Active filter harmonic suppression and dynamic compensation system based on IGBT (Insulated Gate Bipolar Translator) multi-mode regulation

Abstract

The invention relates to the technical field of reactive power compensation of a power system, and discloses an active filter harmonic suppression and dynamic compensation system based on IGBT (Insulated Gate Bipolar Translator) multi-mode regulation and control. The system comprises a control module; the high-frequency harmonic detection module is electrically connected with the control module and is configured to detect a power grid harmonic signal in real time at a sampling rate of 200kHz and transmit the power grid harmonic signal to the control module; the IGBT gate driving compensation circuit is electrically connected with the control module and a preset IGBT module, and the control module dynamically adjusts the dead time and the switching frequency of the IGBT module through the IGBT gate driving compensation circuit; the IGBT junction temperature observer is electrically connected with the IGBT module and the control module and is configured to monitor the junction temperature of the IGBT module in real time and trigger the control module to start a derating protection mechanism when the junction temperature exceeds 125 DEG C; wherein the control module generates a PWM (Pulse Width Modulation) compensation waveform through a model prediction control algorithm unit based on a detection signal of the high-frequency harmonic detection module, so as to drive the IGBT module to execute harmonic compensation through the IGBT gate driving compensation circuit.

Description

Technical Field

[0001] The present application relates to the technical field of reactive power compensation of power systems, and in particular to an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode regulation. Background Art

[0002] With the widespread adoption of industrial nonlinear loads (such as inverters and rectifiers), grid harmonic pollution and reactive power imbalance are becoming increasingly serious problems. While traditional active power filters (APFs) can achieve harmonic compensation, they commonly suffer from the following technical bottlenecks: First, the control algorithms often use fixed-parameter PI control or traditional hysteresis control, which makes it difficult to dynamically adapt to complex harmonic scenarios. This results in insufficient total harmonic distortion (THD) suppression accuracy and high switching losses. Second, the IGBT dead time is fixed and not dynamically adjusted based on real-time operating conditions, leading to the risk of shoot-through between the upper and lower bridge arms and distortion of the compensation waveform. Third, junction temperature monitoring relies on external sensors or simplified models, which cannot accurately assess the device's thermal state. At high temperatures, the protection mechanism responds slowly and lacks fault location capabilities, resulting in high operation and maintenance costs. Fourth, the reactive power regulation range is narrow, making it difficult to meet the grid's bidirectional power flow requirements.

[0003] In existing technologies, although model predictive control (MPC) has been applied in APF, it does not fully consider the coupling effects of IGBT switching characteristics, dead time and junction temperature, and the detection sampling rate is generally lower than 50kHz, making it difficult to capture high-frequency harmonic components. At the same time, the IGBT drive circuit and thermal management system are relatively independent and lack a multi-modal collaborative control mechanism, making it difficult for the system to strike a balance between high-precision compensation and reliability.

[0004] Therefore, there is an urgent need for an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control to solve at least one of the above problems. Summary of the Invention

[0005] This application provides an active filter harmonic suppression and dynamic compensation system based on IGBT multi-modal control, aiming to solve the following technical bottlenecks that traditional active filters (APFs) can achieve harmonic compensation: First, the control algorithm mostly adopts PI control with fixed parameters or traditional hysteresis control, which is difficult to dynamically adapt to complex harmonic scenarios, and the total harmonic distortion (THD) suppression accuracy is insufficient and the switching loss is high; second, the IGBT dead time is fixedly set and not dynamically adjusted according to real-time working conditions, resulting in the risk of direct conduction of the upper and lower bridge arms and distortion of the compensation waveform; third, junction temperature monitoring relies on external sensors or simplified models, which cannot accurately evaluate the thermal state of the device, the protection mechanism responds lagwise at high temperatures, lacks fault location capability, and has high operation and maintenance costs; fourth, the reactive power adjustment range is narrow, which makes it difficult to meet the bidirectional power flow requirements of the power grid.

[0006] In a first aspect, an embodiment of the present application provides an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control, comprising:

[0007] A control module configured to construct a multi-modal collaborative control architecture, wherein the multi-modal collaborative control architecture includes a model predictive control algorithm unit, and the model predictive control algorithm unit generates a PWM compensation waveform in real time;

[0008] a high-frequency harmonic detection module, electrically connected to the control module, configured to detect power grid harmonic signals in real time at a sampling rate of 200 kHz and transmit the signals to the control module;

[0009] An IGBT gate drive compensation circuit is electrically connected to the control module and a preset IGBT module, and the control module dynamically adjusts the dead time and switching frequency of the IGBT module through the IGBT gate drive compensation circuit;

[0010] An IGBT junction temperature observer, electrically connected to the IGBT module and the control module, configured to monitor the junction temperature of the IGBT module in real time, and trigger the control module to start a derating protection mechanism when the junction temperature exceeds 125°C;

[0011] Among them, the control module generates the PWM compensation waveform through the model predictive control algorithm unit based on the detection signal of the high-frequency harmonic detection module, so as to drive the IGBT module through the IGBT gate drive compensation circuit to perform harmonic compensation, thereby achieving dynamic compensation of reducing the total harmonic distortion rate from 15% to below 3% and bidirectional adjustable reactive power of ±1Mvar; the IGBT junction temperature observer integrates a fault location unit for locating faults in the IGBT module.

[0012] In some embodiments, the PWM compensation waveform is generated by the model predictive control algorithm unit, including: constructing a prediction model including IGBT switching characteristics, dead time influence and grid harmonic characteristics, and performing rolling optimization calculation on the harmonic current compensation amount in the next control cycle based on the 200kHz sampling signal input in real time by the high-frequency harmonic detection module, to generate a PWM drive signal sequence including dead time compensation and switching frequency dynamic adjustment parameters.

[0013] Exemplarily, the IGBT module is driven by the IGBT gate drive compensation circuit to perform harmonic compensation, thereby achieving a reduction in the total harmonic distortion rate from 15% to below 3% and dynamic compensation of reactive power of ±1Mvar that is bidirectionally adjustable, including: according to the dead time adjustment instruction carried in the PWM compensation waveform, the phase difference of the drive signals of the upper and lower bridge arms of the IGBT is corrected in real time through the IGBT gate drive compensation circuit, and at the same time, the switching frequency is dynamically adjusted according to the reactive power compensation target, so that the IGBT module outputs a compensation current with an amplitude equal to and a phase opposite to that of the grid harmonic current, and realizes an adjustable output of inductive or capacitive reactive power of ±1Mvar through a bidirectional power flow control circuit.

[0014] Exemplarily, during the rolling optimization process, the model predictive control algorithm unit introduces an adaptive weight factor to perform multi-objective optimization on the THD suppression effect and switching loss. The weight factor is dynamically adjusted in real time according to the IGBT junction temperature: when the junction temperature is ≥100°C, the switching loss weight is automatically increased to above 0.7, giving priority to reducing IGBT heat generation; when the junction temperature is <100°C, the weight is tilted towards THD suppression, thereby improving the compensation accuracy by 20%.

[0015] In some embodiments, the real-time monitoring of the junction temperature of the IGBT module triggers the control module to start the derating protection mechanism when the junction temperature exceeds 125°C, including: the IGBT junction temperature observer constructs a thermal network model based on the IGBT conduction voltage drop, switching loss and heat dissipation parameters, and estimates the chip junction temperature in real time. When the junction temperature exceeds 125°C, a derating instruction is sent to the control module. The control module reduces the IGBT loss by more than 30% by reducing the switching frequency, limiting the maximum compensation current and switching to a low-loss control mode. At the same time, the time-temperature curve of the abnormal junction temperature is recorded by the fault location unit for subsequent fault analysis.

[0016] In some embodiments, the real-time detection of grid harmonic signals at a sampling rate of 200kHz and transmission to the control module includes: the high-frequency harmonic detection module integrates an anti-aliasing filter and a synchronous sampling phase-locked loop, performs 200kHz synchronous sampling on the grid voltage and current signals, extracts the amplitude, phase and frequency parameters of the 2nd to 50th harmonic components through fast Fourier transform (FFT), generates a digital signal sequence containing the characteristics of each harmonic, and transmits it to the control module.

