A new method for surge suppression of power modules

Abstract

This application discloses a new method for surge suppression in a power module. After determining that the current input voltage has reached its peak value, the method utilizes the dynamic variation characteristics and trends between the input voltage timing data and the output voltage timing data over a predetermined time period to determine whether to connect a rectifier device to the rectifier circuit, that is, to determine whether to execute step 4 or return to step 2. This method avoids the drawbacks of traditional fixed setpoint comparison and monitoring control methods, enabling effective surge suppression under a wider range of conditions, thereby helping to improve the stability and efficiency of the power module startup process.

Description

Technical Field

[0001] The present application relates to the field of intelligent identification, and more specifically, to a new power module surge suppression method. Background Art

[0002] In modern power electronics systems, power modules, as core components for energy conversion, have a direct impact on the stability and efficiency of the entire system. Especially during startup, due to grid voltage fluctuations or changes in load characteristics, power modules are prone to generating large inrush currents, which pose a potential threat to both the power module itself and the connected grid. Inrush currents can not only damage components within the power module but can also cause grid voltage drops, impacting the normal operation of other equipment. Therefore, effectively suppressing startup inrush currents has become a key research area for improving power module reliability and grid stability.

[0003] Chinese patent CN112953190A proposes a method for suppressing startup surges in a PFC power module. This method uses a pre-charging circuit to charge a first capacitor during startup, ensuring that the auxiliary power supply can provide stable operating power to the control unit. Once operational, the control unit detects the input and output voltages in real time and, based on these parameters, outputs a control signal to control the rectifier device, enabling it to connect or disconnect. This allows the PFC power module to avoid or suppress startup surges during operation. This method achieves efficient surge suppression without the need for a surge switch, while minimizing waveform distortion.

[0004] However, in the aforementioned PFC power module startup surge suppression method, the decision to connect the rectifier to the rectifier circuit is made by comparing whether the difference between the current input voltage and the output voltage exceeds a fixed first set value. If this set value is set too high, it may cause a delay in connecting the rectifier, thereby prolonging the entire startup process and affecting startup efficiency. Conversely, if the set value is too low, the rectifier may be connected before the input voltage reaches its peak, increasing the risk of generating large inrush currents during startup.

[0005] Therefore, an optimized new power module surge suppression solution is desired. Summary of the Invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a new power module surge suppression method, which, after determining that the current input voltage value has reached a peak value, uses the dynamic change characteristics and trends between the input voltage timing data and the output voltage timing data in a predetermined time period to determine whether to connect the rectifier device to the rectifier circuit, that is, to determine whether to execute step 4 or return to step 2. In this way, the defects of the traditional fixed set value comparison and monitoring control method can be avoided, and effective surge suppression can be achieved under a wider range of conditions, thereby helping to improve the stability and efficiency of the power module startup process.

[0007] According to one aspect of the present application, a new power module surge suppression method is provided, which includes:

[0008] Step 1: Make the auxiliary power supply work;

[0009] Step 2: Continuously collect output voltage and input voltage values;

[0010] Step 3: After determining that the current input voltage value has reached the peak value, determine whether to execute step 4 or return to step 2 based on the time queue of the input voltage and the time queue of the output voltage in the predetermined time period;

[0011] Step 4: Output a control signal to control the rectifier device of the power module to connect to the rectifier circuit to charge the first capacitor;

[0012] Step 5: If the current input voltage value is less than the difference between the current output voltage value and the preset threshold value, disconnect the rectifier device and return to step 3; if not, return to step 4;

[0013] Step 6: Determine whether the difference between the current input voltage value and the output voltage value is less than the predetermined threshold value. If so, keep the rectifier device in the on state so that the power module enters the normal working mode. If not, return to step 3 and continue input surge suppression.

[0014] Compared to the prior art, the new power module surge suppression method provided by this application, after determining that the current input voltage value has reached its peak, utilizes the dynamic change characteristics and trends between the input voltage timing data and the output voltage timing data over a predetermined time period to determine whether to connect the rectifier device to the rectifier circuit, that is, to determine whether to execute step 4 or return to step 2. This avoids the shortcomings of traditional fixed set value comparison and monitoring control methods, enables effective surge suppression under a wider range of conditions, and thus helps improve the stability and efficiency of the power module startup process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 Flowchart of a new power module surge suppression method according to an embodiment of the present application;

[0017] Figure 2 Schematic diagram of data flow of a new power module surge suppression method according to an embodiment of the present application;

[0018] Figure 3 Flowchart of sub-step S3 of the new power module surge suppression method according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0020] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0021] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0022] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0024] In the technical solution of the present application, a new power module surge suppression method is proposed. Figure 1 Flowchart of a new power module surge suppression method according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the new power module surge suppression method according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the new power module surge suppression method according to the embodiment of the present application includes the following steps: S1, making the auxiliary power supply work; S2, continuously collecting the output voltage value and the input voltage value; S3, after judging that the current input voltage value has reached the peak value, based on the time queue of the input voltage and the time queue of the output voltage in the predetermined time period, determining whether to execute step 4 or return to step 2; S4, outputting a control signal to control the rectifier device of the power module to connect to the rectifier circuit to charge the first capacitor; S5, if the current input voltage value is less than the difference between the current output voltage value and the preset threshold value, disconnecting the rectifier device and returning to step 3, if not, returning to step 4; S6, judging whether the difference between the current input voltage value and the output voltage value is less than the predetermined threshold value, if so, keeping the rectifier device in the open state so that the power module enters the normal working mode, if not, returning to step 3 and continuing the input surge suppression.

[0025] Specifically, S1 activates the auxiliary power supply. When the PFC power module is powered on, the auxiliary power supply first begins operating, providing power to the control unit. At this point, the control unit begins operating and selects or obtains a pre-set first set value. This first set value serves as a standard for determining whether the difference between the input voltage and the output voltage is safe.

[0026] It's understandable that enabling the auxiliary power supply is a crucial first step in the power module's startup process. The auxiliary power supply's primary function is to provide the necessary power to the control system before the main power supply establishes a stable output, ensuring the control system can monitor and manage the startup process in a timely manner.

