An all-solid-state power supply large-scale switch intelligent modulation optimization method and device

Abstract

This invention discloses a method and device for large-scale intelligent modulation optimization of all-solid-state power supply switching, belonging to the field of power electronics control technology. The method includes: generating an initial switching timing matrix using the NLM algorithm for any waveform selected by the user; extracting effective action events from the initial switching timing matrix, and obtaining an optimized switching timing matrix by dynamically weighting and uniformly sorting the events with the goal of minimizing the standard deviation of the number of switching actions; the power supply outputting according to the optimized switching timing matrix and acquiring switching parameters in real time; and executing optimization strategies hierarchically according to criteria, repeating the above process until all switch states reach equilibrium. The optimization strategy includes analyzing and locating the abnormal state of a switch when the parameters of a local switch exceed a set threshold, adjusting the number of switch actions according to the migration coefficient criterion, and updating the optimized switching timing matrix. This ensures that all switches of the power supply operate in the optimal state, fully utilizing device performance and guaranteeing stable and reliable operation of the power supply.

Description

Technical Field

[0001] This invention belongs to the field of power electronic control technology, and more specifically, relates to a method and device for large-scale intelligent modulation optimization of all-solid-state power supply switching. Background Technology

[0002] The switching modulation technology of high-voltage pulse power supplies must simultaneously meet the dual requirements of accurate waveform generation and reliable switching operation. In existing research, the conversion from waveform to switching timing can be achieved through the Nearest Level 1 Modulation (NLM) algorithm, but it suffers from the drawback of extremely uneven distribution of switching actions: the step-by-step reduction logic of the traditional NLM algorithm causes more than 80% of the actions to be concentrated in a few switching modules, and the overheating of local modules due to high-frequency actions reduces the switching life to a certain extent.

[0003] Existing optimization solutions often focus on a single dimension: for example, reducing the number of switches through topology modification, but failing to address the fundamental issue of switch optimization action strategies; or only introducing temperature monitoring without a timely feedback mechanism, resulting in delays in switch modulation; and the limited number of related factors leads to limited modulation optimization effects, while introducing switch electrical parameter monitoring is difficult to cover power supplies with a large number of switch modules in engineering.

[0004] Therefore, there is an urgent need for a closed-loop modulation technology that can collaboratively achieve algorithm optimization for uniform action, multi-dimensional feedback to balance switching states, and reliability data iterative optimization strategies, in order to break through the bottleneck of existing control technology for large-scale switching of power electronic power supplies. Summary of the Invention

[0005] To address the shortcomings of related technologies, the present invention aims to provide a method and apparatus for large-scale intelligent modulation optimization of all-solid-state power supplies, which is intended to solve the problems of uneven operation frequency and operation number of multiple switching modules in a power electronic power supply for output target waveform, resulting in local module overheating or local switch damage.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for large-scale switching intelligent modulation optimization of all-solid-state power supplies, comprising:

[0007] S100. For any waveform selected by the user, the NLM algorithm is used to generate the initial switching timing matrix. ;

[0008] S200, Extract the initial switching timing matrix. Valid action events, expressed as the standard deviation of the number of switching actions. With the goal of minimizing the time series, an optimized switching timing matrix is ​​obtained through dynamic weight uniform sorting. The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions.

[0009] S300, the power supply follows the optimized switching timing matrix. Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance;

[0010] The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is judged to have reached the damage condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the damage condition, the power supply automatically terminates the output.

[0011] Optionally, the dynamic weights The expression is:

[0012] ,

[0013] in, For switch module The historical cumulative number of switch actions, for each assigned event Then, the weight of the switch module is updated in real time. In the event of multiple switch modules having the same priority in a single output allocation, the switch event will be preferentially allocated to the switch module with the highest current weight to balance the total number of switch actions.

[0014] Optionally, the extraction of the initial switching timing matrix Valid action events, expressed as the standard deviation of the number of switching actions. With the goal of minimizing the time series, an optimized switching timing matrix is ​​obtained through dynamic weight uniform sorting. ,include:

[0015] Extract the initial switching timing matrix Effective action events form a time series. time series Divide equally to Level switch module, switch module Total number of events allocated As shown in the following formula:

[0016] ,

[0017] ,

[0018] in, For each event, the total number of switching actions based on the NLM algorithm is [number]. This corresponds to a single switch action; For floor function, when When divisible, ;when ,Will Assigned according to dynamic weights Level switch module, For even-numbered components, when When it is even, ,when When the number is odd, for high dynamic weights ,on the contrary .