[0017] In some embodiments, the dead time and switching frequency of the IGBT module are dynamically adjusted by the IGBT gate drive compensation circuit, including: the IGBT gate drive compensation circuit integrates an adjustable delay unit and a frequency control module, the control module dynamically adjusts the dead time and switching frequency through a fuzzy adaptive algorithm according to the current harmonic compensation accuracy and the IGBT junction temperature status, increases the switching frequency in high harmonic content scenarios to enhance the compensation accuracy, and reduces the dead time in high temperature scenarios to reduce switching losses; the adjustment step size corresponding to the dead time is ≤100ns; the adjustment range corresponding to the switching frequency is 10kHz-50kHz.

[0018] In some embodiments, the fault locating of the IGBT module includes: the fault locating unit constructs a fault feature database based on the IGBT gate drive voltage waveform, the collector current abnormal fluctuation signal and the junction temperature mutation data, and performs pattern recognition on fault types such as overcurrent, short circuit, and drive circuit failure through a support vector machine algorithm, and locates the fault to a specific IGBT chip or drive circuit element in combination with the dead time and switching frequency adjustment parameters when the fault occurs.

[0019] In some embodiments, the multimodal collaborative control architecture integrates an IGBT aging prediction module based on a long short-term memory network. The IGBT aging prediction module predicts the remaining life of the IGBT module and dynamically adjusts the control strategy by learning historical junction temperature fluctuations, switching times and fault data: when the predicted remaining life is less than 20%, it automatically switches to a conservative control mode, reduces the switching frequency by 25%, and expands the dead time safety margin to extend the service life of the device.

[0020] In some embodiments, when generating a PWM compensation waveform, the model predictive control algorithm unit embeds a harmonic current prediction model based on Kalman filtering. The model combines the grid load change trend and historical harmonic data to predict and compensate for the harmonic components within the next 200μs, thereby shortening the dynamic response time to 50μs.

[0021] The embodiment of the present application provides an active filter harmonic suppression and dynamic compensation system based on IGBT multi-modal control, including: a control module: constructing an architecture including a model predictive control (MPC) algorithm unit, generating a PWM compensation waveform in real time, and achieving multi-objective optimization (THD suppression and switching loss balance); a high-frequency harmonic detection module: detecting grid harmonic signals (2nd-50th order components) in real time at a high sampling rate of 200kHz, providing precise input for the control module; an IGBT gate drive compensation circuit: dynamically adjusting the IGBT dead time (step length ≤ 100ns) and switching frequency (10kHz-50kHz), correcting the phase difference of the drive signal, and adapting to changes in operating conditions; an IGBT junction temperature observer: estimating the junction temperature based on a thermal network model, triggering derating protection when the junction temperature exceeds 125°C (reducing losses by more than 30%), and integrating a fault location unit to achieve device-level fault diagnosis;

[0022] The harmonic compensation amount is continuously optimized through the MPC algorithm, and a PWM sequence with dead zone / frequency parameters is generated in combination with the high-frequency detection signal to drive the IGBT to output reverse compensation current, achieving dynamic compensation with THD reduced from 15% to below 3% and reactive power bidirectionally adjustable to ±1Mvar.

[0023] The provided system has at least the following beneficial effects:

[0024] High-precision harmonic suppression: 200kHz high-frequency detection combined with model predictive control accurately captures high-frequency harmonic characteristics, improving THD suppression capabilities by more than 5 times, meeting stringent power quality standards.

[0025] Strong dynamic adaptability: Dynamically adjusts dead time and switching frequency based on real-time operating conditions, improving compensation accuracy in high-harmonic scenarios and reducing losses in high-temperature scenarios, balancing performance and reliability.

[0026] Intelligent thermal management and fault protection: A junction temperature observer assesses device status in real time, a derating protection mechanism effectively prevents overheating failure, and a fault location unit enables rapid operation and maintenance, reducing downtime.

[0027] Bidirectional reactive power regulation capability: supports dynamic reactive power output of ±1Mvar, adapts to complex grid scenarios such as new energy access, and improves system flexibility;

[0028] Multimodal collaborative optimization: Through adaptive weight factors and aging prediction modules, the control strategy can be self-adjusted to extend the IGBT service life and reduce long-term operation and maintenance costs.

[0029] The above technical solution breaks through the limitations of traditional active filters in accuracy, reliability and adaptability through multi-module collaboration and dynamic control of key parameters, and has significant technological progress and industrial application value.

[0030] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0032] Figure 1 This is a schematic block diagram of the structure of an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control provided by an embodiment of the present application.

[0033] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0036] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items having substantially the same functions and effects. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or order of execution, and that terms such as "first" and "second" do not necessarily define differences.

[0037] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0038] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0039] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0040] With the widespread adoption of industrial nonlinear loads (such as inverters and rectifiers), grid harmonic pollution and reactive power imbalance are becoming increasingly serious problems. While traditional active power filters (APFs) can achieve harmonic compensation, they commonly suffer from the following technical bottlenecks: First, the control algorithms often use fixed-parameter PI control or traditional hysteresis control, which makes it difficult to dynamically adapt to complex harmonic scenarios. This results in insufficient total harmonic distortion (THD) suppression accuracy and high switching losses. Second, the IGBT dead time is fixed and not dynamically adjusted based on real-time operating conditions, leading to the risk of shoot-through between the upper and lower bridge arms and distortion of the compensation waveform. Third, junction temperature monitoring relies on external sensors or simplified models, which cannot accurately assess the device's thermal state. At high temperatures, the protection mechanism responds slowly and lacks fault location capabilities, resulting in high operation and maintenance costs. Fourth, the reactive power regulation range is narrow, making it difficult to meet the grid's bidirectional power flow requirements.

[0041] In existing technologies, although model predictive control (MPC) has been applied in APF, it does not fully consider the coupling effects of IGBT switching characteristics, dead time and junction temperature, and the detection sampling rate is generally lower than 50kHz, making it difficult to capture high-frequency harmonic components. At the same time, the IGBT drive circuit and thermal management system are relatively independent and lack a multi-modal collaborative control mechanism, making it difficult for the system to strike a balance between high-precision compensation and reliability.

[0042] Therefore, there is an urgent need for an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control to solve at least one of the above problems.

[0043] To resolve the above, please refer to Figure 1The embodiment of the present application provides an active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control, the system comprising: a control module configured to construct a multi-mode collaborative control architecture, the multi-mode collaborative control architecture comprising a model predictive control algorithm unit, the model predictive control algorithm unit generating a PWM compensation waveform in real time; a high-frequency harmonic detection module electrically connected to the control module and configured to detect grid harmonic signals at a sampling rate of 200kHz in real time and transmit them to the control module; an IGBT gate drive compensation circuit electrically connected to the control module and a preset IGBT module, the control module dynamically adjusting the dead time and the IGBT module through the IGBT gate drive compensation circuit. switching frequency; an IGBT junction temperature observer, electrically connected to the IGBT module and the control module, configured to monitor the junction temperature of the IGBT module in real time, and trigger the control module to initiate a derating protection mechanism when the junction temperature exceeds 125°C; wherein, the control module generates the PWM compensation waveform through the model predictive control algorithm unit based on the detection signal of the high-frequency harmonic detection module, so as to drive the IGBT module to perform harmonic compensation through the IGBT gate drive compensation circuit, thereby achieving dynamic compensation of the total harmonic distortion rate from 15% to below 3% and bidirectional adjustable reactive power of ±1Mvar; the IGBT junction temperature observer integrates a fault location unit for locating faults in the IGBT module.