[0027] Specifically, auxiliary power supplies typically utilize switching power supplies (SSMs) due to their advantages, including low power consumption, high efficiency, compact size, light weight, and wide voltage regulation range. The core of a switching power supply is the converter, with common topologies including single-ended flyback and forward. For low-power applications, the single-ended flyback converter is widely adopted due to its simple structure and convenient multiple outputs. The basic structure of an auxiliary power supply includes input rectification and filtering, a converter, transformer isolation, output rectification and filtering, feedback control, and protection circuitry. These components work together to ensure a stable and reliable power supply.

[0028] The AC input voltage is first converted to DC voltage by a rectifier bridge. This bridge, typically composed of four diodes, converts the AC power into pulsating DC. This pulsating DC power requires further filtering to reduce voltage fluctuations. When selecting a filter capacitor, consider its capacity and withstand voltage to ensure it effectively filters out ripple and provides a stable DC voltage.

[0029] The rectified DC voltage is initially filtered by a filter capacitor to remove high-frequency ripple. When selecting a filter capacitor, consider its capacity and withstand voltage to ensure it effectively filters ripple and provides a stable DC voltage. For example, for a 220VAC input, the rectified DC voltage is approximately 300V, so a 470μF / 400V electrolytic capacitor can be selected for filtering.

[0030] The filtered DC voltage is fed into the converter, where it is converted to a high-frequency AC voltage through the switching action of a switching transistor (such as a MOSFET). The converter operates by controlling the on and off states of the switching transistor to change the frequency and duty cycle of the input voltage, thereby achieving voltage conversion. When selecting a switching transistor, factors such as its withstand voltage, on-resistance, switching speed, and maximum current should be considered.

[0031] High-frequency AC voltage is transformed and electrically isolated through a transformer. The primary and secondary sides of the transformer transfer energy through magnetic coupling, achieving voltage step-up or step-down. Transformer design requires consideration of core material, winding turns, and insulation performance to ensure efficient and stable operation. For example, a transformer with a ferrite core can have 100 primary turns and 20 and 40 secondary turns, corresponding to output voltages of 15V and 24V, respectively.

[0032] The low-voltage, high-frequency square wave on the secondary side of the transformer is then rectified into low-voltage DC power by rectifier diodes. This is further filtered by filter capacitors, resulting in a stable DC output voltage. When selecting rectifier diodes, consider their withstand voltage, current capability, and reverse recovery time to ensure efficient and reliable operation. For example, two fast-recovery diodes with a withstand voltage of 100V and a maximum current of 1A can be selected to rectify the 15V and 24V outputs, respectively. The selection of filter capacitors is equally important. Two 1000μF / 25V electrolytic capacitors can be used to filter the 15V and 24V outputs, respectively, to achieve a stable DC output voltage.

[0033] Auxiliary power supply circuits typically include a feedback loop to monitor output voltage changes. Common feedback components include optocouplers and TL431s. This feedback loop compares the monitored output voltage signal with a reference voltage, generating an error signal through an error amplifier. This signal is then used to adjust the duty cycle of the PWM controller, controlling the on-time of the switching transistor and achieving output voltage stability. For example, a PC817 optocoupler and a TL431 precision adjustable voltage regulator can be used to generate the reference voltage. A UC3842 PWM controller can be used to generate the control signal.

[0034] Specifically, S2 continuously collects output and input voltage values. The control unit continuously collects input and output voltage values. The input voltage is an AC voltage, which is converted to a DC voltage value by an isolated voltage sensor; the output voltage is a DC voltage, which is directly collected by a voltage divider circuit.

[0035] It's clear that continuously collecting output and input voltage values ​​during the power module's startup and operation is a critical step in ensuring system stability and efficiency. By monitoring these voltages in real time, the control system can make timely adjustments to avoid inrush currents and other abnormalities.

[0036] In one example, to continuously acquire output and input voltage values, appropriate hardware components must first be configured. These components include a voltage sensor, an analog-to-digital converter (ADC), a microcontroller (MCU), and a power supply. Voltage sensors are used to measure input and output voltages. Commonly used sensors include resistor dividers, Hall-effect sensors, and isolation amplifiers. The analog-to-digital converter (ADC) converts analog voltage signals into digital signals for processing by a microcontroller or other digital processing unit. The microcontroller (MCU) is responsible for controlling the ADC sampling process and processing and analyzing the collected data. The power supply provides a stable power supply to the above components, ensuring the normal operation of the system.

[0037] The input voltage is typically an AC voltage and requires a series of steps to be collected. First, a rectifier bridge is used to convert the AC voltage into a pulsating DC voltage. This rectifier bridge typically consists of four diodes, converting the AC power into unidirectional pulsating DC power. For example, for a 220VAC input, the rectified DC voltage is approximately 300V. Next, a filter capacitor performs preliminary filtering on the pulsating DC voltage to reduce high-frequency ripple. When selecting a filter capacitor, consider its capacity and withstand voltage to ensure it can effectively filter out ripple and provide a stable DC voltage. For example, a 470μF / 400V electrolytic capacitor can be used for filtering.

[0038] A resistor divider is then used to reduce the high DC voltage to a range suitable for the ADC input. The voltage divider design must consider the voltage divider ratio and power consumption to ensure a stable and reliable voltage signal. For example, a resistor divider with a 1:10 ratio can be used to reduce a 300V DC voltage to 30V. To ensure system electrical safety, an isolation amplifier or optocoupler can be used to isolate the input voltage signal from the control system. Isolation amplifiers provide electrical isolation, preventing interference from the high-voltage side from affecting the control system. For example, an isolation amplifier can be used to isolate a 30V voltage signal from the control system.

[0039] Finally, the divided voltage signal is fed into an ADC for sampling. When selecting an ADC, consider its resolution, sampling rate, and accuracy to ensure accurate acquisition of the voltage signal. For example, an ADC with 12-bit resolution and a sampling rate of 1kHz can be selected for sampling.

[0040] The output voltage is typically a DC voltage and can be directly acquired. First, a resistor divider is used to reduce the output voltage to a range suitable for the ADC input. The voltage divider design must consider the voltage divider ratio and power consumption to ensure a stable and reliable voltage signal. For example, a resistor divider with a 1:2 ratio can be used to reduce a 24V output voltage to 12V. Next, a filter capacitor is used to filter the divided voltage signal to reduce noise interference. When selecting the filter capacitor, consider its capacity and voltage rating to ensure it can effectively filter out noise and provide a stable voltage signal. For example, a 100μF / 25V electrolytic capacitor can be used for filtering.