[0019] Optionally, step S300 specifically includes:

[0020] S310, The power supply follows the optimized switching timing matrix. The output is used to acquire the electrical parameters of the switch in real time via sensors and signal acquisition systems. The host computer reads the event interval between two actions of all switching modules in real time. The thermal status parameters of all switching modules are collected in real time using digital temperature sensors. ; wherein, the electrical parameters This includes the switching operating voltage, on-state voltage drop, turn-on current, and trigger voltage;

[0021] S320, Set rated values ​​for electrical stress parameters Equivalent frequency rating and temperature rating As shown in the following formula:

[0022] ,

[0023] ,

[0024] Among them, switching to burst frequency ;

[0025] S330, Based on the rated value of the electrical stress parameter Equivalent frequency rating and temperature rating Given threshold values ​​for state parameters, optimization is performed hierarchically based on state criteria:

[0026] ,

[0027] ,

[0028] ,

[0029] in, This indicates the threshold value for electrical stress parameters. This indicates that the equivalent frequency is set to a threshold. Indicates the temperature setting threshold;

[0030] S340. Based on the criteria and parameter settings, thresholds are set for hierarchical optimization. The above process is repeated until all switch states reach a balance.

[0031] Optionally, step S340 specifically includes:

[0032] Level 1, all state variables All are less than the set threshold :

[0033] ,

[0034] Then the switching timing matrix remains unchanged; where, The value ranges from 1 to 3. When taking different values, Indicate electrical parameters Event interval of the action or thermal state parameters ;

[0035] At the second level, when no more than 50% of the state variables of a local switch exceed a set threshold, the abnormal switch module is analyzed and located, and the migration coefficient is used as a criterion. Reduce the number of times the switch operates, and adjust the reduced number of operations to the normal switching module to obtain the state feedback optimization matrix. Optimize the matrix based on state feedback Update and optimize the switching timing matrix Controls the switching operation of the power supply; the updated switching module action is represented as follows:

[0036] ,

[0037] ,

[0038] ,

[0039] in, The total number of events allocated to the switch module after feedback optimization based on switch status monitoring is indicated by the superscript, where 1 indicates algorithm timing optimization, 2 indicates status detection optimization, and the subscript indicates that the number of switch actions may be different for different parameter transitions. The total number of events to be migrated to other modules. The migration coefficient; For electrical parameter migration coefficient, For frequency parameter mobility coefficients, The temperature parameter migration coefficient; the local switch refers to a quantity less than 25% of the total number of switch modules;

[0040] Level 3: When more than 50% of the state variables of a local switch are greater than the set threshold, the switch module is judged to have reached the damage condition, the switch module is stopped, a switch damage warning is issued, all action events are migrated to the normal module, the NLM algorithm is re-executed, and the execution of S100 is returned.

[0041] Level 4: When more than 25% of the switching modules trigger a switch failure warning, the power supply will automatically stop outputting power.

[0042] Among them, when performing migration action events, priority is given to assigning them to the current one. The highest level of normal module; if multiple abnormal states occur in the second level of optimization, the temperature criterion will be used sequentially. Frequency criterion Electrical criteria The migration strategy is executed sequentially.

[0043] Optional, also includes:

[0044] S400. After the power supply is interrupted, record the switching history data of each round of power output according to different power interruption types; based on the historical data, use a neural network algorithm to obtain the data influence factors of the switching module, optimize the criterion, and optimize the criterion when the power supply is started next time.

[0045] Among them, the different power supply interruption types include: after the power supply outputs for one round, three shutdown types are defined and cycled according to the switching state;

[0046] The shutdown types include:

[0047] The first type is normal exit; the power supply completes all output tasks in this round normally and updates historical data.

[0048] The second category is automatic termination and exit; when the power supply reaches the fourth level of optimization conditions, a warning is issued that more than 25% of the switching modules are disabled, and the weight of the corresponding switching modules is increased when updating historical data, thus... ;

[0049] The third category is damage-based exit, where the device malfunctions, there is no power output or a power level is missing, and the boundary conditions for the next action are set based on historical data of this category.

[0050] Optionally, the historical switching data for each round of power output recording includes:

[0051] For each round of power supply output, record the total number of switching actions of all switches. The number of times the master switch was activated. Including: the temperature corresponding to each switching action. The burst frequency of the switch Electrical parameters of the switch ;

[0052] ,

[0053] ,

[0054] ,

[0055] ,

[0056] ,

[0057] ,

[0058] in, This is the maximum temperature when the switch is activated. For the maximum burst frequency, Refers to the maximum value of electrical parameters; This represents the average burst frequency. The average temperature during switch operation. Similarly, it is the average value of multiple electrical parameters.

[0059] Optionally, the step of obtaining various data influence factors of the switching module based on the historical data using a neural network algorithm includes:

[0060] The historical data is imported into a neural network algorithm to obtain the switch health coefficient. as well as Various data influence factors ;

[0061] set up The criterion corresponding to the third-level optimization condition is based on the influence factor. Adjusting the priority of the aforementioned criteria; Influence factor as follows:

[0062] ,

[0063] in, The initial values ​​of all influencing factors are set to zero, corresponding to the first round of power supply output, and the criterion is that there is no historical data for adjustment.

[0064] Optionally, the criterion can be optimized by including:

[0065] based on The predictive model adjusts the set threshold coefficient in the power output for each switch in the next cycle. and data weights , The deep learning objective is to provide early warnings based on minimizing power supply switching during operation.

[0066] ,

[0067] in, This indicates the proportion of warning switches in one round of output. The optimization coefficients are automatically updated in each round.