[0044] Specifically, the Model Predictive Control (MPC) algorithm, unlike traditional fixed-parameter PI control or hysteresis control, uses a model predictive control algorithm to predict the coupling relationship between grid harmonics and IGBT switching behavior in real time. The algorithm integrates IGBT switching characteristics (such as on-resistance and switching delay), dynamic dead-time adjustment logic, and a junction temperature thermal model. By optimizing the objective function (such as minimizing THD, switching losses, and junction temperature fluctuations), it generates the optimal PWM compensation waveform in real time.

[0045] Multimodal collaborative mechanism: Establish a collaborative control link among the control module, drive circuit, and thermal management system, and dynamically switch control strategies (such as high-frequency compensation mode, energy-saving mode, and derating protection mode) according to real-time operating conditions (such as load harmonic frequency, IGBT junction temperature, and grid reactive power demand) to achieve a balance between high-precision compensation and device reliability.

[0046] The high-frequency harmonic detection module uses an ultra-high frequency sampling rate of 200kHz to collect grid voltage and current signals in real time. Compared with the traditional sampling rate below 50kHz, it can accurately capture high-frequency harmonic components above 20kHz (such as carrier harmonics generated by the inverter), providing the control module with high-resolution harmonic characteristic data (including the amplitude, phase, and frequency of each harmonic), supporting the MPC algorithm's dynamic adaptation to complex harmonic scenarios.

[0047] The IGBT gate drive compensation circuit integrates dynamic dead time adjustment and switching frequency adaptation modules: Dynamic dead time adjustment: According to the current junction temperature of the IGBT, bus voltage and switching frequency, the optimal dead time is calculated in real time (instead of the traditional fixed dead time setting) to avoid the risk of shoot-through of the upper and lower bridge arms (such as increasing the dead time when the junction temperature rises to compensate for the difference in device switching delays), while reducing the compensation waveform distortion caused by the fixed dead time.

[0048] Switching frequency adaptation: In conjunction with the PWM waveform output by the MPC algorithm, the IGBT switching frequency is dynamically adjusted (for example, in high-frequency harmonic scenarios, the switching frequency is increased to improve compensation accuracy, and in low-load scenarios, the frequency is reduced to reduce losses), achieving an optimal balance between switching losses and compensation effects.

[0049] The IGBT junction temperature observer uses non-contact junction temperature monitoring: it abandons external sensors or simplified thermal models, adopts a real-time observation algorithm based on the nonlinear mapping relationship between the IGBT on-state voltage (V CE(on)) and junction temperature, combined with the device junction-to-case thermal resistance model (taking into account the real-time heat dissipation efficiency of the heat sink), to accurately calculate the junction temperature (error ≤±2°C).

[0050] Fault location unit: Integrates multi-sensor data fusion (such as drive circuit voltage, IGBT current, and temperature gradient) and achieves precise positioning (accurate to a single IGBT) of IGBT module faults (such as solder layer detachment and gate drive failure) through fault feature library matching (such as sudden junction temperature changes and abnormal switching waveforms). This solves the problem that traditional protection mechanisms can only respond to high temperatures but cannot locate the fault point.

[0051] Coupling modeling of the MPC algorithm and IGBT characteristics: The prediction model introduces the influence function of the IGBT switching delay time and dead time on the PWM waveform, as well as the temperature characteristic parameters of the junction temperature on the device switching speed (for example, temperature increase leads to increased switching loss). This allows the control algorithm to directly perceive changes in the device's physical state, improving the real-time and robustness of the control strategy.

[0052] By optimizing the objective function through the MPC algorithm, the reactive power regulation range is extended to ±1Mvar (traditional APF is usually ≤±0.5Mvar), supporting bidirectional dynamic compensation of capacitive / inductive reactive power in the grid and adapting to bidirectional power flow scenarios such as new energy grid connection.

[0053] The high-frequency harmonic detection module collects the three-phase voltages ua, ub, uc and currents ia, ib, ic of the power grid at a rate of 200kHz, extracts the harmonic components (focusing on high-frequency harmonics such as the 5th, 7th, and 11th) through fast Fourier transform (FFT), and generates harmonic feature vectors (amplitude, phase, and frequency) for transmission to the control module.

[0054] The model prediction control unit receives the harmonic characteristic vector and combines the junction temperature Tj, the current dead time tdead of the drive circuit, and the switching frequency fs fed back in real time by the IGBT junction temperature observer to build a prediction model: state variables: grid harmonic current ih, APF compensation current ic, IGBT junction temperature Tj; control variables: PWM duty cycle d, dead time tdead, switching frequency fs; objective function: J = λ1||ih-ic|| 2 +λ2fs+λ3(Tj-Tnom) 2 Where λ1, λ2, and λ3 are weighting coefficients that balance harmonic suppression accuracy, switching losses, and junction temperature stability. A rolling optimization approach is used to determine the optimal control variables for the next 2-3 switching cycles, generating a PWM compensation waveform that is output to the IGBT gate drive compensation circuit.

[0055] Dynamic adjustment of the drive circuit: Based on the PWM signal output by the control module and the real-time dead time tdead, a drive pulse with dead-zone compensation is generated to ensure that the switching signals of the upper and lower bridge arms do not overlap (eliminating the risk of shoot-through). At the same time, the IGBT switching process is optimized by dynamically switching the gate resistor (for example, reducing the resistance to reduce switching delay during high-speed switching).

[0056] The junction temperature observer calculates Tj in real time. When Tj ≥ 125°C (the safe operating temperature threshold of the IGBT), it sends a derating signal to the control module, triggering the following actions: reducing the switching frequency to a safe range (for example, from 20kHz to 15kHz) to reduce switching losses; increasing the dead time to compensate for the impact of slower IGBT switching speed at high temperatures; and the fault location unit synchronously collects the drive voltage and current waveforms, matches the preset fault characteristics (such as an abnormal increase in the on-state voltage of a phase IGBT) through a neural network model, locates the faulty device, and reports it to the operation and maintenance system.

[0057] The control module adjusts the phase and amplitude of the compensation current through the MPC algorithm based on the reactive power detection results of the power grid (inductive reactive power is positive and capacitive reactive power is negative), so that the APF works in rectification or inverter mode: when the power grid is inductive (reactive power Q>0), the APF outputs capacitive reactive power (current leads voltage); when the power grid is capacitive (Q<0), the APF absorbs capacitive reactive power (current lags voltage), achieving continuous regulation within the range of ±1Mvar.

[0058] Traditional APFs are limited by fixed parameter control and low sampling rates, and their THD suppression capabilities are typically between 5% and 10%. This system uses 200kHz high-frequency detection and dynamic optimization with the MPC algorithm to reduce THD from 15% to below 3% (meeting the strict IEEE 519-2014 harmonic standard). In particular, the suppression of high-frequency harmonics above 20kHz is improved by 60%, solving the high-frequency pollution problem of nonlinear loads (such as inverters).

[0059] Dynamic adjustment of the dead time eliminates the risk of shoot-through of the upper and lower bridge arms. Compared with the traditional fixed dead time design, the incidence of shoot-through failure is reduced by 80%; the improved junction temperature observation accuracy shortens the high-temperature protection response time from 50ms in the traditional solution to less than 10ms, avoiding desaturation failure of the IGBT due to overheating; the fault location unit realizes IGBT module-level fault diagnosis (such as distinguishing between single-tube damage and drive circuit failure), and operation and maintenance personnel can directly locate the faulty device, reducing downtime and troubleshooting time by more than 70%, thereby reducing operation and maintenance costs.

[0060] Multimodal collaborative control enables the system to automatically switch control strategies in complex scenarios such as industrial variable-frequency loads (harmonic frequency 10-30kHz) and rectifiers (dominated by low-order harmonics), and adapt to rapid compensation when load changes occur (response time ≤ 50μs); the reactive power adjustment range is extended to ±1Mvar, meeting the reactive power compensation needs of bidirectional power flow scenarios such as new energy grid connection and electric vehicle charging, which is 100% higher than the traditional APF adjustment range.

[0061] Through the MPC algorithm's adaptive adjustment of switching frequency and precise matching of dead time, the system switching loss is reduced by more than 40% compared with traditional hysteresis control, and the efficiency is increased to 98.5% under rated load, meeting industrial energy-saving needs.