[0041] To ensure system electrical safety, an isolation amplifier or optocoupler can be used to isolate the output voltage signal from the control system. For example, an isolation amplifier can be used to isolate the 12V voltage signal from the control system. Finally, the divided voltage signal is fed into an ADC for sampling. When selecting an ADC, consider its resolution, sampling rate, and accuracy to ensure accurate voltage signal acquisition. For example, an ADC with 12-bit resolution and a sampling rate of 1kHz can be selected for sampling. Choosing the sampling frequency requires consideration of several factors. First, according to the Nyquist theorem, the sampling frequency should be at least twice the highest frequency of the signal to avoid aliasing. For example, for a 50Hz AC input voltage, a sampling frequency of 1kHz can be selected to accurately capture changes in the voltage waveform. Second, the sampling frequency should be high enough to ensure that the control system can respond promptly to voltage changes. For example, for systems requiring fast response, the sampling frequency can be increased to 5kHz or higher. Finally, the sampling frequency should not be too high, as it may exceed the processing capabilities of the processor. For example, for microcontrollers with limited processing power, the sampling frequency can be appropriately reduced to ensure real-time data processing.

[0042] To reduce noise interference, the collected data can be filtered. Common filtering methods include low-pass filtering, averaging filtering, and median filtering. Low-pass filtering removes high-frequency noise while retaining low-frequency signals. When designing a low-pass filter, the cutoff frequency and order must be considered to ensure effective filtering. For example, a second-order low-pass filter with a cutoff frequency of 100 Hz can be selected. An averaging filter reduces the impact of random noise by averaging data from multiple sampling points. When designing an averaging filter, the window size must be considered to balance filtering effectiveness and response speed. For example, an averaging filter with a window size of 10 can be selected. A median filter sorts the data from multiple sampling points and takes the median value to remove the impact of outliers. When designing a median filter, the window size must be considered to balance filtering effectiveness and response speed. For example, a median filter with a window size of 5 can be selected.

[0043] To facilitate subsequent processing and analysis, the collected data can be stored in memory or transmitted to a host computer via a communication interface. Common data storage and transmission methods include internal storage, external storage, and communication interfaces. Internal storage stores data in the microcontroller's RAM or Flash memory and is suitable for short-term storage and local processing. For example, the data of the most recent 100 sampling points can be stored in RAM for real-time analysis. External storage stores data in external memory, such as an SD card or EEPROM, and is suitable for long-term storage and offline analysis. For example, daily voltage data can be stored in an SD card for subsequent analysis. The communication interface transmits data to the host computer via a communication interface such as UART, SPI, I2C, or CAN, and is suitable for remote monitoring and real-time analysis. For example, voltage data can be transmitted to the host computer via a UART interface for real-time display and recording.

[0044] Specifically, in step S3, after determining that the current input voltage has reached its peak value, the system determines whether to proceed to step 4 or return to step 2 based on the input voltage time series and the output voltage time series over a predetermined time period. In step 3, the input voltage is continuously monitored, and when the input voltage reaches its peak value within a cycle, the next step is initiated. This ensures that the comparison occurs when the input voltage is at its highest, as this is when a large inrush current is most likely to occur. The original solution compared the difference between the current input and output voltages to see if it is greater than a first set value. If the difference is greater than the first set value, it indicates that the input voltage is relatively high relative to the output voltage, posing a risk of generating a large inrush current, and thus the next step is executed. If the difference is less than the first set value, it indicates that the input voltage is relatively low, and input voltage monitoring and acquisition continue. However, inappropriate set values ​​can affect the stability and efficiency of the startup process and may even cause startup failure. Therefore, in the technical solution of the present application, after determining that the current input voltage has reached its peak value, the system utilizes the dynamic change characteristics and trends between the input and output voltage time series data over a predetermined time period to determine whether to connect the rectifier device to the rectifier circuit, that is, to determine whether to proceed to step 4 or return to step 2. Specifically, after the current input voltage value reaches its peak value, by utilizing the time queue data of the input voltage and output voltage, and introducing a data processing and analysis algorithm based on artificial intelligence and deep learning in the back end to analyze these time queue data of the input voltage and output voltage, the fine-grained comparison interaction feature representation between the multi-scale time series semantics of the input voltage and the multi-scale time series semantics of the output voltage is learned and captured, thereby achieving a more accurate capture and analysis of the time series dynamic characteristics between the input voltage and the output voltage, and then making a state judgment based on the feature representation to decide whether to execute step 4 or return to step 2. In a specific example of the present application, as Figure 3As shown, the S3 includes: S31, respectively performing sequence encoding on the time queue of the input voltage and the time queue of the output voltage to obtain multi-scale time series implicit coding features of the input voltage and multi-scale time series implicit coding features of the output voltage; S32, performing feature interaction comparison processing on the multi-scale time series implicit coding features of the input voltage and the multi-scale time series implicit coding features of the output voltage using external knowledge modulation to obtain input voltage-output voltage time series comparison fine-grained interaction features; S33, performing state discrimination based on the input voltage-output voltage time series comparison fine-grained interaction features to determine whether to execute step 4 or return to step 2.

[0045] Specifically, the S31 is to perform sequence encoding on the time queue of the input voltage and the time queue of the output voltage respectively to obtain the multi-scale temporal implicit coding features of the input voltage and the multi-scale temporal implicit coding features of the output voltage. Considering that the input voltage and the output voltage have the dynamic change characteristics and trends of the time series in the time dimension, the dynamic characteristics of this time series not only exist in the temporal dependency relationship in the short term, but also in the long-term change trend. Based on this, in order to more fully capture the temporal dynamic characteristics of the input voltage and the output voltage, thereby helping to more comprehensively understand the temporal dynamic characteristics of the input voltage and the output voltage, and provide a more accurate basis for predicting future voltage changes and state discrimination tasks, in the technical solution of the present application, the time queue of the input voltage and the time queue of the output voltage in the predetermined time period are further input into the sequence encoder based on the RNN-LSTM hybrid model to obtain the multi-scale temporal implicit coding vector of the input voltage and the multi-scale temporal implicit coding vector of the output voltage. It's worth noting that in the sequence encoder based on the RNN-LSTM hybrid model, the RNN is able to process the time series data for the input and output voltages, capturing the temporal dependencies between the input and output voltages. This is very useful for analyzing the dynamic changes of the input and output voltages, as these voltage values ​​are continuous and correlated over time. The LSTM is particularly adept at handling long-term dependencies and can effectively memorize the characteristic information of the long-term dependencies between the input and output voltages, which is crucial for capturing the long-term trends of the input and output voltages. This multi-scale analysis helps to more comprehensively understand the dynamic characteristics of the input and output voltages, thereby more accurately predicting future voltage changes. It also improves the model's robustness and generalization capabilities under different operating conditions, which is crucial for determining whether to connect a rectifier device to the rectifier circuit to improve the stability of the power module startup process.