[0068] Secondly, the present invention also provides a large-scale switching intelligent modulation optimization device for all-solid-state power supplies, comprising:

[0069] The initialization module is used to generate an initial switching timing matrix for any waveform selected by the user using the NLM algorithm. ;

[0070] The dynamic optimization module is used to extract the initial switching timing matrix. Valid action events, expressed as the standard deviation of the number of switching actions. With the goal of minimizing the time series, an optimized switching timing matrix is ​​obtained through dynamic weight uniform sorting. The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions.

[0071] The hierarchical control module is used to control the power supply according to the optimized switching timing matrix. Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance;

[0072] The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is judged to have reached the damage condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the damage condition, the power supply automatically terminates the output.

[0073] Compared with the prior art, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0074] 1. This invention provides a method for large-scale switching intelligent modulation optimization of all-solid-state power supplies, which optimizes the initial switching timing matrix generated by NLM by setting dynamic weights. The algorithm optimization time matrix is ​​obtained. Power button After output begins, the hardware module collects monitorable related data in real time, including operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature, and continuous switching interval. When the parameters of a local switch exceed the set threshold, the host computer analyzes and locates the switch in the abnormal state, and determines the migration coefficient based on the criterion. By reducing the number of switching actions and adjusting these actions accordingly to ensure the switches are functioning well, a state feedback optimization matrix can be obtained. ,Will The waveform is transmitted to the power control module, and the above process is repeated until the states of all switches reach equilibrium. This ensures that all switches in the power electronic power supply operate in their optimal state, making it suitable for large-scale switching power electronic power supplies that output arbitrary waveforms.

[0075] 2. This invention provides a large-scale intelligent modulation optimization method for all-solid-state power supplies. A historical switching database is established on a host computer. Based on different power supply termination types, the historical switching data for each round of power supply output is processed and recorded. The criteria for optimizing the status monitoring feedback are then used during the next startup. A four-level timing optimization strategy is executed based on these criteria, achieving a closed-loop optimization process of prediction, adjustment, early warning, and shutdown. Through multiple rounds of automatic iterative optimization, all switches in the power electronic power supply are ensured to operate in their optimal state, thereby fully utilizing device performance and ensuring stable and reliable power supply operation.

[0076] 3. The present invention provides a method for large-scale switching intelligent modulation optimization of all-solid-state power supplies, which predetermines the circuit topology. This approach involves real-time monitoring of individual switches to achieve predictive sparse sensor status monitoring. This solves the problems of high difficulty and cost associated with large-scale monitoring of switch electrical parameters. Based on switch health coefficients provided by a historical state optimization model, the system dynamically selects switches for monitoring electrical parameters during a training test or in the previous power output cycle, iteratively setting thresholds to optimize switches under multi-dimensional status monitoring feedback. This solution simplifies switch data, enabling intelligent algorithms to obtain relatively accurate prediction results even with a smaller training set. The method can obtain the influence factors of state parameters and failure probability for any switch model, facilitating the investigation of the impact of peak and mean values ​​on the switch. Based on different power outage types, historical switch data for each power output cycle is recorded. Health coefficients are calculated based on the switch health coefficients or failure probabilities of the historical data weighted by the shutdown type, optimizing the criteria for the next cycle. This optimization is specific to individual switch differences, making it more targeted. Attached Figure Description

[0077] Figure 1 This is a flowchart illustrating a large-scale switching intelligent modulation optimization method for all-solid-state power supplies provided in an embodiment of the present invention.

[0078] Figure 2 The rectangular oscillation waveform and the action heatmap of the three-time output of the switch based on the NLM algorithm are provided for embodiments of the present invention.

[0079] Figure 3 The image shows the action heatmap of the three outputs after the algorithm optimization under the rectangular oscillation waveform provided in the embodiment of the present invention, and the comparison diagram before and after optimization.

[0080] Figure 4 A schematic diagram of the switch status monitoring experimental debugging platform provided for an example of the present invention.

[0081] Figure 5 This is a comparison chart showing the process before and after the first round of iterative optimization based on switch state monitoring feedback, provided for an embodiment of the present invention.

[0082] Figure 6 The present invention provides a parameter correlation heatmap based on historical data and an influence factor calculation result based on a neural network algorithm; wherein, (a) is a switch parameter correlation heatmap, and (b) is... The calculation results of the influence factors of the prediction model are shown in the figure.

[0083] Figure 7 A comparison diagram of important switch states before and after three rounds of optimization criteria cyclic optimization provided in this embodiment of the invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0085] The following description, in conjunction with a preferred embodiment, illustrates the content involved in the above embodiments.

[0086] Example 1

[0087] refer to Figure 1 This invention provides a method for large-scale switching intelligent modulation optimization of all-solid-state power supplies, comprising:

[0088] S100. For any waveform selected by the user, the NLM algorithm is used to generate the initial switching timing matrix. ;

[0089] S200, Extract the initial switching timing matrix. Valid action events, expressed as the standard deviation of the number of switching actions. With the goal of minimizing the time series, an optimized switching timing matrix is ​​obtained through dynamic weight uniform sorting. The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions.