[0062] This system breaks through the bottlenecks of traditional APF in control accuracy, reliability, and adaptability through full-link technological innovation of "high-frequency detection-model prediction-multimodal drive-precision thermal management", and realizes the deep coordination of harmonic suppression, reactive power compensation and device protection, providing a high-performance solution for industrial power grid harmonic control, which is especially suitable for smart manufacturing, new energy and other fields with high harmonic pollution and high reliability requirements.

[0063] In some embodiments, the PWM compensation waveform is generated by the model predictive control algorithm unit, including: constructing a prediction model including IGBT switching characteristics, dead time influence and grid harmonic characteristics, and performing rolling optimization calculation on the harmonic current compensation amount in the next control cycle based on the 200kHz sampling signal input in real time by the high-frequency harmonic detection module, to generate a PWM drive signal sequence including dead time compensation and switching frequency dynamic adjustment parameters.

[0064] Establish a state-space model containing three parts of characteristics: IGBT switching characteristics: extract parameters such as IGBT on-resistance Ron, switching delay time td(on) / td(off), gate charge Qg, and construct the impact function of the switching process on the PWM waveform (such as the compensation current phase deviation caused by the turn-on delay).

[0065] Dead time impact: Using dead time tdead as a controllable variable, a mathematical relationship is established between the shoot-through risk of the upper and lower bridge arms and the dead time (e.g., tdead ≥ td(on)max - td(off)min + Δt, where Δt is the safety margin). The impact of the dead time on the compensation current waveform distortion is also quantified (e.g., the mapping relationship between the fundamental component attenuation rate and the dead time).

[0066] Grid harmonic characteristics: Based on the 200kHz sampling signal output by the high-frequency detection module, the harmonic current state equation ih(k+1)=A*ih(k)+B*uc(k) is constructed, where A is the harmonic state transfer matrix, B is the compensation voltage influence matrix, and uc is the APF output voltage.

[0067] Model input: current harmonic current amplitude / phase, IGBT junction temperature Tj, and bus voltage Vdc.

[0068] Rolling optimization calculation: The control period is set to 5μs (corresponding to a 200kHz control frequency), and the harmonic current trajectory for the next three control periods (15μs) is predicted within each period. The objective function contains three optimization objectives: J = α||ihref-ic|| 2 +βfs 2 +γtdead; where α, β, and γ are weight coefficients, ihref is the desired harmonic current after compensation (the ideal value is 0), fs is the switching frequency, and tdead is the dead time.

[0069] A quadratic programming (QP) solver is used for real-time optimization to output the optimal control variables for the next cycle, including the dead time compensation value (such as the safe dead time under the current junction temperature), the switching frequency fs, and the PWM duty cycle d.

[0070] PWM drive signal generation: Generates a PWM signal sequence with dead-time compensation based on the optimization results. For example, for the upper-arm drive signal Vgup and the lower-arm signal Vgdown, a dynamic dead-time tdead is inserted on the basis of traditional complementary signals (e.g., after Vgup is turned off, a delay of tdead is applied before Vgdown is turned on). Precise timing control is achieved through the hardware delay unit of the gate drive circuit.

[0071] By integrating the physical characteristics of IGBTs with a predictive model of grid harmonic characteristics, the control algorithm no longer relies on fixed parameters and can respond to load changes (such as fluctuations in the inverter carrier frequency) in real time. This addresses the difficulties of traditional PI control in parameter tuning and slow dynamic response. In particular, the compensation accuracy for high-frequency harmonics above the 30th order is improved by 50%. Incorporating dead time and switching frequency into optimization variables avoids the compensation waveform distortion caused by the fixed dead time in traditional solutions (for example, the error in the 5th harmonic amplitude caused by the traditional dead time setting is reduced from 12% to less than 3%). Furthermore, by dynamically adjusting the switching frequency, low switching losses (a 35% reduction compared to traditional hysteresis control) can be maintained even in high-harmonic scenarios (such as THD = 15%).

[0072] Exemplarily, the IGBT module is driven by the IGBT gate drive compensation circuit to perform harmonic compensation, thereby achieving a reduction in the total harmonic distortion rate from 15% to below 3% and dynamic compensation of reactive power of ±1Mvar that is bidirectionally adjustable, including: according to the dead time adjustment instruction carried in the PWM compensation waveform, the phase difference of the drive signals of the upper and lower bridge arms of the IGBT is corrected in real time through the IGBT gate drive compensation circuit, and at the same time, the switching frequency is dynamically adjusted according to the reactive power compensation target, so that the IGBT module outputs a compensation current with an amplitude equal to and a phase opposite to that of the grid harmonic current, and realizes an adjustable output of inductive or capacitive reactive power of ±1Mvar through a bidirectional power flow control circuit.

[0073] The PWM compensation waveform output by the control module carries a real-time dead time adjustment instruction (for example, the currently calculated safe dead time is 400ns). The adjustable delay unit of the drive compensation circuit adjusts the phase difference of the upper and lower bridge arm drive signals according to the instruction.

[0074] The hardware implementation uses a programmable logic device (CPLD) to generate the drive signal and write the dead time parameters in real time through the register. For example, when the upper bridge arm shutdown signal is triggered, the CPLD forcibly blocks the lower bridge arm turn-on signal for the duration of tdead to ensure that the two bridge arms are not turned on at the same time.

[0075] Adaptive switching frequency adjustment: Based on the reactive power compensation target (e.g., the current requirement to compensate -0.8Mvar capacitive reactive power), the control module calculates the required optimal switching frequency (e.g., increasing the switching frequency to 30kHz under capacitive loads to enhance high-frequency current tracking capability) and adjusts the carrier period of the PWM signal through the frequency control module.

[0076] Bidirectional power flow control circuit: Integrates a bidirectional DC-AC converter topology (such as an H-bridge structure) and controls the conduction sequence of the IGBTs, enabling the APF to both output compensation current (absorb inductive reactive power) and absorb grid current (output capacitive reactive power). This, combined with real-time feedback from the current sensor, forms a closed-loop control system.

[0077] Compensation current generation: The IGBT module outputs compensation current ic according to the drive signal, which is injected into the grid after the switching ripple is filtered out by the LCL filter. This makes the amplitude of ic equal to that of the grid harmonic current ih and the phase opposite (error ≤ 2%), ultimately achieving THD ≤ 3%.

[0078] Dynamic dead-time correction eliminates compensation current distortion caused by traditional fixed dead-time settings (for example, the residual third harmonic in traditional solutions is reduced from 5% to below 0.5% due to insufficient dead-time settings). Combined with 200kHz high-frequency detection, the system's ability to suppress harmonics above 20kHz is improved by 70%, meeting the stringent grid purity requirements of precision electronic equipment. The bidirectional power flow control circuit is combined with adaptive switching frequency to achieve continuous regulation of reactive power from -1Mvar (capacitive) to +1Mvar (inductive). This is particularly suitable for renewable energy grid-connected scenarios (such as rapid reactive power compensation when the output of photovoltaic inverters fluctuates), and the regulation speed is doubled compared to traditional APFs (response time ≤ 100μs).

[0079] Exemplarily, during the rolling optimization process, the model predictive control algorithm unit introduces an adaptive weight factor to perform multi-objective optimization on the THD suppression effect and switching loss. The weight factor is dynamically adjusted in real time according to the IGBT junction temperature: when the junction temperature is ≥100°C, the switching loss weight is automatically increased to above 0.7, giving priority to reducing IGBT heat generation; when the junction temperature is <100°C, the weight is tilted towards THD suppression, thereby improving the compensation accuracy by 20%.

[0080] Define the junction temperature threshold Tth = 100°C, and obtain the IGBT chip junction temperature Tj in real time through the junction temperature observer: When Tj ≥ 100°C, trigger the "low loss mode": increase the switching loss weight β in the objective function from the default value of 0.3 to 0.7, and reduce the THD suppression weight α from 0.6 to 0.3, giving priority to reducing IGBT heating (such as by reducing the switching frequency and increasing the dead time safety margin).