[0046] Specifically, in S32, the input voltage multi-scale time series implicit coding features and the output voltage multi-scale time series implicit coding features are subjected to feature interactive comparison processing using external knowledge modulation to obtain input voltage-output voltage time series comparison fine-grained interactive features. It should be understood that since the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector respectively contain multi-scale time series dynamic feature information about the input voltage and output voltage in the time dimension, however, since the input voltage and output voltage not only have their own time series dynamic characteristics in the time dimension, that is, the voltage data at the current time point will be affected by the time series changes of the previous voltage, but also there is interactive information and dynamic correlation between the voltage time series semantics between the two. Therefore, in order to perform fine-grained interactive comparison of the input voltage multi-scale time series features and the output voltage multi-scale time series features, thereby providing more refined support for subsequent state discrimination and decision-making, in the technical solution of the present application, the input voltage multi-scale time series implicit coding features and the output voltage multi-scale time series implicit coding features are further subjected to feature interactive comparison processing using external knowledge modulation to obtain fine-grained interactive features of the input voltage-output voltage time series comparison. In particular, the feature interactive comparison processing using external knowledge modulation can further capture the subtle but important interactive information between the input voltage multi-scale time series features and the output voltage multi-scale time series features. This fine-grained interactive information helps to more accurately understand the dynamic relationship between the input voltage and output voltage, thereby providing more refined support for subsequent state discrimination tasks and decision-making. It is worth mentioning that in the feature interactive comparison processing process using external knowledge modulation, not only the interaction and comparison information of the time series multi-scale features between the input voltage and output voltage are strengthened, but also these interactions are optimized by integrating external knowledge. External knowledge, such as the experience of domain experts and historical data, is effectively integrated into the interactive comparison of input and output voltage timing features through an attention mechanism. This enables a better understanding and handling of complex state discrimination tasks in various scenarios, improving its performance in practical applications. This enhanced generalization capability enables the model to maintain high accuracy and stability across diverse operating conditions and grid environments, providing reliable data support for subsequent control decisions.

[0047] In an embodiment of the present application, the specific steps of performing feature interaction comparison processing on the input voltage multi-scale temporal implicit coding features and the output voltage multi-scale temporal implicit coding features using external knowledge modulation include: first, performing fine-grained feature interaction and optimization representation on the input voltage multi-scale temporal implicit coding vector and the output voltage multi-scale temporal implicit coding vector to obtain an external knowledge optimized input voltage-output voltage temporal fine-grained feature interaction matrix. That is, in a specific example of the present application, the input voltage multi-scale temporal implicit coding vector and the output voltage multi-scale temporal implicit coding vector are first subjected to fine-grained feature interaction to obtain an input voltage-output voltage temporal fine-grained feature interaction matrix; then, the input voltage-output voltage temporal fine-grained feature interaction matrix is ​​passed through an attention unit based on external knowledge to obtain the external knowledge optimized input voltage-output voltage temporal fine-grained feature interaction matrix.

[0048] More specifically, the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector are subjected to fine-grained feature interaction and optimized representation using the following fine-grained feature interaction optimization formula to obtain an external knowledge optimized input voltage-output voltage time series fine-grained feature interaction matrix; wherein the fine-grained feature interaction optimization formula is:

[0049]

[0050] Among them, V1 and V2 are the multi-scale time series implicit coding vector of the input voltage and the multi-scale time series implicit coding vector of the output voltage, respectively. p is the input voltage-output voltage timing comparison fine-grained feature interaction matrix, M k and M v represents the first external knowledge attention learnable memory parameter matrix and the second external knowledge attention learnable memory parameter matrix, norm(·) represents the normalization function, M y Optimizing the input voltage-output voltage timing fine-grained feature interaction matrix for external knowledge.

[0051] Next, based on the external knowledge, the input voltage-output voltage time series fine-grained feature interaction matrix is ​​optimized, and the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector are modulated to obtain the optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector. That is, in a specific example of the present application, the input voltage-output voltage time series fine-grained feature interaction matrix is ​​first optimized based on the external knowledge, and the input voltage multi-scale time series implicit coding vector is fine-grained modulated and optimized to obtain the optimized input voltage multi-scale time series implicit coding vector; and then, based on the external knowledge, the input voltage-output voltage time series fine-grained feature interaction matrix is ​​optimized based on the external knowledge, and the output voltage multi-scale time series implicit coding vector is fine-grained modulated and optimized to obtain the optimized output voltage multi-scale time series implicit coding vector.

[0052] Among them, based on the external knowledge optimization of the input voltage-output voltage time series fine-grained feature interaction matrix, the input voltage multi-scale time series implicit coding vector is fine-grained modulated and optimized to obtain the optimized input voltage multi-scale time series implicit coding vector, including: linearly transforming the input voltage multi-scale time series implicit coding vector to obtain an input voltage multi-scale time series query feature vector and an input voltage multi-scale time series value feature vector; using the external knowledge optimization of the input voltage-output voltage time series fine-grained feature interaction matrix as a key matrix, inputting the input voltage multi-scale time series query feature vector, the input voltage multi-scale time series value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized input voltage multi-scale time series implicit coding vector; using And, based on the external knowledge-optimized input voltage-output voltage timing fine-grained feature interaction matrix, the output voltage multi-scale timing implicit coding vector is fine-grained modulated and optimized to obtain the optimized output voltage multi-scale timing implicit coding vector, including: linearly transforming the output voltage multi-scale timing implicit coding vector to obtain an output voltage multi-scale timing query feature vector and an output voltage multi-scale timing value feature vector; using the external knowledge-optimized input voltage-output voltage timing fine-grained feature interaction matrix as a key matrix, the output voltage multi-scale timing query feature vector, the output voltage multi-scale timing value feature vector and the key matrix are input into the Transformer-based fine-grained modulation module to obtain the optimized output voltage multi-scale timing implicit coding vector.