[0090] S300, the power supply follows the optimized switching timing matrix. Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance;

[0091] The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is judged to have reached the damage condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the damage condition, the power supply automatically terminates the output.

[0092] In this embodiment, an arbitrary waveform generator is provided to output a rectangular oscillating wave with an output voltage of 16kV and a pulse width of 10µs, such as... Figure 2 As shown in (a) of the diagram. The initial switching timing matrix is ​​generated using the traditional NLM algorithm. For ease of comparison, a rectangular heatmap (with time on the horizontal axis) is used instead of the timing matrix. The switching timing heatmap for three consecutive outputs at 1Hz is given below. Figure 2 (b) Figure 2 (c) Figure 2 As shown in (d), it can be observed that the level transition of the output oscillation section based on the sequential logic of the traditional NLM algorithm is only handled by the last 5 stages of switches. The unevenness of the operation of each switch module is relatively high, and the standard deviation of the number of switch operations is as follows: .

[0093] While maintaining waveform accuracy, an optimized switching timing matrix is ​​obtained through dynamic weighted uniform sorting. At this time, the standard deviation of the number of switching actions At the minimum, the host computer transmits data via serial port. To the power control module.

[0094] The dynamic weight As shown in the following formula:

[0095] ,

[0096] in, For module The historical cumulative number of switch actions, each time a switch action event is assigned. Then, update the module's weight in real time. In the event of multiple switch modules having the same priority in a single output allocation, the switch event will be preferentially allocated to the switch module with the highest current weight to balance the total number of switch actions.

[0097] extract Effective action events form a time series. time series Divide equally to Level switch module, module Total number of events allocated As shown in the following formula:

[0098] ,

[0099] ,

[0100] in, The total number of switching actions for each rectangular oscillating wave output by the power supply, per event. This corresponds to one switching action. A switched-on switch must eventually be switched off; therefore, the number of switching actions must be even. To ensure waveform fitting accuracy, as many switches as possible should be used. Here, it is assumed that the voltage rise segment in the waveform is caused by a switched-on event generated by a previously switched-off switch. In summary, switched-on and switched-off events always occur in pairs. This is the floor function. For the remainder, when When divisible, ;when Not divisible by any other number, and with a remainder of 0. At that time, Assigned according to dynamic weights Level switch module; For even-numbered components, when When it is even, ,when When the number is odd, for high dynamic weights ,on the contrary The total number of events satisfies .

[0101] In this embodiment, the algorithm-optimized timing matrix for three consecutive outputs is compared with the traditional NLM timing matrix. Figure 3 As shown, the uniformity of switching action is significantly improved. However, there are differences between switches, and from a physical perspective, the number of operations is not the most critical factor affecting the lifespan of a switch. Therefore, it is necessary to monitor and optimize the working status of the switches.

[0102] Feedback optimization is performed based on switch operation status monitoring, and switch parameters are collected and monitored in real time. When 50% of the switch parameters of a local switch exceed a set threshold, the host computer analyzes and locates the switch in the abnormal state, and determines the appropriate switch based on the migration coefficient criterion. By reducing the number of switching actions and adjusting these actions accordingly to ensure the switches are functioning well, a state feedback optimization matrix can be obtained. ,Will The waveform is transmitted to the power control module and the above process is repeated until the states of each switch reach equilibrium.

[0103] See Figure 4 This is a schematic diagram of a state monitoring and debugging platform for an all-solid-state power supply provided in this embodiment. A high-voltage probe and an oscilloscope read the electrical parameters of the switch in real time. Specifically, this includes: switching operating voltage, on-state voltage drop, turn-on current, and trigger voltage (for simplicity, the formulas will be consistently represented by...). (This indicates that the host computer reads the event interval between two actions of all switching modules in real time.) Switch to burst frequency This indicates that the digital temperature sensor collects the thermal status parameters of all switching modules in real time. , here This only needs to represent the relative temperature between different switching modules; no specific physical meaning is required. Refer to the technical manual to set the rated values ​​for electrical stress parameters. Set the equivalent frequency rating. and temperature rating As shown in the following formula:

[0104] ,

[0105] ,

[0106] Based on the rated values, the threshold values ​​for the state parameters are set as follows, and optimization is performed hierarchically according to the state criteria:

[0107] ,

[0108] ,

[0109] ,

[0110] in, This indicates the threshold value for electrical stress parameters. This indicates that the equivalent frequency is set to a threshold. This indicates the temperature setting threshold.

[0111] For ease of calculation, the first round of power output is taken as... Subsequent values ​​are optimized based on data feedback. Thresholds are set according to criteria and parameters for tiered optimization, and the above process is repeated until all switch states reach equilibrium.