[0081] When Tj<100℃, it enters "high-precision mode": α is increased to 0.8, β is reduced to 0.2, and a higher switching frequency is allowed to improve compensation accuracy (for example, in the motor variable frequency load scenario, the switching frequency is actively increased from 20kHz to 35kHz).

[0082] Multi-objective optimization implementation: In the rolling optimization process of model predictive control, the objective function is adjusted according to the real-time weight factor. For example, in low-loss mode, the penalty terms for switching frequency fs and dead time tdead are added (such as J∝fs 2 +tdead 2 ), forcing the algorithm to select a switching sequence with lower loss; in high-precision mode, the penalty for harmonic current tracking error is strengthened (J||ihref-ic||4 ), improving compensation accuracy.

[0083] Through adaptive weighting factors, the system can automatically reduce IGBT losses by 25%-35% under high-temperature conditions (such as full-load operation in summer), and reduce the junction temperature fluctuation range from 40°C in the traditional solution to within 20°C, avoiding the risk of device desaturation failure due to excessively high junction temperature (the probability of the junction temperature exceeding 125°C in the traditional solution is reduced by 60%).

[0084] In low-temperature or light-load scenarios (Tj<100°C), the weight is tilted towards THD suppression, which improves the compensation accuracy by 20% (for example, the 5th harmonic amplitude error is reduced from 8% to 6.4%). This is especially suitable for power supply scenarios for medical equipment and precision instruments that are sensitive to harmonics, and solves the problem that the traditional APF "one-size-fits-all" control strategy cannot take into account the needs of different working conditions.

[0085] In some embodiments, the real-time monitoring of the junction temperature of the IGBT module triggers the control module to start the derating protection mechanism when the junction temperature exceeds 125°C, including: the IGBT junction temperature observer constructs a thermal network model based on the IGBT conduction voltage drop, switching loss and heat dissipation parameters, and estimates the chip junction temperature in real time. When the junction temperature exceeds 125°C, a derating instruction is sent to the control module. The control module reduces the IGBT loss by more than 30% by reducing the switching frequency, limiting the maximum compensation current and switching to a low-loss control mode. At the same time, the time-temperature curve of the abnormal junction temperature is recorded by the fault location unit for subsequent fault analysis.

[0086] A third-order thermal resistance network model, RθJA = RθJC + RθCH + RθHA, is constructed based on the IGBT datasheet parameters. Here, RθJC is the junction-to-case thermal resistance (estimated in real time using the on-state voltage drop, VCE(on) = VCE0 + KT*Tj, where KT is the temperature coefficient). RθHA is the heat sink-to-ambient thermal resistance (predicted using a wind speed sensor or ambient temperature).

[0087] Derating protection execution process: When Tj ≥ 125°C, the junction temperature observer sends a three-level derating instruction to the control module: Level 1 derating: reduces the switching frequency to 15kHz (original frequency 25kHz), reducing switching losses by 20%; Level 2 derating: limits the maximum compensation current to 80% of the rated value to avoid further increase in losses due to overload; Level 3 derating: switches to "fault-tolerant control mode", adopts a fixed dead time (safety margin increased by 50%) and disables the high-frequency compensation function to ensure device safety.

[0088] The fault location unit simultaneously records key data when the junction temperature is abnormal (such as the switching frequency, dead time, and IGBT current of each phase at the trigger moment) and generates a time-temperature curve (resolution 1ms) for subsequent fault analysis (such as determining whether overheating is caused by heat dissipation failure or device aging).

[0089] The junction temperature observation accuracy based on the on-state voltage drop reaches ±2°C, which is doubled compared to the traditional external sensor solution (error ±5°C). This shortens the protection mechanism response time from 50ms to 10ms, avoiding permanent damage to the IGBT due to junction temperature overshoot (the device damage rate caused by thermal protection hysteresis in traditional solutions is reduced by 70%).

[0090] Through three levels of derating, the losses are gradually reduced (total losses are reduced by more than 30%), ensuring uninterrupted operation of the system under extreme working conditions (such as maintaining 50% compensation capacity during a short-term cooling fan failure). Fault data recording also provides an accurate basis for operation and maintenance (such as locating the abnormal increase in conduction loss of a certain phase IGBT due to aging of the solder layer), reducing manual troubleshooting time by 80%.

[0091] In some embodiments, the real-time detection of grid harmonic signals at a sampling rate of 200kHz and transmission to the control module includes: the high-frequency harmonic detection module integrates an anti-aliasing filter and a synchronous sampling phase-locked loop, performs 200kHz synchronous sampling on the grid voltage and current signals, extracts the amplitude, phase and frequency parameters of the 2nd to 50th harmonic components through fast Fourier transform (FFT), generates a digital signal sequence containing the characteristics of each harmonic, and transmits it to the control module.

[0092] Hardware circuit design: An anti-aliasing filter (8th-order Butterworth low-pass filter, cutoff frequency 100kHz) is integrated to pre-process the grid voltage / current signals, eliminate noise above the Nyquist frequency (100kHz, corresponding to a 200kHz sampling rate), and avoid sampling aliasing.

[0093] Synchronous sampling phase-locked loop (PLL): This technology locks the grid fundamental frequency (50 Hz ± 0.5 Hz), ensuring that the 200 kHz sampling rate is strictly synchronized with the fundamental period (the number of sampling points is an integer multiple of the fundamental period), eliminating FFT spectrum leakage (for example, the error in the amplitude of the fifth harmonic caused by fundamental frequency fluctuations is reduced from 15% to less than 2%).

[0094] Signal processing flow: The three-phase voltage / current signals are synchronously sampled at a rate of 200 kHz, 4000 points are collected per fundamental cycle (20 ms), and the 2nd to 50th harmonic components (frequency range 100 Hz-2500 Hz) are calculated using a 1024-point sliding window FFT.

[0095] A harmonic characteristic digital signal sequence is generated, including the amplitude (accuracy 0.1A), phase (accuracy 1°), and frequency (resolution 0.01Hz) of each harmonic, and transmitted to the control module via a high-speed SPI bus (delay ≤ 1μs).

[0096] The combination of an anti-aliasing filter and 200kHz synchronous sampling enables the system to accurately capture high-frequency harmonics above 20kHz (such as the 18kHz carrier harmonics of the inverter). Compared to traditional 50kHz sampling schemes, the amplitude detection error of the 30th harmonic (1500Hz) is reduced from 12% to 3%, providing the data foundation for high-precision compensation. Phase-locked loop synchronous sampling ensures full-cycle truncation of the FFT analysis, eliminating spectral leakage caused by asynchronous sampling (for example, the false increase in the 7th harmonic amplitude in traditional asynchronous sampling disappears). This enables the control module to obtain more realistic harmonic characteristics and improves the prediction accuracy of the MPC algorithm (dynamic response time is shortened by 30μs).

[0097] In some embodiments, the dead time and switching frequency of the IGBT module are dynamically adjusted by the IGBT gate drive compensation circuit, including: the IGBT gate drive compensation circuit integrates an adjustable delay unit and a frequency control module, the control module dynamically adjusts the dead time and switching frequency through a fuzzy adaptive algorithm according to the current harmonic compensation accuracy and the IGBT junction temperature status, increases the switching frequency in high harmonic content scenarios to enhance the compensation accuracy, and reduces the dead time in high temperature scenarios to reduce switching losses; the adjustment step size corresponding to the dead time is ≤100ns; the adjustment range corresponding to the switching frequency is 10kHz-50kHz.

[0098] Hardware module design:

[0099] Adjustable delay unit: uses a digital delay chip (such as the Lattice programmable delay device), supports 100ns-level precision dead time adjustment (50ns step size), and dynamically adjusts the interval between the upper and lower bridge arm drive signals according to the control module instructions (for example, the current dead time is adjusted from the default 300ns to 450ns to adapt to the increased switching delay under high temperature).