[0053] More specifically, the input voltage-output voltage time series fine-grained feature interaction matrix is ​​optimized based on the external knowledge, and the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector are modulated using the following modulation formula to obtain optimized input voltage multi-scale time series implicit coding vector and optimized output voltage multi-scale time series implicit coding vector; wherein the modulation formula is:

[0054]

[0055] Among them, W 1q and b 1q They represent the input voltage multi-scale time series query weight matrix and the input voltage multi-scale time series query bias vector, W 1v and b 1v Respectively represent the input voltage multi-scale time series value weight matrix and the input voltage multi-scale time series value bias vector, is the matrix multiplication, V 1q and V 1v are the input voltage multi-scale time series query feature vector and the input voltage multi-scale time series value feature vector, (·) T is the matrix transpose, d1 is V 1q The length of , softmax(·) represents the softmax function, V′1 is the optimized input voltage multi-scale time series implicit encoding vector, W 2q and b 2q They represent the output voltage multi-scale time series query weight matrix and the output voltage multi-scale time series query bias vector, W 2v and b 2v Represent the output voltage multi-scale time series value weight matrix and the output voltage multi-scale time series value bias vector, V 2q and V 2v are the output voltage multi-scale time series query feature vector and the output voltage multi-scale time series value feature vector, d2 is V 2q , V′2 is the optimized output voltage multi-scale time series implicit coding vector.

[0056] Furthermore, semantic contrast and association coding is performed on the optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector to obtain an input voltage-output voltage time series comparison fine-grained interaction feature vector as the input voltage-output voltage time series comparison fine-grained interaction feature. That is, in a specific example of the present application, the input voltage-output voltage time series comparison fine-grained interaction feature vector is obtained by calculating the position point division between the optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector.

[0057] More specifically, semantic contrast and association coding is performed on the optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector using the following semantic contrast and association coding formula to obtain an input voltage-output voltage time series contrast fine-grained interaction feature vector; wherein the semantic contrast and association coding formula is:

[0058]

[0059] Among them, V s It is the fine-grained interaction feature vector of the input voltage-output voltage timing comparison.

[0060] Specifically, the S33 performs state discrimination based on the input voltage-output voltage timing comparison fine-grained interaction feature to determine whether to execute step 4 or return to step 2. That is, in a specific example of the present application, the input voltage-output voltage timing comparison fine-grained interaction feature vector is input into a state discriminator based on a classifier to obtain a discrimination result, and the discrimination result is used to indicate whether to execute step 4 or return to step 2. In other words, the semantic features of the timing fine-grained comparison interaction between the input voltage and the output voltage are used for classification processing, so as to perform state judgment to decide whether to execute step 4 or return to step 2. In this way, a more intelligent new power module surge suppression control can be achieved, which can avoid the defects of the traditional fixed set value comparison to determine whether to connect the rectifier device to the rectifier circuit, and thus can achieve effective surge suppression under a wider range of conditions, which helps to improve the stability of the power module startup process.

[0061] In one example, a classifier-based state discriminator can be used to determine whether the next action is to execute step 4 or return to step 2. Specifically, a large amount of input voltage and output voltage data is collected, and the correct action corresponding to each set of data is labeled (execute step 4 or return to step 2). For example, when the difference between the input voltage and output voltage is less than a preset threshold, it is labeled as executing step 4; otherwise, it is labeled as returning to step 2. The data is divided into a training set and a test set. The training set is used to train the classifier, and the test set is used to evaluate the classifier's performance.

[0062] In the technical solution of the present application, the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector respectively represent the multi-scale time series correlation characteristics of the input voltage and the output voltage. When performing fine-grained feature interaction based on external knowledge modulation, the time series multi-scale load type of the input voltage and the output voltage will make the fine-grained modulation of the external knowledge difficult, resulting in insufficient representation of the long-distance fine-grained interaction response, thereby reducing the expression effect of the input voltage-output voltage time series comparison fine-grained interaction feature vector, and affecting the accuracy of the discrimination result obtained by its input classifier-based state discriminator.

[0063] Therefore, in one example, when the input voltage-output voltage timing sequence comparison fine-grained interaction feature vector is input into a state discriminator based on a classifier, the input voltage-output voltage timing sequence comparison fine-grained interaction feature vector is optimized, and the optimization includes the following steps:

[0064] Arranging the eigenvalues ​​of the input voltage-output voltage time series comparison fine-grained interaction eigenvalue vector in ascending order to form an input voltage-output voltage time series comparison fine-grained interaction sequence encoding vector;

[0065] In response to the absolute value of the difference between the i-th eigenvalue and the i+1-th eigenvalue of the input voltage-output voltage timing comparison fine-grained interaction sequential encoding vector being less than or equal to the distance difference hyperparameter ε, the weighted sum between the i-th eigenvalue and the i+1-th eigenvalue is calculated as the optimized i+1-th eigenvalue v′ i+1 =ω1×v i +ω2×v i+1 , where v i and v i+1 They respectively represent the i-th eigenvalue and the i+1-th eigenvalue of the input voltage-output voltage time series contrast fine-grained interaction sequence encoding vector, ω1 represents the first weighted hyperparameter, and ω2 represents the second weighted hyperparameter;

[0066] In response to the input voltage-output voltage time series comparison, the absolute value of the difference between the i-th eigenvalue and the i+1-th eigenvalue of the fine-grained interactive sequential encoding vector is greater than the distance difference hyperparameter ε:

[0067] Calculate the square root of the sum of squares of all eigenvalues ​​of the input voltage-output voltage timing contrast fine-grained interaction eigenvector Among them, v1, v2 and v n Respectively represent the first, second and nth eigenvalues ​​of the input voltage-output voltage time series comparison fine-grained interaction eigenvector;

[0068] Multiply the square root by 2 and then divide it by the square of the length of the input voltage-output voltage timing comparison fine-grained interaction feature vector to obtain the input voltage-output voltage timing comparison fine-grained interaction space primitive value base=2×root / L 2 , where L represents the length of the input voltage-output voltage time series comparison fine-grained interaction feature vector;

[0069] After multiplying the input voltage-output voltage time series comparison fine-grained interaction space primitive value by the i-th eigenvalue, the weighted subtraction between the product and the i+1th eigenvalue is calculated to obtain the optimized i+1th eigenvalue v i+1 ′=ω3×vi ×base-ω4×v i+1 , where ω3 represents the third weighted hyperparameter and ω4 represents the fourth weighted hyperparameter;

[0070] Based on v′1=v1, where v′1 represents the first eigenvalue of the optimized input voltage-output voltage timing contrast fine-grained interaction eigenvector, the i+1th eigenvalue v of the combined optimization is i+1 ′ is used to obtain the optimized input voltage-output voltage timing comparison fine-grained interaction feature vector.