[0112] The hierarchical optimization performed based on the set threshold is as follows:

[0113] Level 1, all state variables All are less than the set threshold :

[0114] ,

[0115] Then the switching timing matrix remains unchanged; where, The value ranges from 1 to 3. When taking different values, Indicate electrical parameters Event interval of the action or thermal state parameters ;

[0116] At the second level, when no more than 50% of the state variables of a local switch exceed a set threshold, the abnormal switch module is analyzed and located, and the migration coefficient is used as a criterion. Reduce the number of times the switch operates, and adjust the reduced number of operations to the normal switching module to obtain the state feedback optimization matrix. Optimize the matrix based on state feedback Update and optimize the switching timing matrix Controls the switching operation of the power supply; the updated switching module action is represented as follows:

[0117] ,

[0118] ,

[0119] ,

[0120] in, The total number of events allocated to the switch module after feedback optimization based on switch status monitoring is indicated by the superscript, where 1 indicates algorithm timing optimization, 2 indicates status detection optimization, and the subscript indicates that the number of switch actions may be different for different parameter transitions. The total number of events to be migrated to other modules. The migration coefficient; For electrical parameter migration coefficient, For frequency parameter mobility coefficients, The temperature parameter migration coefficient; the local switch refers to a quantity less than 25% of the total number of switch modules;

[0121] Level 3: When more than 50% of the state variables of a local switch are greater than the set threshold, the switch module is judged to have reached the damage condition, the switch module is stopped, a switch damage warning is issued, all action events are migrated to the normal module, the NLM algorithm is re-executed, and the execution of S100 is returned.

[0122] Level 4: When more than 25% of the switching modules trigger a switch failure warning, the power supply will automatically stop outputting power.

[0123] Among them, when performing migration action events, priority is given to assigning them to the current one. The highest level of normal module; if multiple abnormal states occur in the second level of optimization, the temperature criterion will be used sequentially. Frequency criterion Electrical criteria The migration strategy is executed sequentially.

[0124] Because monitoring all electrical stress parameters in engineering is too costly and complex, this invention proposes a sparse sensing feedback based on predictive maintenance during the real-time acquisition of switch parameters. Specifically, the monitoring of electrical stress parameters is dynamically selected, initially predetermined based on the circuit topology. The switches that need to be monitored should be selected, for example, those with the highest collector and emitter overvoltages in the circuit topology. After establishing a model based on historical state data, the switches to be monitored in real time are selected according to the model. A switch.

[0125] See Figure 5 In this embodiment, when the arbitrary waveform pulse generator outputs a rectangular oscillating wave at 10Hz for 1000 cycles in the first round, without considering electrical parameters, the peak frequency, peak temperature, and average temperature of each switch before and after state monitoring are recorded. and average frequency The calculation comparison results show that the average temperature and average frequency The definition is as follows ( ):

[0126] ,

[0127] ,

[0128] It is evident that after incorporating temperature and frequency operating status feedback, the uniformity of state variables among the switches has significantly improved. Although electrical parameters were not measured in the first round of testing, based on the switch turn-off loss formula, electrical parameters and frequency jointly affect temperature. Similarly, it can be inferred that the uniformity of electrical parameters will also significantly improve. Comparing the three optimization results, it is clear that algorithm optimization only targets the number of switch actions. While it has some effect on temperature and frequency control, the effect is poor. Therefore, state monitoring and control are essential.

[0129] The threshold coefficients or migration coefficients of various criteria in condition monitoring significantly affect power supply operation. Higher threshold coefficients may lead to higher peak temperatures in some switches, reducing their lifespan, while lower anomaly criteria may cause most switches to fail simultaneously, resulting in warning failure, significantly reducing the accuracy of the fitted waveform, or prematurely entering the third criterion level (power supply automatic shutdown). Therefore, it is necessary to introduce a criterion optimization loop based on historical data, using the failure probability to characterize the switch state and its correspondence with the graded criteria, thereby continuously optimizing the criteria until the optimal coefficient set for the power supply is reached. Simultaneously, to improve the switching modulation strategy system, it is also necessary to establish a model of historical state data to achieve sparse sensor feedback for predictive maintenance and improve the measurement of electrical parameters in condition monitoring.

[0130] For example, during the next power output cycle, the neural network model based on the historical operating status data of the switch established from 1000 power output cycles in this cycle can update the various coefficients and feedback strategies (criteria priority and classification standards) of the aforementioned status monitoring, and design threshold coefficients specifically for different switch models. , , and migration coefficient .

[0131] Optionally, following the S300, the following may also be included:

[0132] S400. After the power supply is interrupted, record the switching history data of each round of power output according to different power interruption types; based on the historical data, use a neural network algorithm to obtain the data influence factors of the switching module, optimize the criterion, and optimize the criterion when the power supply is started next time.

[0133] Among them, the different power supply interruption types include: after the power supply outputs for one round, three shutdown types are defined and cycled according to the switching state;

[0134] The shutdown types include:

[0135] The first type is normal exit; the power supply completes all output tasks in this round normally and updates historical data (basic weights). );

[0136] The second category is automatic termination and exit; when the power supply reaches the fourth level of optimization conditions, a warning is issued that more than 25% of the switching modules are disabled, and the weight of the corresponding switching modules is increased when updating historical data, thus... ;

[0137] The third category is damage-based exit, where the device malfunctions, there is no power output or a power level is missing, and the boundary conditions for the next action are set based on historical data of this category.

[0138] After each output cycle, the power supply exits the operating cycle from any stop type.