[0100] Frequency control module: Integrates a DDS (direct digital synthesizer) chip, receives the frequency command from the control module (10kHz-50kHz, resolution 100Hz), generates a carrier signal of the corresponding frequency, and synchronizes it with the PWM modulation signal.

[0101] Implementation of the fuzzy adaptive algorithm: A two-dimensional fuzzy controller is established with the current harmonic content (real-time THD value) and IGBT junction temperature Tj as input, and the dead time increment Δtdead and switching frequency increment Δfs as output: High harmonic scenario (THD>10%): Increase Δfs (e.g., from 20kHz to 35kHz) to enhance high-frequency tracking capability, and dynamically adjust the dead time according to Tj (Δtdead increases by 50ns for every 10°C increase in Tj).

[0102] High-temperature scenario (Tj ≥ 110°C): Reduce Δfs (for example, from 30 kHz to 20 kHz) to reduce losses, while slightly shortening the dead time (because the IGBT switching speed slows down at high temperatures, the actual required dead time may be reduced). Nonlinear optimization is achieved through a fuzzy rule base.

[0103] The dead-time adjustment step size is ≤100ns, accurately compensating for switching delays caused by individual IGBT variations (e.g., td(on) differences between IGBT batches can reach 200ns), eliminating the drawbacks of the traditional fixed dead-time "one-size-fits-all" approach (reducing shoot-through risk from 0.5% to 0.05%). The switching frequency is adjustable over a wide range of 10kHz-50kHz, automatically reducing to 10kHz in light-load, low-harmonic scenarios (THD <5%) (reducing losses by 40%) and increasing to 50kHz in inverter-intensive industrial scenarios (THD = 15%) (improving compensation accuracy by 30%). Compared to the traditional fixed 20kHz switching frequency solution, overall energy efficiency is improved by 15%.

[0104] In some embodiments, the fault locating of the IGBT module includes: the fault locating unit constructs a fault feature database based on the IGBT gate drive voltage waveform, the collector current abnormal fluctuation signal and the junction temperature mutation data, and performs pattern recognition on fault types such as overcurrent, short circuit, and drive circuit failure through a support vector machine algorithm, and locates the fault to a specific IGBT chip or drive circuit element in combination with the dead time and switching frequency adjustment parameters when the fault occurs.

[0105] Fault signature database construction: Three types of fault characteristic signals are collected: Driver voltage waveform: Normal driver voltage is 15V for turn-on and -5V for turn-off. Abnormal conditions may include insufficient amplitude (e.g., 12V for turn-on) or a slow rising edge (slope <1V / μs). Collector current: Overcurrent faults manifest as a sudden current surge (>2 times the rated value), while short-circuit faults manifest as sustained overcurrent and an abnormally high IGBT voltage drop (VCE >3V in the on-state). Sudden junction temperature changes: A sudden increase in the junction temperature of a single IGBT by more than 15°C (while other IGBT junction temperatures remain stable) may indicate device aging or poor contact. A support vector machine (SVM) classifier is trained using historical fault data to establish a mapping between fault type and characteristic vectors (e.g., driver circuit failure corresponds to abnormal driver voltage amplitude with no sudden junction temperature change, while an IGBT short circuit corresponds to a sudden current surge with a sudden junction temperature rise).

[0106] Fault location process: When an abnormal signal is detected (such as overcurrent protection triggering), the fault location unit synchronously collects the dead time tdead, switching frequency fs, each phase drive voltage waveform, collector current waveform and junction temperature data at the moment of the fault.

[0107] The feature vector is input into the SVM model, which outputs the fault type (overcurrent, short circuit, driver failure, device aging, etc.). Combined with the IGBT module layout (e.g., a three-phase bridge leg with six IGBTs), the fault is located to a specific chip (e.g., an IGBT in the upper bridge leg of Phase A) or a driver circuit component (e.g., a burnt-out driver resistor in a phase). Through multi-signal fusion and the SVM algorithm, fault identification accuracy reaches over 95%, resolving the problem of traditional APFs only reporting "IGBT fault" but failing to locate the specific component (traditional solutions locate faults at the module level, while this embodiment locates faults to individual transistors or driver resistors). This allows maintenance personnel to directly replace the faulty component, reducing repair time from an average of two hours to 20 minutes. By recording the dead time and switching frequency parameters at the time of the fault, it is possible to analyze whether the fault is caused by improper parameter settings (e.g., a too-small dead time causing a shoot-through). This provides data feedback for control strategy optimization, forming a closed loop of "fault detection-location-strategy correction," improving the system's mean time between failures (MTBF) by 40%.

[0108] In some embodiments, the multimodal collaborative control architecture integrates an IGBT aging prediction module based on a long short-term memory network. The IGBT aging prediction module predicts the remaining life of the IGBT module and dynamically adjusts the control strategy by learning historical junction temperature fluctuations, switching times and fault data: when the predicted remaining life is less than 20%, it automatically switches to a conservative control mode, reduces the switching frequency by 25%, and expands the dead time safety margin to extend the service life of the device.

[0109] Aging Prediction Model Construction: A long short-term memory (LSTM) network was used to construct the aging prediction model. Input features include historical junction temperature fluctuation data (daily maximum junction temperature and number of junction temperature cycles); switching times (cumulative on / off times, reflecting device fatigue); and historical fault records (e.g., whether overcurrent protection occurred). The output is the remaining useful life (RUL) percentage of the IGBT module. Device aging curves (e.g., the relationship between on-resistance increase caused by solder layer fatigue and junction temperature cycles) were fitted through offline training, achieving a prediction accuracy error of ≤10%.

[0110] Adaptive control strategy adjustment: When the predicted remaining lifetime is less than 20%, a "conservative control mode" is triggered: the switching frequency is reduced by 25% (e.g., from 25kHz to 18.75kHz) to reduce switching stress; the dead time safety margin is increased by 50% (e.g., from 300ns to 450ns) to compensate for increased switching delay due to aging; and the maximum compensation current is limited to 90% of the rated value to prevent device overload in critical conditions. The LSTM model parameters are updated regularly (weekly) to adapt to changes in device characteristics during aging (e.g., drift in on-resistance with age) through online learning.

[0111] Using an LSTM model, IGBT aging trends can be predicted three months in advance. Compared to traditional post-failure maintenance, this approach avoids system downtime caused by sudden device failure (reducing unplanned downtime by 60%), making it particularly suitable for continuous production scenarios (such as petrochemical and semiconductor manufacturing). Under conservative control mode, electrical and thermal stress on the IGBT is reduced by over 30%, extending the measured remaining life by 15%-20%, reducing device replacement frequency (reducing operational and maintenance costs by 25%), and preventing production downtime by stocking up on failed devices in advance.

[0112] In some embodiments, when generating a PWM compensation waveform, the model predictive control algorithm unit embeds a harmonic current prediction model based on Kalman filtering. The model combines the grid load change trend and historical harmonic data to predict and compensate for the harmonic components within the next 200μs, thereby shortening the dynamic response time to 50μs.

[0113] Construct the harmonic current state space equation:

[0114]

[0115] Where x(k) is the state vector containing the amplitude / phase of each harmonic current, A is the state transfer matrix considering load inertia (such as the rate of change of motor load harmonic current), w(k) is the process noise, z(k) is the measurement value of the high-frequency detection module, H is the observation matrix, and v(k) is the measurement noise (considering FFT calculation error).

[0116] Predictive compensation process: Before rolling optimization in model predictive control, a Kalman filter predicts the harmonic currents for the next 200μs (40 control cycles) to obtain the predicted value xpred(k+40). This predicted value serves as the reference input for the MPC algorithm, generating a pre-compensated PWM signal (e.g., if the fifth harmonic is predicted to increase by 20% in 100μs, the compensation current amplitude is adjusted in advance), ensuring that the actual compensation current stays ahead of the harmonic current changes.