[0071] In this way, in order to address the problem of insufficient global semantic search representation capability of the feature set of the input voltage-output voltage timing comparison fine-grained interaction feature vector under a predetermined eigenvalue sequential distribution due to a long distance exceeding a predetermined local distribution interval threshold, the high-dimensional feature space primitive representation based on self-inner product fusion of the input voltage-output voltage timing comparison fine-grained interaction feature vector is used to capture the complex structure of the global network interaction of its eigenvalues, thereby reconstructing the dynamic semantic search relationship between the eigenvalues ​​of the input voltage-output voltage timing comparison fine-grained interaction feature vector by simulating the scale-based high-dimensional feature space potential primitives, so as to realize the encoding reconstruction of the real sequence distribution behavior of the input voltage-output voltage timing comparison fine-grained interaction feature vector under a long distance, improve the encoding expression effect of the input voltage-output voltage timing comparison fine-grained interaction feature vector, and improve the accuracy of the discrimination result obtained by its input into the classifier-based state discriminator.

[0072] In particular, step S4 outputs a control signal to control the rectifier device of the power module to connect to the rectifier circuit and charge the first capacitor. During the startup and operation of the power module, outputting a control signal to control the rectifier device to connect to the rectifier circuit and charge the first capacitor is a key step in ensuring system stability and efficiency. By precisely controlling the timing of connecting the rectifier device, inrush current can be effectively avoided, ensuring a smooth startup of the power module.

[0073] In order to output the control signal, control the rectifier device to connect to the rectifier circuit, and charge the first capacitor, appropriate hardware components need to be configured. These components include a microcontroller (MCU), a drive circuit, a rectifier device, and a first capacitor. The microcontroller is responsible for generating control signals to control the connection and disconnection of the rectifier device. The drive circuit is used to amplify the control signal of the microcontroller and drive the rectifier device. The rectifier device is used to convert AC voltage into DC voltage. Common rectifier devices include MOSFET, IGBT, and diodes. The first capacitor is used to store electrical energy to ensure the stability of the output voltage. It usually has a large capacity to absorb the inrush current during startup.

[0074] In one example, the microcontroller first initializes the ADC and timer, setting the sampling frequency and sampling interval. For example, the sampling frequency can be set to 1kHz and the sampling interval to 1ms. The microcontroller then continuously collects input and output voltage values ​​using the ADC and stores the data in RAM. The collected data is then filtered to reduce noise. Common filtering methods include low-pass filtering, averaging filtering, and median filtering. For example, a second-order low-pass filter with a cutoff frequency of 100Hz can be used to filter the data. Finally, the filtered data is stored in RAM for subsequent analysis and control. Next, input and output voltage thresholds are set based on system requirements. For example, the input voltage threshold can be set to 300V and the output voltage threshold to 22V. The microcontroller compares the collected voltage values ​​with the preset thresholds in real time to determine whether the access conditions are met. For example, the access condition is triggered when the input voltage reaches 300V and the output voltage falls below 22V.

[0075] The microcontroller calculates the voltage trend based on the collected voltage values. For example, it calculates the difference between the input and output voltages to determine the voltage trend. Based on the voltage trend and preset conditions, it determines whether to connect a rectifier. For example, when the input voltage reaches its peak and the difference between the output and input voltages is less than a preset threshold, the connection condition is triggered.

[0076] When the access conditions are met, the microcontroller generates a control signal and outputs it to the driver circuit through an IO port. This control signal is typically a high or low level signal that controls the on / off state of the driver circuit. The driver circuit amplifies the microcontroller's control signal and drives the rectifier device to connect to the rectifier circuit. For example, the driver circuit amplifies the microcontroller's control signal and drives the gate of a MOSFET, turning it on.

[0077] After the rectifier device is connected to the rectifier circuit, it converts the AC voltage into a DC voltage to charge the first capacitor. For example, when the MOSFET is turned on, the AC voltage is converted to a DC voltage by the rectifier bridge, charging the 470μF / 400V first capacitor. The microcontroller continuously monitors the voltage across the first capacitor using the ADC to determine whether charging is complete. When the voltage across the first capacitor reaches a preset value, the microcontroller generates a disconnect signal, disconnecting the rectifier device through the drive circuit to prevent overcharging.

[0078] In particular, in S5, if the current input voltage value is less than the difference between the current output voltage value and the preset threshold value, the rectifier device is disconnected and the process returns to step 3. If not, the process returns to step 4. In the initial startup of the power module, the input voltage gradually increases and the output voltage gradually builds up. In order to prevent inrush current caused by excessively high input voltage, it is necessary to connect the rectifier device at an appropriate time to charge the first capacitor. At the same time, in some cases, it is necessary to disconnect the rectifier device to prevent overcharging or other abnormal conditions. Therefore, a precise control logic is required to determine when to connect and disconnect the rectifier device.

[0079] In one example, a microcontroller continuously collects input and output voltage values ​​using an ADC to monitor voltage changes in real time. Specifically, the microcontroller collects input and output voltage values ​​using the ADC and stores the data in RAM. To reduce noise interference, the collected data is filtered. Common filtering methods include low-pass filtering, averaging filtering, and median filtering. For example, a second-order low-pass filter with a cutoff frequency of 100 Hz can be used to filter the data. The filtered data is stored in RAM for subsequent analysis and control.