[0139] For each round of power supply output, record the total number of switching actions of all switches. The temperature corresponding to each switching action The burst frequency of the switch Electrical parameters of the switch Electrical parameters do not need to be considered if they are not recorded.

[0140] The recorded historical data is as follows:

[0141] ,

[0142] ,

[0143] ,

[0144] ,

[0145] ,

[0146] ,

[0147] in, This is the maximum temperature when the switch is activated. For the maximum burst frequency, This refers to the maximum value of each electrical parameter; This represents the average burst frequency. The average temperature during switch operation. Similarly, it is the average value of multiple electrical parameters.

[0148] Optionally, the step of obtaining various data influence factors of the switching module based on the historical data using a neural network algorithm includes:

[0149] The historical data is imported into a neural network algorithm to obtain the switch health coefficient (probability of damage). as well as Various data influence factors ;

[0150] set up The criterion corresponding to the third-level optimization condition is based on the influence factor. Adjusting the priority of the aforementioned criteria; Influence factor as follows:

[0151] ,

[0152] in, The initial values ​​of all influencing factors are set to zero, corresponding to the first round of power supply output, and the criterion is that there is no historical data for adjustment.

[0153] Furthermore, the criterion is optimized, including:

[0154] based on The predictive model adjusts the set threshold coefficient in the power output for each switch in the next cycle. and data weights , The deep learning objective is to provide early warnings based on minimizing power supply switching during operation.

[0155] ,

[0156] in, This indicates the proportion of warning switches in one round of output. The optimization coefficients are automatically updated in each round.

[0157] In this embodiment, four low-rated (vulnerable) switches (switch model: AIKQ120N75CP2XKSA1, rated value 750V, output below 1kV) are used. Replace the four working switches and monitor them. The electrical stress parameters measured by this method will have a larger influence factor, but it is easier to obtain experimental results for different power supply interruption types. Switching timing based on the NLM algorithm... Locate the faulty switch. After three switches fail, perform Pearson correlation calculations based on historical data processed by the host computer. Figure 6 As shown in (a) and The results of the priority classification calculation of the impact factor criteria are as follows: Figure 6 As shown in (b), it can be seen that in the test method provided in this embodiment, the peak value has a greater impact on the switch life than the average value, and the temperature has a greater impact on the switch life than the frequency and electrical parameters. This is because, theoretically, temperature depends on the frequency and electrical parameters, which can be equivalent to three different effects on the switch.

[0158] See Figure 7 This study describes an arbitrary waveform pulse generator that outputs a rectangular oscillation wave of 1000 times at 10Hz. Through a comparison of state feedback optimization during three rounds of power supply operation, and after three rounds of criterion optimization, the power supply state parameters in each output cycle are closer to the expected values. Figure 6 We can extract the peak parameters with larger influencing factors as the analysis objects. We can see that the peak temperature of each switch approaches the expected optimized temperature by 1℃ per round, and the peak frequency also optimizes with each round. The difference in peak frequency is generally because the initial migration coefficient value is set too low in the first round, causing some high-frequency module migration events to fail. This excessively high frequency contributes significantly to increasing the frequency migration coefficient and decreasing the frequency threshold coefficient in historical data. Therefore, in the next round, the high-frequency, unmigrated action events at this point will successfully migrate, thus reducing the peak frequency. Simultaneously, we can see that the number of switch actions does not follow a clear pattern in each round of optimization. This is because temperature and frequency have higher priority, and the number of actions is controlled by the former. Figure 7 It is evident that after three rounds of optimization based on historical data, the target effect of switching modulation has been largely achieved, proving the high efficiency and feasibility of the scheme.

[0159] Compared with traditional switching control strategies, the switching modulation strategy of the present invention can improve the lifespan of pulse power switches by about 1000 cycles. Compared with traditional switching control strategies, the switching modulation strategy of the present invention can increase the uniformity of the number of switching operations by 100%-300%, reduce the maximum burst frequency of the switch by 50%-500%, and reduce the average temperature of the switch by 10%-15%. Moreover, the more switching modules there are, the better the uniformity effect.

[0160] This invention provides a method for large-scale switching intelligent modulation optimization of all-solid-state power supplies, which optimizes the initial switching timing matrix generated by NLM by setting dynamic weights. The algorithm optimization time matrix is ​​obtained. Power button After output begins, the hardware module collects monitorable related data in real time, including operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature, and continuous switching interval. Based on criteria, a four-level timing optimization strategy is executed to achieve an optimized closed loop of prediction, adjustment, early warning, and stop. When the parameters of a local switch exceed a set threshold, the host computer analyzes and locates the switch in the abnormal state, and determines the appropriate action based on the migration coefficient criterion. By reducing the number of switching actions and adjusting these actions accordingly to ensure the switches are functioning well, a state feedback optimization matrix can be obtained. ,Will The waveform is transmitted to the power control module, and the above process is repeated until the states of all switches reach equilibrium. A switch history database is established on the host computer. Based on different power interruption types, the historical switch data of each round of power output is processed and recorded, and the criteria for state monitoring feedback are optimized during the next startup. Through multiple rounds of automatic iterative optimization, all switches of the power electronic power supply are ensured to operate in the optimal state, thereby fully utilizing device performance and ensuring stable and reliable power supply operation. This method is suitable for large-scale switching power electronic power supplies that output arbitrary waveforms.