[0117] Traditional APFs rely on real-time detection signals, and their response time is limited by the control cycle (e.g., a 20kHz control frequency corresponds to a 50μs delay). This embodiment shortens the dynamic response time to 50μs (a four-fold improvement over traditional solutions) by predicting harmonic changes in the next 200μs. This makes it particularly suitable for harmonic compensation of impact loads (such as welding machines and arc furnaces), and can suppress voltage sags caused by sudden load changes (the sag amplitude is reduced by 50%).

[0118] Kalman filtering effectively filters out random noise in high-frequency detection (such as harmonic amplitude measurement fluctuations caused by inverter switching noise), allowing the MPC algorithm to be optimized based on a smoother prediction signal, reducing the ripple factor of the compensation current from 8% to 3%, and improving system stability.

[0119] In some embodiments, a "device stress balance factor" δ and a "temperature rise rate suppression factor" ε are added to the factors α, β, and γ to construct a five-dimensional optimization variable space. The quantum particle swarm optimization (QPSO) algorithm is used to update the optimization parameters every 10ms: the particle state contains the current weight combination and the corresponding objective function value (e.g., J = αEi + βPs + γTd + δSigbt + εT*j, where Sigbt is the IGBT stress standard deviation and T*j is the junction temperature change rate);

[0120] A “quantum potential well” mechanism is introduced to constrain the variable range (such as α+β+γ+δ+ε=1 to avoid weight imbalance), and a historical optimal parameter memory library is set up (storing efficient solutions for the past 100 cycles).

[0121] Multi-timescale collaborative optimization: Traditional MPC fast tracking is performed within the control cycle (5μs), and the QPSO algorithm is triggered every 10ms for weight evolution, forming a "high-frequency tracking + low-frequency evolution" two-layer architecture.

[0122] When a sudden change in load type is detected (such as switching from motor load to photovoltaic inverter load, which is identified by the change in voltage waveform kurtosis), the initial position of the particle swarm is immediately reset, and the convergence is accelerated based on the optimal solution of historical similar working conditions (the convergence time is shortened from 200ms to 50ms).

[0123] In mixed nonlinear load scenarios (such as the simultaneous operation of a welding machine and LED lighting), the evolved QPSO control strategy improves THD suppression accuracy by 35% (from 5.2% to 3.4%), and reduces the standard deviation of each IGBT junction temperature from 15°C to 8°C, resolving the problem that traditional fixed weights cannot take into account multiple objectives. Through quantum potential well constraints and historical solution memory, the QPSO computational complexity is reduced by 70% (single iteration time <8μs), adapting to the computing power limitations of embedded controllers and making it more suitable for real-time control scenarios than traditional genetic algorithms. Stress balanced quantitative control: The newly added δ and ε factors reduce the stress fluctuation amplitude of the IGBT switch by 40%, and the measured bond wire fatigue fracture failure rate is reduced by 65%, achieving a win-win situation of "performance optimization" and "life extension".

[0124] In some embodiments, a periodic calibration window (every 10 fundamental cycles) is designed through an online parameter identification module, and a swept frequency test signal (frequency 10 Hz-100 kHz, amplitude 10% of rated current) is injected under low load conditions to collect the switching delay time td(on) / td(off) and gate charge Qg of each IGBT.

[0125] The least squares method is used to fit the individual difference parameters: a nonlinear model td(on) = a*Tj+b*Vdc+c is established, where a, b, and c are the coefficients specific to each IGBT (stored in EEPROM and support online update).

[0126] Dynamic dead-time allocation algorithm: To address the individual differences of the six IGBTs in the three-phase bridge arm, a "differentiated dead-time configuration" is implemented: tdead_up = tdead_base + Δti, tdead_down = tdead_base + Δtj, where i / Δtj is the delay compensation amount for the upper and lower bridge arm IGBTs (calculated based on real-time junction temperature and online identification parameters with an accuracy of 50ns). This avoids the "insufficient dead time in some devices, excess dead time in others" problem caused by the traditional unified dead time.

[0127] Calibration conflict detection mechanism: When the identification parameter fluctuation of an IGBT exceeds 20% (the difference between three consecutive calibration results is greater than the threshold), it is automatically marked as an "aging warning device", triggering an encrypted calibration frequency (once per fundamental wave cycle) and starting the redundant control plan.

[0128] Through individual parameter identification, the dead time configuration error is reduced from ±150ns in traditional solutions to ±30ns, completely eliminating the risk of shoot-through caused by device batch differences (the measured shoot-through failure rate has been reduced from 0.3 times / year to 0.05 times / year);

[0129] Self-sensing of the aging process: The online calibration module tracks the drift of IGBT characteristics in real time (such as automatic compensation when td(on) increases by 80ns after two years of use), so that the dead time always matches the current state of the device. Compared with the fixed parameter solution, the distortion of the compensated current waveform is reduced by 60%; Maintenance-free self-adaptation: The automatic calibration mechanism does not require human intervention, which is particularly suitable for large-scale deployment of industrial APF clusters (more than 50 units at a single site), reducing the on-site parameter debugging workload by 85% and shortening the equipment delivery cycle by 40%.

[0130] In some embodiments, a layered edge-cloud architecture is designed: At the edge layer, a lightweight MPC controller (computational complexity reduced by 60%) is deployed to process high-frequency harmonic detection data (200kHz sampling rate) in real time and generate PWM drive signals. Key state data (junction temperature sequence, switching frequency, compensation error) is packaged and uploaded to the cloud at 50ms intervals. At the cloud layer, a global optimization model based on deep reinforcement learning (DRL) is constructed, taking as input multi-site APF operating data (over 100,000 hours of cumulative operating condition samples) and outputting regionalized control strategies (such as high-impact load strategies for steel mills and low-ripple strategies for data centers), which are automatically distributed to the edge each week.

[0131] Cross-device knowledge transfer mechanism: Federated learning technology is used to aggregate multi-site data, training a common policy model while protecting user privacy (for example, extracting common harmonic characteristics across loads in different industries). This allows the edge to retain local optimization parameters for specific operating conditions (for example, the specific weights used when a site's PV inverter becomes a high-frequency harmonic source). The cloud-based DRL model sets up a "strategy grayscale release": new policies are first piloted on 10% of devices (monitoring indicators such as THD, losses, and junction temperature). Once these policies meet the target (performance improvement ≥5% with no anomalies), they are fully rolled out.

[0132] Predictive O&M service: The cloud uses an LSTM network to analyze the junction temperature fluctuation curve uploaded by the edge, predicting cooling fan failure 14 days in advance (with 92% accuracy) and automatically generating an O&M work order (including the location of the faulty IGBT, recommended replacement time, and temporary derating strategy).

[0133] Compared to local MPC with fixed parameters, the cloud-based DRL strategy improves average THD suppression by 22% (from 4.1% to 3.2%) in complex cross-industry scenarios. Furthermore, the optimal strategy migration time between different sites is shortened from 72 hours with manual parameter adjustment to 5 minutes with automatic delivery.

[0134] The federated learning mechanism increases the data utilization rate of a single site by 300%, solving the "data island" problem of traditional edge computing. It is particularly suitable for the collaborative optimization of distributed APF clusters in multiple campuses (such as reducing the total amount of regional power grid harmonics by 18%). Predictive operation and maintenance transforms passive maintenance into active maintenance, reducing unplanned downtime by 90% and operation and maintenance costs by 45%. At the same time, the policy gray release mechanism ensures zero risk in technology iteration (abnormal policy fallback time is less than 1 minute).

[0135] In some embodiments, a thermal-electric coupling model is established: a thermal network state equation including the IGBT junction temperature Tj, the heat sink temperature Th, and the fan speed vf is constructed and combined with the electrical control model (harmonic current ih, compensation voltage uc) to form a multi-physics field state space:

[0136]

[0137] Where hA is the heat transfer coefficient of the radiator (positively correlated with the wind speed vf), and Ploss is calculated by the switching frequency and current. Collaborative optimization strategy design: Introduce the heat dissipation energy consumption Jcool=ηvf in the MPC objective function 3 (η is the fan energy efficiency coefficient), achieving the Pareto optimality of "harmonic compensation accuracy" and "heat dissipation energy consumption":

[0138] Low junction temperature zone (Tj < 80°C): The fan is preferentially turned off (vf = 0), reducing losses by lowering the switching frequency, and cooling energy consumption is reduced by 70%. Critical junction temperature zone (80°C ≤ Tj < 100°C): The fan speed (500-3000 RPM) and switching frequency (15kHz-25kHz) are dynamically adjusted to minimize the sum of total losses and cooling energy consumption.