[0080] The microcontroller compares the collected voltage value with the preset threshold in real time to determine whether the access condition is met. Specifically, the collected voltage value is compared with the preset threshold to determine whether the access condition is met. For example, when the input voltage reaches 300V and the output voltage is less than 22V, the access condition is triggered. Based on the voltage trend and preset conditions, it further determines whether the rectifier device needs to be connected. For example, when the input voltage reaches its peak and the difference between the output voltage and the input voltage is less than the preset threshold, the access condition is triggered.

[0081] When the access conditions are met, the microcontroller generates a control signal and outputs it to the driver circuit via the IO port, driving the rectifier device to connect to the rectifier circuit. Specifically, when the access conditions are met, the microcontroller generates a control signal and outputs it to the driver circuit via the IO port. The control signal is typically a high or low level signal that controls the on / off state of the driver circuit. The driver circuit amplifies the microcontroller's control signal and drives the rectifier device to connect to the rectifier circuit. For example, the driver circuit amplifies the microcontroller's control signal to drive the gate of a MOSFET, turning it on.

[0082] After the rectifier device is connected to the rectifier circuit, it converts the AC voltage into a DC voltage to charge the first capacitor. Specifically, after the rectifier device is connected to the rectifier circuit, it converts the AC voltage into a DC voltage to charge the first capacitor. For example, when the MOSFET is turned on, the AC voltage is converted into a DC voltage through the rectifier bridge to charge the first capacitor of 470μF / 400V. The microcontroller continuously monitors the voltage value across the first capacitor through the ADC to determine whether charging is complete. For example, the voltage value across the first capacitor is collected through the ADC to determine whether charging is complete.

[0083] To prevent overcharging or other abnormal conditions, the rectifier device needs to be disconnected at an appropriate time. Specifically, the microcontroller uses the ADC to collect the current input voltage and output voltage values. The microcontroller calculates the difference between the current output voltage and a preset threshold. For example, if the preset threshold is 22V and the current output voltage is 24V, the difference is 24V - 22V = 2V. The current input voltage is compared with the calculated difference. For example, if the current input voltage is 290V and the difference is 2V, then 290V < (24V - 2V) = 22V. If the current input voltage is less than the difference between the current output voltage and the preset threshold, the microcontroller generates a disconnect signal, which disconnects the rectifier device through the drive circuit. For example, when 290V < 22V, a disconnect signal is generated to drive the gate of the MOSFET, turning it off. After disconnecting the rectifier device, the process returns to step 3 and repeats the threshold comparison and status determination. If the current input voltage is not less than the difference between the current output voltage and the preset threshold, the process returns to step 4, where control signals are continuously generated to maintain the rectifier device connected and continue charging the first capacitor.

[0084] Specifically, step S6 determines whether the difference between the current input voltage and the output voltage is less than a predetermined threshold. If so, the rectifier device remains in the on state, allowing the power module to enter normal operation. If not, the process returns to step 3 to continue input surge suppression. In one example, using a typical PFC power module as an example, the specific implementation steps of step S6 are as follows: First, the microcontroller is initialized, and the sampling frequency is set to 1kHz and the sampling interval is 1ms. Based on system requirements, the input voltage threshold is set to 300V, the output voltage threshold is set to 22V, and the preset threshold difference is 10V. The microcontroller continuously acquires the input and output voltage values ​​using the ADC and stores the data in RAM. The acquired data is low-pass filtered to reduce noise interference. The filtered data is stored in RAM for subsequent analysis and control. The microcontroller compares the acquired voltage values ​​with preset thresholds in real time to determine whether the access condition is met. For example, the access condition is triggered when the input voltage reaches 300V and the output voltage is less than 22V. Based on the voltage trend and preset conditions, the microcontroller further determines whether the rectifier device needs to be engaged. For example, when the input voltage reaches its peak and the difference between the output voltage and the input voltage is less than a preset threshold, the access condition is triggered. When the access condition is met, the microcontroller generates a control signal and outputs it to the driver circuit through the IO port. The driver circuit amplifies the microcontroller's control signal and drives the gate of the MOSFET, turning it on. After the MOSFET turns on, it converts the AC voltage into DC voltage, charging the 470μF / 400V first capacitor. The microcontroller continuously monitors the voltage across the first capacitor using the ADC to determine whether charging is complete. To ensure a smooth transition to normal operating mode, the microcontroller uses the ADC to collect the current input and output voltage values. The microcontroller calculates the difference between the current output voltage and the preset threshold. For example, if the preset threshold is 22V and the current output voltage is 24V, the difference is 24V - 22V = 2V. The current input voltage is compared with the calculated difference. For example, if the current input voltage is 300V and the difference is 2V, then 300V > (24V - 2V) = 22V. If the difference between the current input voltage and the output voltage is less than the preset threshold, the rectifier remains on, allowing the power module to enter normal operating mode. Specifically, when 300V-24V < 10V, the MOSFET gate remains on, allowing the power module to enter normal operating mode. At this point, the power module's output voltage has stabilized and can supply power normally. If the difference between the current input voltage and the output voltage is not less than the preset threshold, the process returns to step 3 and continues with input surge suppression. Specifically, when 300V-24V > 10V, a disconnect signal is generated, driving the MOSFET gate to turn off, and the process returns to step 3 to continue with input surge suppression.At this time, the power module continues to monitor the changes in input voltage and output voltage, waiting for the right time to connect to the rectifier device again.

[0085] In summary, the new power module surge suppression method according to the embodiment of the present application is explained. After determining that the current input voltage value has reached a peak value, the method utilizes the dynamic change characteristics and trends between the input voltage timing data and the output voltage timing data over a predetermined time period to determine whether to connect the rectifier device to the rectifier circuit, that is, to determine whether to execute step 4 or return to step 2. In this way, the shortcomings of traditional fixed set value comparison and monitoring control methods can be avoided, and effective surge suppression can be achieved under a wider range of conditions, thereby helping to improve the stability and efficiency of the power module startup process.