[0161] Example 2

[0162] This invention also provides a large-scale switching intelligent modulation optimization device for all-solid-state power supplies, comprising:

[0163] The initialization module is used to generate an initial switching timing matrix for any waveform selected by the user using the NLM algorithm. ;

[0164] The dynamic optimization module is used to extract the initial switching timing matrix. Valid action events, expressed as the standard deviation of the number of switching actions. With the goal of minimizing the time series, an optimized switching timing matrix is ​​obtained through dynamic weight uniform sorting. The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions.

[0165] The hierarchical control module is used to control the power supply according to the optimized switching timing matrix. Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance;

[0166] The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is judged to have reached the damage condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the damage condition, the power supply automatically terminates the output.

[0167] The all-solid-state power supply large-scale switching intelligent modulation optimization device provided in this embodiment of the invention is used to execute the all-solid-state power supply large-scale switching intelligent modulation optimization method provided in Embodiment 1, and has the same beneficial effects.

[0168] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent modulation optimization of a large-scale switch of an all-solid-state power supply, characterized in that, include: S100. For any waveform selected by the user, the NLM algorithm is used to generate the initial switching timing matrix. ; S200, extracting the initial switching timing matrix effective action events, forming a time series , the time series evenly divided level switching module, switching module total number of events assigned as follows: ， ， wherein, is the total number of switching actions based on the NLM algorithm, each event corresponds to one switching action; is the floor function, when is an integer division, ; when , the is assigned to the level switching module according to the dynamic weight, is the even sub-item, when is even, , when is odd, for the high dynamic weight , and vice versa ; The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions. S300, power supply according to optimized switching timing matrix Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance; The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is determined to have reached the failure condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the failure condition, the power supply automatically terminates the output. Specifically, when no more than 50% of the state variables of a local switch exceed a set threshold, the abnormal switch module is analyzed and located, and the migration coefficient is used as a criterion. Reduce the number of times the switch operates, and adjust the reduced number of operations to the normal switching module to obtain the state feedback optimization matrix. Optimize the matrix based on state feedback Update and optimize the switching timing matrix Controls the switching operation of the power supply; the updated switching module action is represented as follows: ， ， ， in, The total number of events allocated to the switch module after feedback optimization based on switch status monitoring is indicated by the superscript, where 1 indicates algorithm timing optimization, 2 indicates status detection optimization, and the subscript indicates that the number of switch actions may be different for different parameter transitions. The total number of events to be migrated to other modules. The migration coefficient; For electrical parameter migration coefficient, For frequency parameter mobility coefficients, The temperature parameter migration coefficient; the local switch is less than 25% of the total number of switch modules.

2. The method as described in claim 1, characterized in that, The dynamic weight The expression is: ， in, For switch module The historical cumulative number of switch actions, for each assigned event Then, the weight of the switch module is updated in real time. In the event of multiple switch modules having the same priority in a single output allocation, the switch event will be preferentially allocated to the switch module with the highest current weight to balance the total number of switch actions.

3. The method as described in claim 1, characterized in that, Step S300 specifically includes: S310, The power supply follows the optimized switching timing matrix. The output is used to acquire the electrical parameters of the switch in real time via sensors and signal acquisition systems. The host computer reads the event interval between two actions of all switching modules in real time. The thermal status parameters of all switching modules are collected in real time using digital temperature sensors. ; wherein, the electrical parameters This includes the switching operating voltage, on-state voltage drop, turn-on current, and trigger voltage; S320, Set rated values ​​for electrical stress parameters Equivalent frequency rating and temperature rating As shown in the following formula: ， ， Among them, switching to burst frequency ; S330, Based on the rated value of the electrical stress parameter Equivalent frequency rating and temperature rating Given threshold values ​​for state parameters, optimization is performed hierarchically based on state criteria: ， ， ， in, This indicates the threshold value for electrical stress parameters. This indicates that the equivalent frequency is set to a threshold. Indicates the temperature setting threshold; S340. Based on the criteria and parameter settings, thresholds are set to perform hierarchical optimization. The above process is repeated until all switch states reach a balance.

4. The method as described in claim 3, characterized in that, Step S340 specifically includes: When all state variables All are less than the set threshold : ， Then the switching timing matrix remains unchanged; where, The value ranges from 1 to 3. When taking different values, Indicate electrical parameters Event interval of the action or thermal state parameters ; When more than 50% of the state variables of a local switch are greater than the set threshold, the switch module is determined to have reached the damage condition, the switch module is stopped, a switch damage warning is issued, all action events are migrated to the normal module, the NLM algorithm is re-executed, and the execution of S100 is returned. When more than 25% of the switching modules trigger a switch failure warning, the power supply will automatically stop outputting power. Among them, when performing migration action events, priority is given to assigning them to the current one. The highest level of normal module; if multiple abnormal states occur in the second level of optimization, the temperature criterion will be used sequentially. Frequency criterion Electrical criteria The migration strategy is executed sequentially.

