Design method and device of digital active EMI filter based on online delay compensation

Abstract

The application provides a digital active EMI filter design method and device based on online time delay compensation, and belongs to the technical field of electromagnetic interference of power electronic devices. The online compensation of time delay is realized without needing to obtain the impedance and transfer function information of an inverter and an AEF in advance. Firstly, when the inverter is normally working, electromagnetic interference is collected by using a sensor, and main frequency points of the interference are extracted by using an FFT algorithm; subsequently, a notch filter corresponding to the interference frequency points is generated based on an LMS algorithm; finally, the control parameters of each notch channel are adjusted online, so that the compensation current and the interference current are aligned in the time domain, the residual interference after compensation is minimized, the outputs of the channels are combined, the multi-frequency point interference can be cancelled, and the online compensation of the algorithm to the time delay is completed. The application can adjust the control parameters of the LMS algorithm online and search for the optimal update direction, so that the efficient suppression of electromagnetic interference is realized under the condition that the time delay is unknown.

Description

Technical Field

[0001] This invention belongs to the field of electromagnetic interference technology for power electronic devices, and more specifically, relates to a design method and device for a digital active EMI filter based on online time delay compensation. Background Technology

[0002] Electromagnetic interference (EMI) is an unavoidable problem in inverter operation because pulse width modulation (PWM) means that power devices operate in a high-speed switching state, resulting in large voltage (dv / dt) and current (di / dt) jumps, forming EMI sources rich in high-frequency components. Currently, common-mode and differential-mode inductors, along with X and Y capacitors, are commonly used to construct passive EMI filters to suppress EMI and meet electromagnetic compatibility (EMC) standards. However, differential-mode EMI typically has high amplitudes at the switching frequency and its first few harmonics. This requires the EMI filter to provide sufficient attenuation at low frequencies, necessitating larger capacitors and inductors to lower the filter's cutoff frequency. This significantly increases the filter's size and hinders power density improvements. Active EMI filters (AEF) are based on the detection-injection principle, using operational amplifiers, transistors, and other active devices to generate signals with the same amplitude but opposite directions to cancel out electromagnetic interference. Therefore, compared to passive EMI filters, AEFs are smaller in size and weight, effectively improving the power density of devices.

[0003] Based on different signal processing methods, AEFs can be divided into two types: analog and digital. Analog AEFs use analog devices such as operational amplifiers and push-pull amplifiers to amplify and output interference signals, while digital AEFs typically use digital chips such as MCUs and FPGAs to control the SiC / GaN inverter to generate inverse cancellation signals corresponding to the interference, achieving EMI suppression. The suppression performance of analog AEFs depends on the internal component parameters of the AEF, the impedance of the inverter's interference source, and the transfer function, making the process complex and parameter design difficult. In contrast, digital AEFs rely on digital chips and mainly have three design methods: 1) Without relying on control algorithms, only using an ADC to collect interference, converting it into a digital signal, inverting it, and then using a DAC to restore it to an analog signal and injecting it into the power loop to compensate for the interference; 2) Using offline algorithms, measuring the impedance and transfer function of the interference source in advance and storing them in the digital chip, then using an FFT algorithm to calculate the interference compensation signal that the AEF should output; 3) Using adaptive filtering algorithms, represented by the least mean square (LMS) algorithm, which does not require obtaining transfer function and impedance information, only sampling the interference signal and combining it with the algorithm to achieve electromagnetic interference cancellation. In summary, digital AEF offers greater controllability and stronger adaptability, making it an important development direction for next-generation active EMI suppression technology.

[0004] However, time delays are inevitably introduced in each stage of a digital AEF, causing the compensation current and interference current to misalign in the time domain. This severely impacts the stability and interference suppression performance of the AEF. Existing time delay compensation methods rely on precise time delay measurements, complex algorithm designs, and additional hardware circuits, resulting in high design costs and limited compensation effects, even weakening the performance of the LMS algorithm itself. Therefore, there is an urgent need to propose a simple digital AEF design method to compensate for time delays online and improve the AEF's suppression performance. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a design method and device for a digital active EMI filter based on online time delay compensation, thereby solving the technical problem that digital active electromagnetic interference filters (AEF) have a time delay problem when suppressing electromagnetic interference from power electronic devices, which causes the compensation current and the interference current to be misaligned in the time domain, seriously affecting the stability and interference suppression effect of AEF.