[0086] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A new power module surge suppression method, characterized in that: include: Step 1: Make the auxiliary power supply work; Step 2: Continuously collect output voltage and input voltage values; Step 3: After determining that the current input voltage value has reached the peak value, determine whether to execute step 4 or return to step 2 based on the time queue of the input voltage and the time queue of the output voltage in the predetermined time period; Step 4: Output a control signal to control the rectifier device of the power module to connect to the rectifier circuit to charge the first capacitor; Step 5: If the current input voltage value is less than the difference between the current output voltage value and the preset threshold value, disconnect the rectifier device and return to step 3; if not, return to step 4; Step 6: Determine whether the difference between the current input voltage value and the output voltage value is less than the preset threshold value; if so, keep the rectifier device in the on state so that the power module enters the normal working mode; if not, return to step 3 and continue to perform input surge suppression; Wherein, the step 3 includes: Sequentially encoding the time queue of the input voltage and the time queue of the output voltage to obtain multi-scale time series implicit coding features of the input voltage and multi-scale time series implicit coding features of the output voltage; Performing feature interactive contrast processing on the input voltage multi-scale time series implicit coding features and the output voltage multi-scale time series implicit coding features using external knowledge modulation to obtain input voltage-output voltage time series contrast fine-grained interactive features; Performing state discrimination based on the input voltage-output voltage time sequence comparison fine-grained interaction feature to determine whether to execute step 4 or return to step 2; Among them, the time queue of the input voltage and the time queue of the output voltage are sequence encoded respectively to obtain an input voltage multi-scale time series implicit coding vector and an output voltage multi-scale time series implicit coding vector, including: inputting the time queue of the input voltage and the time queue of the output voltage into a sequence encoder based on an RNN-LSTM hybrid model respectively to obtain an input voltage multi-scale time series implicit coding vector as the input voltage multi-scale time series implicit coding feature and an output voltage multi-scale time series implicit coding vector as the output voltage multi-scale time series implicit coding feature.

2. The new power module surge suppression method according to claim 1 is characterized in that: Performing feature interactive comparison processing on the input voltage multi-scale time series implicit coding features and the output voltage multi-scale time series implicit coding features using external knowledge modulation to obtain input voltage-output voltage time series comparison fine-grained interactive features, including: Performing fine-grained feature interaction and optimization representation on the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector to obtain an external knowledge optimized input voltage-output voltage time series fine-grained feature interaction matrix; Optimizing the input voltage-output voltage time series fine-grained feature interaction matrix based on the external knowledge, and modulating the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector respectively to obtain an optimized input voltage multi-scale time series implicit coding vector and an optimized output voltage multi-scale time series implicit coding vector; The optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector are semantically contrasted and associated coded to obtain an input voltage-output voltage time series contrast fine-grained interaction feature vector as the input voltage-output voltage time series contrast fine-grained interaction feature.

3. The new power module surge suppression method according to claim 2, characterized in that: Performing fine-grained feature interaction and optimization representation on the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector to obtain an external knowledge optimized input voltage-output voltage time series fine-grained feature interaction matrix, including: Performing fine-grained feature interaction on the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector to obtain an input voltage-output voltage time series fine-grained feature interaction matrix; The input voltage-output voltage timing fine-grained feature interaction matrix is ​​passed through an attention unit based on external knowledge to obtain the external knowledge optimized input voltage-output voltage timing fine-grained feature interaction matrix.

4. The new power module surge suppression method according to claim 3 is characterized in that: The method includes optimizing the input voltage-output voltage time series fine-grained feature interaction matrix based on the external knowledge, modulating the input voltage multi-scale time series implicit coding vector and the output voltage multi-scale time series implicit coding vector respectively to obtain an optimized input voltage multi-scale time series implicit coding vector and an optimized output voltage multi-scale time series implicit coding vector, including: Optimizing the input voltage-output voltage time series fine-grained feature interaction matrix based on the external knowledge and performing fine-grained modulation optimization on the input voltage multi-scale time series implicit coding vector to obtain the optimized input voltage multi-scale time series implicit coding vector; Based on the external knowledge, the input voltage-output voltage time series fine-grained feature interaction matrix is ​​optimized to perform fine-grained modulation optimization on the output voltage multi-scale time series implicit coding vector to obtain the optimized output voltage multi-scale time series implicit coding vector.

5. The new power module surge suppression method according to claim 4 is characterized in that: Optimizing the input voltage-output voltage time series fine-grained feature interaction matrix based on the external knowledge and performing fine-grained modulation optimization on the input voltage multi-scale time series implicit coding vector to obtain the optimized input voltage multi-scale time series implicit coding vector includes: Performing a linear transformation on the input voltage multi-scale time series implicit coding vector to obtain an input voltage multi-scale time series query feature vector and an input voltage multi-scale time series value feature vector; The external knowledge is used to optimize the input voltage-output voltage timing fine-grained feature interaction matrix as a key matrix, and the input voltage multi-scale timing query feature vector, the input voltage multi-scale timing value feature vector and the key matrix are input into a fine-grained modulation module based on the Transformer structure to obtain the optimized input voltage multi-scale timing implicit coding vector.

6. The new power module surge suppression method according to claim 5, characterized in that: Optimizing the input voltage-output voltage time series fine-grained feature interaction matrix based on the external knowledge and performing fine-grained modulation optimization on the output voltage multi-scale time series implicit coding vector to obtain the optimized output voltage multi-scale time series implicit coding vector includes: Performing a linear transformation on the output voltage multi-scale time series implicit coding vector to obtain an output voltage multi-scale time series query feature vector and an output voltage multi-scale time series value feature vector; The external knowledge is used to optimize the input voltage-output voltage timing fine-grained feature interaction matrix as a key matrix, and the output voltage multi-scale timing query feature vector, the output voltage multi-scale timing value feature vector and the key matrix are input into the Transformer-based fine-grained modulation module to obtain the optimized output voltage multi-scale timing implicit coding vector.

7. The new power module surge suppression method according to claim 6, characterized in that: The optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector are semantically contrasted and associated coded to obtain an input voltage-output voltage time series contrast fine-grained interaction feature vector as the input voltage-output voltage time series contrast fine-grained interaction feature, including: calculating the position point division between the optimized input voltage multi-scale time series implicit coding vector and the optimized output voltage multi-scale time series implicit coding vector to obtain the input voltage-output voltage time series contrast fine-grained interaction feature vector.

8. The new power module surge suppression method according to claim 7, characterized in that: Based on the input voltage-output voltage timing comparison fine-grained interaction feature, state discrimination is performed to determine whether to execute step 4 or return to step 2, including: inputting the input voltage-output voltage timing comparison fine-grained interaction feature vector into a classifier-based state discriminator to obtain a discrimination result, and the discrimination result is used to indicate whether to execute step 4 or return to step 2.

Images