGTMOSFET switch to complete one turn-on or turn-off transition, which is obtained by querying the parameter configuration of the device. This time is used to represent the actual duration of voltage and current overlap; in this embodiment, according to the hard switching overlap loss theory, the voltage surge amplitude is multiplied by the surge current peak value to obtain the peak power at that moment, and then the peak power is integrated over the preset switching time to obtain the instantaneous energy loss estimate; this value is used to quantify the energy loss caused by voltage and current overlap during a single switch-in or switch-out action.

[0102] Simultaneously, the product of the voltage surge amplitude and the recovered charge of the body diode is calculated to obtain the estimated recovery loss energy.

[0103] The recovery loss energy estimate is obtained by multiplying the voltage surge amplitude and the body diode recovery charge based on the known relationship between charge and voltage in existing theories. This value is used to represent the additional energy consumed by the module due to the reverse recovery of the diode at the moment of switching in or out.

[0104] The estimated switching impact loss for this transition interval is obtained by summing the estimated instantaneous energy loss and the estimated recovery energy loss.

[0105] The switching impact loss estimate for the corresponding handover transition interval is obtained by summing the instantaneous energy loss estimate and the recovery energy loss estimate, which is used to represent the total energy loss caused by the cutting-in or cutting-out action.

[0106] The weighted sum of the estimated switching impact losses of all handover transition sections and the estimated steady-state switching losses of all stable power supply sections in the same candidate scheduling scheme is used to obtain the global loss assessment result of the candidate scheduling scheme.

[0107] In this process, the estimated switching impact loss and steady-state switching loss for all handover transition intervals and stable power supply intervals in the same candidate scheduling scheme are assigned different weights and then weighted and summed. It should be noted that both values ​​were normalized before the weighted summation to eliminate the difference in dimensions.

[0108] In this embodiment, different weights are set for different intervals based on the specific distribution of sub-task types. Steady-state switching loss mainly reflects the average heat dissipation generated by continuous conduction and normal switching during stable operation of the module. It has a long duration but changes gradually. Switching impulse loss mainly reflects the short-term high-density energy impulse generated by factors such as voltage change, current surge and reverse recovery of body diode at the moment of switching in or out of the module. Its duration is extremely short, but it is more likely to cause phenomena such as rapid rise in local junction temperature, which leads to a reduction in device life. Therefore, the weight of the handover transition interval needs to be greater than that of the stable power supply interval. The specific value can be set according to the corresponding operating conditions based on information such as temperature rise records, protection action records and fault warning records in historical operation records. The final evaluation value of the overall loss is obtained by weighted summation, that is, the global loss evaluation result.

[0109] Methods for determining the target scheduling scheme based on global loss assessment results include: All candidate scheduling schemes are sorted in ascending order according to the numerical value of the global loss assessment results. The candidate scheduling scheme at the top is selected, and the corresponding set of callable modules is checked to see if it is online. If it is online, the candidate scheduling scheme is determined as the target scheduling scheme; otherwise, the next candidate scheduling scheme is checked until the corresponding set of callable modules is determined to be online.

[0110] In this embodiment, all corresponding candidate scheduling schemes are sorted in ascending order according to the numerical value of the global loss assessment result. The higher the ranking of the candidate scheduling scheme, the lower the total loss. First, the first group of candidate scheduling schemes is selected, and the state of each charging module in the scheme before the scheme is officially executed is determined. If all the charging modules of the scheme are online, it means that the scheme can be executed immediately, so the candidate scheduling scheme is determined as the target scheduling scheme. If the charging modules of the scheme are not online, the next candidate scheduling scheme is checked in turn until a scheme that can be executed immediately is found as the target scheduling scheme.

[0111] The methods for generating module control commands that match the target scheduling scheme and using these module control commands to perform coordinated management and control of the corresponding charging modules include: It should be noted that in the multi-parallel charging module set corresponding to this embodiment, in order to achieve switching loss optimization and task collaborative management at the module cluster level, there are two core control dimensions: accurately determining when each module is put into operation and withdrawn to avoid handover conflicts, and dynamically changing the driving target of the SGT MOSFET according to the assigned task type. Therefore, this embodiment specifically generates timing pulse instructions and drive configuration instructions as an executable instruction set for the execution target scheduling scheme, so as to complete the global collaborative management from module peak shifting to device loss optimization with minimal communication overhead.

[0112] For each charging module in the target scheduling scheme, a timing pulse command is generated to execute grid connection or grid disconnection actions based on the corresponding cut-in and cut-out times.