5. The method as described in claim 1, characterized in that, Also includes: S400: After the power supply is interrupted, record the switching history data of each round of power output according to different power interruption types. Based on the historical data, a neural network algorithm is used to obtain the various data influence factors of the switching module, optimize the criterion, and optimize the criterion again when the power is started next time. Among them, the different power supply interruption types include: after the power supply outputs for one round, three shutdown types are defined and cycled according to the switching state; The shutdown types include: The first type is normal exit; the power supply completes all output tasks in this round normally and updates historical data. The second category is automatic termination and exit; when the power supply reaches the fourth level of optimization conditions, a warning is issued that more than 25% of the switching modules are disabled, and the weight of the corresponding switching modules is increased when updating historical data, thus... ; The third category is damage-based exit, where the device malfunctions, there is no power output or a power level is missing, and the boundary conditions for the next action are set based on historical data of this category.

6. The method as described in claim 5, characterized in that, The recorded switching history data for each round of power output includes: For each round of power supply output, record the total number of switching actions of all switches. The number of times the master switch was activated. Including: the temperature corresponding to each switching action. The burst frequency of the switch Electrical parameters of the switch ; ， ， ， ， ， ， in, This is the maximum temperature when the switch is activated. For the maximum burst frequency, Refers to the maximum value of electrical parameters; This represents the average burst frequency. The average temperature during switch operation. Similarly, it is the average value of multiple electrical parameters.

7. The method as described in claim 5, characterized in that, The data influence factors of the switching module obtained by using a neural network algorithm based on the historical data include: The historical data is imported into a neural network algorithm to obtain the switch health coefficient. as well as Various data influence factors ; set up The criterion corresponding to the third-level optimization condition is based on the influence factor. Adjusting the priority of the aforementioned criteria; Influence factors as follows: ， in, The initial values ​​of all influencing factors are set to zero, corresponding to the first round of power supply output, and the criterion is that there is no historical data for adjustment.

8. The method as described in claim 5, characterized in that, Optimize the criterion, including: based on The predictive model adjusts the set threshold coefficient in the power output for each switch in the next cycle. and data weights , The deep learning objective is to provide early warnings based on minimizing power supply switching during operation. ， in, This indicates the proportion of warning switches in one round of output. The optimization coefficients are automatically updated in each round.

9. A large-scale switching intelligent modulation optimization device for all-solid-state power supplies, characterized in that, include: The initialization module is used to generate an initial switching timing matrix for any waveform selected by the user using the NLM algorithm. ; The dynamic optimization module is used to extract the initial switching timing matrix. Effective action events form a time series. time series Divide equally to Level switch module, switch module Total number of events allocated As shown in the following formula: ， ， in, For each event, the total number of switching actions based on the NLM algorithm is [number]. This corresponds to a single switch action; For floor function, when When divisible, ;when ,Will Assigned according to dynamic weights Level switch module, For even-numbered components, when When it is even, ,when When the number is odd, for high dynamic weights ,on the contrary ; The dynamic weighted uniform sorting includes: prioritizing the allocation of switch events to the switch module with the highest current weight; the weight is inversely proportional to the number of switch actions. A tiered control module is used to optimize the power supply's switching timing matrix. Output and collect switch parameters in real time; the switch parameters include operating voltage, on-state voltage drop, on-state current, trigger voltage, temperature and continuous operation interval of the switch; execute the optimization strategy according to the criteria, and repeat the above process until the state of each switch reaches a balance; The optimization strategy includes: when all switching parameters are below the corresponding set threshold, the switching timing matrix remains unchanged; when no more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching timing matrix is ​​adjusted according to the criterion migration coefficient. Reduce the number of actions performed by switches in abnormal states, and redistribute the reduced actions to other normal switches, then update and optimize the switch timing matrix. When more than 50% of the switching parameters in the switching module exceed the corresponding set threshold, the switching module is determined to have reached the failure condition, the switching module is disabled, and all action events are migrated to the normal switching module; when more than 25% of the switching modules reach the failure condition, the power supply automatically terminates the output. Specifically, when no more than 50% of the state variables of a local switch exceed a set threshold, the abnormal switch module is analyzed and located, and the migration coefficient is used as a criterion. Reduce the number of times the switch operates, and adjust the reduced number of operations to the normal switching module to obtain the state feedback optimization matrix. Optimize the matrix based on state feedback Update and optimize the switching timing matrix Controls the switching operation of the power supply; the updated switching module action is represented as follows: ， ， ， in, The total number of events allocated to the switch module after feedback optimization based on switch status monitoring is indicated by the superscript, where 1 indicates algorithm timing optimization, 2 indicates status detection optimization, and the subscript indicates that the number of switch actions may be different for different parameter transitions. The total number of events to be migrated to other modules. The migration coefficient; For electrical parameter migration coefficient, For frequency parameter mobility coefficients, The temperature parameter migration coefficient; the local switch is less than 25% of the total number of switch modules.

Images