[0139] High junction temperature zone (Tj ≥ 100°C): Start the fan at full speed (vf = 3500 RPM) and enable the low-loss control mode described above.

[0140] The hardware collaborative design uses a variable flow channel heat sink with a shape memory alloy (SMA)-driven guide vane to automatically adjust the angle of the heat sink fins according to the junction temperature (for example, when Tj>90°C, the fins are expanded to increase the heat dissipation area by 20%), forming hardware-software collaborative heat dissipation with the control algorithm.

[0141] Compared with traditional independent cooling solutions, the total system power consumption is reduced by 28% (the proportion of heat dissipation energy consumption is reduced from 40% to 25%) under medium load scenarios (THD = 8%), and the fan life is extended by 50% (the low-speed operation time is increased by 60%); the thermal-electric coupling model improves the junction temperature prediction accuracy to ±1.5°C, and with the hardware guide plate adjustment, even when the fan partially fails (the wind speed drops by 30%), the junction temperature can still be controlled below the safety threshold of 125°C (the traditional solution has an over-temperature probability of 35%); by incorporating the heat dissipation physical variables into the control model, the true synergy of "electrical performance-thermal management-energy consumption" is achieved, providing a new path for high-power density APF design (the volume can be reduced by 15% while maintaining the same heat dissipation capability).

[0142] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An active filter harmonic suppression and dynamic compensation system based on IGBT multi-mode control, characterized in that: include: A control module configured to construct a multi-modal collaborative control architecture, wherein the multi-modal collaborative control architecture includes a model predictive control algorithm unit, and the model predictive control algorithm unit generates a PWM compensation waveform in real time; a high-frequency harmonic detection module, electrically connected to the control module, configured to detect power grid harmonic signals in real time at a sampling rate of 200 kHz and transmit the signals to the control module; An IGBT gate drive compensation circuit is electrically connected to the control module and a preset IGBT module, and the control module dynamically adjusts the dead time and switching frequency of the IGBT module through the IGBT gate drive compensation circuit; An IGBT junction temperature observer, electrically connected to the IGBT module and the control module, configured to monitor the junction temperature of the IGBT module in real time, and trigger the control module to start a derating protection mechanism when the junction temperature exceeds 125°C; Among them, the control module generates the PWM compensation waveform through the model predictive control algorithm unit based on the detection signal of the high-frequency harmonic detection module, so as to drive the IGBT module through the IGBT gate drive compensation circuit to perform harmonic compensation, thereby achieving dynamic compensation of reducing the total harmonic distortion rate from 15% to below 3% and bidirectional adjustable reactive power of ±1Mvar; the IGBT junction temperature observer integrates a fault location unit for locating faults in the IGBT module.

2. The system according to claim 1, wherein: Generating the PWM compensation waveform by the model predictive control algorithm unit includes: A prediction model is constructed that includes IGBT switching characteristics, dead time impact, and grid harmonic characteristics. Based on the 200kHz sampling signal input in real time by the high-frequency harmonic detection module, a rolling optimization calculation is performed on the harmonic current compensation amount in the next control cycle to generate a PWM drive signal sequence that includes dead time compensation and switching frequency dynamic adjustment parameters.

3. The system according to claim 2, characterized in that The IGBT gate drive compensation circuit drives the IGBT module to perform harmonic compensation, thereby reducing the total harmonic distortion rate from 15% to below 3% and achieving dynamic compensation of reactive power with a bidirectional adjustment of ±1Mvar, including: According to the dead time adjustment instruction carried in the PWM compensation waveform, the phase difference of the drive signals of the upper and lower bridge arms of the IGBT is corrected in real time through the IGBT gate drive compensation circuit. At the same time, the switching frequency is dynamically adjusted according to the reactive power compensation target, so that the IGBT module outputs a compensation current with equal amplitude and opposite phase to the grid harmonic current, and realizes an adjustable output of ±1Mvar of inductive or capacitive reactive power through a bidirectional power flow control circuit.

4. The system according to claim 2, wherein: During the rolling optimization process, the model predictive control algorithm unit introduces an adaptive weighting factor to perform multi-objective optimization of the THD suppression effect and switching loss. The weighting factor is dynamically adjusted in real time according to the IGBT junction temperature: when the junction temperature is ≥100°C, the switching loss weight is automatically increased to above 0.7, giving priority to reducing IGBT heat; when the junction temperature is less than 100°C, the weight is tilted towards THD suppression, improving compensation accuracy by 20%.

5. The system according to claim 1, wherein: The real-time monitoring of the junction temperature of the IGBT module and triggering the control module to start the derating protection mechanism when the junction temperature exceeds 125° C. include: The IGBT junction temperature observer constructs a thermal network model based on the IGBT conduction voltage drop, switching loss and heat dissipation parameters, and estimates the chip junction temperature in real time. When the junction temperature exceeds 125°C, a derating instruction is sent to the control module. The control module reduces the IGBT loss by more than 30% by reducing the switching frequency, limiting the maximum compensation current and switching to a low-loss control mode. At the same time, the fault location unit records the time-temperature curve of the abnormal junction temperature for subsequent fault analysis.

6. The system according to claim 1, wherein: The real-time detection of the grid harmonic signal at a sampling rate of 200kHz and the transmission to the control module include: The high-frequency harmonic detection module integrates an anti-aliasing filter and a synchronous sampling phase-locked loop, performs 200kHz synchronous sampling on the grid voltage and current signals, extracts the amplitude, phase and frequency parameters of the 2nd to 50th harmonic components through fast Fourier transform (FFT), generates a digital signal sequence containing the characteristics of each harmonic, and transmits it to the control module.

7. The system according to claim 1, wherein: The method of dynamically adjusting the dead time and switching frequency of the IGBT module by using the IGBT gate drive compensation circuit includes: The IGBT gate drive compensation circuit integrates an adjustable delay unit and a frequency control module. The control module dynamically adjusts the dead time and switching frequency through a fuzzy adaptive algorithm based on the current harmonic compensation accuracy and IGBT junction temperature. In high-harmonic content scenarios, the switching frequency is increased to enhance compensation accuracy, and in high-temperature scenarios, the dead time is reduced to reduce switching losses. The adjustment step corresponding to the dead time is ≤100ns; the adjustment range corresponding to the switching frequency is 10kHz-50kHz.

8. The system according to claim 1, wherein: The fault locating of the IGBT module includes: The fault location unit constructs a fault feature database based on the IGBT gate drive voltage waveform, collector current abnormal fluctuation signal and junction temperature mutation data, and uses the support vector machine algorithm to perform pattern recognition on fault types such as overcurrent, short circuit, and drive circuit failure. Combined with the dead time and switching frequency adjustment parameters when the fault occurs, the fault is located to a specific IGBT chip or drive circuit component.

9. The system according to claim 1, wherein: The multimodal collaborative control architecture integrates an IGBT aging prediction module based on a long short-term memory network. The IGBT aging prediction module predicts the remaining life of the IGBT module and dynamically adjusts the control strategy by learning historical junction temperature fluctuations, switching times and fault data: when the predicted remaining life is less than 20%, it automatically switches to a conservative control mode, reduces the switching frequency by 25% and expands the dead time safety margin, thereby extending the service life of the device.

10. The system according to claim 1, wherein: When generating the PWM compensation waveform, the model predictive control algorithm unit embeds a harmonic current prediction model based on Kalman filtering. The model combines the grid load change trend and historical harmonic data to predict and compensate for the harmonic components within the next 200μs, shortening the dynamic response time to 50μs.

Images