[0113] For each charging module's respective entry and exit times, timing pulse commands are generated for control. The entry or exit time nodes arranged in the target scheduling scheme are translated into grid connection or degrid commands that the hardware can directly execute. Grid connection means controlling the corresponding charging module to connect to the current parallel scenario and start output. Degrid means controlling the corresponding charging module to exit the parallel scenario and stop power output. The timing pulse commands are executed according to the peak-shaving logic in the target scheduling scheme to avoid conflicts between multiple modules.

[0114] Based on the basic power supply task or switch-intensive compensation subtask allocated by the target scheduling scheme, drive configuration instructions are generated to adjust the switching resistance and dead time of the SGT MOSFET.

[0115] In this embodiment, the switching resistor refers to the gate drive resistor of the SGT MOSFET, which is used to adjust the turn-on and turn-off speed of the switch. The drive configuration command refers to the parameter configuration command issued to the underlying drive controller based on the specific task type assigned to different charging modules. For example, when a charging module is assigned to undertake the basic power supply task, since this task requires stable and continuous power output, the drive configuration command controls the module to operate in a low switching stress mode. At this time, it is necessary to increase the gate drive resistor or increase the dead time to slow down the change rate on the switching side, reduce the risk of cross-conduction, and suppress unnecessary high-frequency response. Conversely, if a charging module is assigned to undertake the switch-dense compensation sub-task, this type of task requires the ability to quickly track grid or load disturbances. Therefore, the drive configuration command is used to control it to improve dynamic response capability, reduce the gate drive resistor, or shorten the dead time to accelerate the turn-on and turn-off speed of the device.

[0116] The dead time value can be set based on the parameter configuration specifications of the corresponding device, such as being dynamically adjusted in the range of 100 nanoseconds to 500 nanoseconds; the specific value of the gate drive resistor can be switched in the range of a few ohms to tens of ohms; by executing drive configuration instructions, dynamic control from scheduling decision to lower-level device drive parameters is realized, so that the target scheduling scheme not only determines the entry and exit of the module, but also determines the optimal posture of each module.

[0117] This embodiment recalculates the switching loss state based on the characteristics of operating condition transitions, while preventing the reuse of loss labels across cycles. This improves upon the existing technology's problem of incorrect module scheduling and localized overheating caused by the continued use of old labels after sudden changes in operating conditions. Specifically, output requests are split into basic power supply and switch-intensive compensation sub-tasks for targeted allocation, avoiding the problem of traditional methods performing overall allocation that leads to unsuitable modules undertaking dynamic compensation tasks, thus drastically amplifying switching losses. The actual thermal recovery state is calculated by combining cooling process information, while identifying and intercepting pseudo-usable modules. This ensures that modules with surface cooling but residual internal thermal stress are not misjudged as healthy, preventing them from suffering repeated residual heat shocks. Segmented loss prediction is performed using separate handover transition intervals and stable power supply intervals, solving the problem of conventional methods omitting power consumption at the moment of switching in and out, leading to distorted evaluation results of the scheduling scheme. This improves the accuracy and interpretability of switching loss scheduling in multi-path parallel cluster scenarios, while ensuring that the final output collaborative control commands have effective energy-saving effects and hardware operational reliability in real-world scenarios.

[0118] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0119] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0120] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. An AI scheduling and management system for optimizing SGT MOSFET switching losses, characterized in that, include: The status refresh module acquires the operating characteristic data of multiple parallel charging modules in real time and identifies the operating condition transition characteristics of the real-time scheduling cycle. When the operating condition transition characteristics meet the preset refresh conditions, the switching loss state of each charging module is recalculated based on the operating characteristic data, the loss label across cycles is blocked, and the real module loss sequence of the real-time scheduling cycle is obtained. The task splitting module obtains the output requests to be executed, splits the output requests to be executed by type, and obtains the basic power supply task and the switch dense compensation sub-task respectively, which are combined into a sub-task distribution demand set. The thermal recovery verification module extracts cooling process information from the operational characteristic data of each charging module after completing historical high-frequency tasks, and evaluates the true thermal recovery state inside the SGT MOSFET switch; it performs pseudo-use module interception based on the true thermal recovery state; and it combines the interception results and the true module loss sequence to screen callable modules and obtain a set of callable modules. The loss assessment module constructs candidate scheduling schemes based on the set of callable modules and the subtask distribution requirement set; for each group of candidate scheduling schemes, it separates the handover transition interval and the stable power supply interval, and combines the operating characteristic data to predict the switching loss of each group of candidate scheduling schemes, and outputs the global loss assessment result of the corresponding candidate scheduling scheme. The collaborative execution module determines the target scheduling scheme based on the global loss assessment results; Generate module control instructions that match the target scheduling scheme, and use these module control instructions to perform coordinated management and control of the corresponding charging modules; the modules are connected to each other via wired and / or wireless means.

2. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 1, characterized in that, The methods for identifying the operational transition characteristics of the real-time scheduling cycle include: Extract the first sampled dataset from the end of the previous scheduling cycle and the second sampled dataset from the beginning of the real-time scheduling cycle from the running characteristic data. The sampled dataset includes DC bus voltage, total output current and duty cycle of SGT MOSFET. Calculate the bus voltage step difference, current output abrupt change rate, and duty cycle offset between the second sampling dataset and the first sampling dataset, respectively. If any one of the bus voltage step difference, current output mutation rate, or duty cycle offset exceeds the corresponding preset steady-state tolerance threshold, the corresponding physical quantity type and value will be combined into a working condition transition characteristic.

3. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 2, characterized in that, When the operating condition transition characteristics meet the preset refresh conditions, the switching loss state of each charging module is recalculated based on the operating characteristic data. The methods to prevent the use of loss labels across cycles include: The difference between the calculated value of the working condition transition characteristic and the corresponding preset steady-state tolerance threshold is used to obtain the excess limit. If the excess limit is higher than the preset recalculation trigger threshold, a tag refresh instruction is generated and the initial loss tag of the corresponding charging module in the previous scheduling cycle is cleared. After receiving the tag refresh command, extract the real-time switching frequency, switching transient voltage and peak shutdown current of each charging module from the real-time scheduling cycle's operational characteristic data; Obtain the preset on-resistance of the SGT MOSFET switch and the recovery charge of the body diode. Combine the real-time switching frequency, switching transient voltage and peak turn-off current to calculate the estimated instantaneous on-resistance loss and instantaneous switching loss of the corresponding charging module. Then sum them up to obtain the recalculated value of the total module loss. Charging modules whose total recalculated loss value is lower than the preset safe loss limit are filtered out and sorted in ascending order based on the size of the total recalculated loss value to obtain the true module loss sequence.

4. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 3, characterized in that, The methods for splitting the type of the output request to be executed include: Identify the target output power command in the output request to be executed, and extract the low-frequency power component and high-frequency power component of the target output power command within a preset time window; The number of steady-state modules is obtained by calculating the ratio of the low-frequency power component to the rated power of the corresponding charging module. The average power allocation value of the low-frequency power component in each steady-state module is calculated. The average power allocation value is encapsulated with the constraint condition of reducing the duty cycle adjustment frequency to generate the basic power supply task. Frequency domain analysis is performed on the high-frequency power components to output the fluctuation frequency characteristics corresponding to the dynamic compensation requirements; the amplitude of the high-frequency power components is bound to the fluctuation frequency characteristics, and driving conditions to improve the switching response speed are added to generate a switch-intensive compensation subtask. The power demand nodes corresponding to the basic power supply subtask and the frequency response nodes corresponding to the switch-dense compensation subtask are time-aligned and combined to obtain the subtask distribution demand set.

5. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 4, characterized in that, The methods for evaluating the true thermal recovery state inside the SGT MOSFET switch include: Locate the shutdown time of each charging module after it last completed a historical high-frequency task. At the same time, extract the ambient temperature and heat sink surface temperature from the shutdown time to the current time from the operation feature data, and plot them as a temperature decay curve to obtain cooling process information. Calculate the real-time temperature difference between the heat sink surface temperature and the ambient temperature at the current moment in the temperature decay curve, as well as the temperature drop gradient of the heat sink surface at the current moment. Combine this with the preset chip thermal resistance parameters to estimate the internal junction temperature at the current moment. The residual thermal coefficient is obtained by statistically analyzing the duration of continuous high-frequency switching and the average commutation frequency of the corresponding charging module during the most recent execution of a historical high-frequency task, and then multiplying them. The time interval from the shutdown moment to the current moment is calculated as the cooling interval. The ratio of the residual heat coefficient to the cooling interval is calculated to obtain the residual heat factor. The residual heat factor is multiplied by the preset heat mapping coefficient to obtain the residual heat compensation temperature. The real-time core junction temperature is obtained by summing the residual heat compensation temperature and the estimated internal junction temperature, and this real-time core junction temperature is taken as the true thermal recovery state.

6. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 5, characterized in that, The method of intercepting pseudo-usage modules based on the actual thermal recovery state and filtering callable modules by combining the interception results and the actual module loss sequence includes: The thermal deviation is obtained by calculating the temperature difference between the heat sink surface temperature of each charging module at the current moment and the real-time core junction temperature corresponding to the thermal recovery state. If the surface temperature of the heat sink is lower than the preset safe temperature threshold at the current moment, and the corresponding thermal deviation is higher than the preset internal thermal safety threshold, then the corresponding charging module will be identified as a fake usage module, and a scheduling lock interception command will be generated to execute on the fake usage module. The remaining charging modules, except for the pseudo-use modules, are marked as available candidate modules, and the sorting number of each available candidate module is identified by mapping it to the real module loss sequence. All available candidate modules are sorted in ascending order of their serial numbers. Available candidate modules belonging to the first category of preset number ranges are assigned to the switch intensive compensation subtask and used as a high-priority calling subset. Available candidate modules belonging to the second category of preset number ranges are assigned to the basic power supply task and used as a low-priority calling subset. Standby rotation instructions are sent to the remaining available candidate modules and assigned to the forced rotation subset. Integrate the high-priority call subset and the low-priority call subset into a set of callable modules.

7. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 6, characterized in that, The method for constructing candidate scheduling schemes based on the set of callable modules and the subtask distribution requirement set includes: Extract the power demand node sequence corresponding to the basic power supply task and the frequency response node sequence corresponding to the switch-dense compensation subtask. Traverse the high-priority call subset and combine its modules into a high-frequency response array; traverse the low-priority call subset and combine its modules into a basic conduction array; cross-match the high-frequency response array and the basic conduction array to generate a preliminary scheduling array group; The system compiles a list of currently running online modules and compares it with the time sequence of each array in the initial scheduling array group. If a module is marked as being called at any time but is not in the list of online modules, then that time is marked as the entry node for the corresponding module. If a module is marked as being exited at any time but is in the list of online modules, then that time is marked as the exit node for the corresponding module. If several charging modules simultaneously trigger the cut-in node or cut-out node within the same preset very short time window, it is determined that the corresponding array has a conflict, and the array is removed from the initial scheduling array group; the output of each remaining array group is used as a candidate scheduling scheme.

8. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 7, characterized in that, The method of separating the handover transition interval and the stable power supply interval for each group of candidate scheduling schemes, and predicting the switching loss of each group of candidate scheduling schemes in combination with operational characteristic data, includes: Obtain the preset dead time of the SGT MOSFET switch, extract the entry node and exit node in each candidate scheduling scheme, and extend the time according to the preset dead time with each node as the center to obtain the handover transition interval, while taking the remaining interval as the stable power supply interval. Within a stable power supply range, the expected operating frequency and expected effective current in the operational characteristic data corresponding to the candidate scheduling schemes are statistically analyzed, and the steady-state conduction loss is calculated in combination with the preset conduction internal resistance. At the same time, the steady-state switching loss is calculated by multiplying the unit switching energy loss corresponding to the expected effective current with the expected operating frequency. The steady-state conduction loss and steady-state switching loss are summed to obtain the estimated steady-state switching loss for the stable power supply range. Within the handover transition interval, the voltage fluctuation amplitude and surge current peak of the corresponding charging module at the moment of switching in or out are extracted, and the instantaneous energy loss estimate is obtained by integration according to the preset switching time; at the same time, the product of the voltage fluctuation amplitude and the recovery charge of the body diode is calculated to obtain the recovery energy loss estimate; the instantaneous energy loss estimate and the recovery energy loss estimate are summed to obtain the switching impact loss estimate of the handover transition interval. The weighted sum of the estimated switching impact losses of all handover transition sections and the estimated steady-state switching losses of all stable power supply sections in the same candidate scheduling scheme is used to obtain the global loss assessment result of the candidate scheduling scheme.

9. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 8, characterized in that, The method for determining the target scheduling scheme based on the global loss assessment results includes: All candidate scheduling schemes are sorted in ascending order according to the numerical value of the global loss assessment results. The candidate scheduling scheme at the top is selected, and the corresponding set of callable modules is checked to see if it is online. If it is online, the candidate scheduling scheme is determined as the target scheduling scheme; otherwise, the next candidate scheduling scheme is checked until the corresponding set of callable modules is determined to be online.

10. The SGT MOSFET switching loss optimization AI scheduling management system according to claim 9, characterized in that, The method of generating module control instructions that match the target scheduling scheme and using these module control instructions to perform coordinated management and control of the corresponding charging modules includes: For the switching-in and switching-out times of each charging module in the target scheduling scheme, generate timing pulse commands to execute grid connection or grid disconnection actions; Based on the basic power supply task or switch-intensive compensation subtask allocated by the target scheduling scheme, drive configuration instructions are generated to adjust the switching resistance and dead time of the SGT MOSFET.

Images