[0006] To achieve the above objectives, according to one aspect of the present invention, a design method for a digital active EMI filter based on online time delay compensation is provided, comprising: The interference current signal of the power electronic converter is collected, and frequency domain analysis is performed on the interference current signal to extract multiple interference main frequency points; A multi-channel notch filter is constructed based on multiple interference master frequencies using the least mean square algorithm, so that each channel corresponds to one interference master frequency. For each interference master frequency, the interference frequency of each interference master frequency is obtained, and the interference frequency generates a waveform signal of the same frequency as the reference signal of the notch filter. The phase of the reference signal for each channel is adjusted online to align the reference signal with the error signal in the time domain, thereby changing the iteration direction of the notch filter weights so that the actual iteration direction of the weights converges to the ideal iteration direction. The adjusted reference signals of each channel are combined as compensation signals and output to the interference current loop, thereby achieving multi-frequency interference suppression.

[0007] Preferably, constructing a multi-channel notch filter using the least mean square algorithm includes: Based on the least mean square algorithm, a secondary path modeling method is introduced to obtain the weight update formula for the notch filter.

[0008] in, P x ( ω ) is the frequency ω The power spectral amplitude of the reference signal at that time. μ The step size parameter is the step size parameter of the least mean square algorithm. A ω In order to be in ω Secondary path identification amplitude error at a given frequency. P ω This is the primary path. S ω For secondary paths, φ a For the reference signal phase, j The number of interference main frequency points.

[0009] Preferably, generating a waveform signal of the same frequency as the interference frequency includes: After obtaining the interference frequency, a reference sine wave signal of the same frequency is generated by a sine wave generator, and then the phase is shifted by 90° to obtain the cosine wave signal corresponding to the sine wave signal.

[0010] Preferably, the expression of the waveform signal is as follows:

[0011] in, x ksin '( n )and x kcos '( n ) respectively represent the first k The phase-shifted reference signal for each channel Ts represent Sampling time, ω k Representing the k The frequency of each channel, φ ak Representing the k The built-in phase shift of each channel; only the phase shift of each channel needs to be adjusted. x sin ( n Phase shifting can be used to update the iteration direction of the notch filter weights.

[0012] Preferably, the reference signal phase of each channel is adjusted online. φ a Changing the iteration direction of the notch filter weights involves the following steps: During the initialization phase, the notch filter weight parameters are not updated, and the iteration continues. N Time of each sampling period T N ,calculate T N Interference signals within a time period e ( n The sum of all the terms and the average of the terms are denoted as . S min ; During the iterative direction search phase, a phase is selected. φ a The same iteration T N After a period of time, calculate T N Interference signals within e ( n The sum of all the terms and the average of the terms are denoted as . S o ;like S o < S min ,but S min = S o ,on the contrary S min The process remains unchanged until the search in all directions is completed, at which point the final result is recorded. S min Corresponding φ a .

[0013] During the normal operation of the filter, the selected recording... φ a And continue iterating until a steady state is reached.

[0014] Preferably, if the interference signal continues to increase in a certain update direction, it is determined that the direction does not converge, and the update is terminated immediately.

[0015] Preferably, the frequency domain analysis uses a fast Fourier transform algorithm to extract the main interference frequency point.

[0016] Preferably, the output signals of the multi-channel notch filter are combined by direct superposition.

[0017] Another aspect of the present invention proposes a digital active EMI filter device obtained by a digital active EMI filter design method based on online time delay compensation.

[0018] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. This invention proposes a digital active EMI filter design method based on online time delay compensation. During normal inverter operation, electromagnetic interference is collected using sensors, and the dominant interference frequency is extracted using an FFT algorithm. A notch filter corresponding to the interference frequency is generated based on the LMS algorithm. Finally, the control parameters of each notch filter channel are adjusted online to align the compensation current with the interference current in the time domain, minimizing the residual interference after compensation. The outputs of each channel are combined to cancel out multi-frequency interference, completing the online time delay compensation algorithm. This invention can adaptively adjust the LMS algorithm control parameters online and search for its optimal update direction, thereby achieving efficient suppression of electromagnetic interference even when the time delay is unknown.

[0019] 2. The digital active EMI filter design method based on online time delay compensation proposed in this invention uses the least mean square algorithm to construct a multi-channel notch filter, and introduces secondary path modeling on the basis of the least mean square algorithm to obtain the weight update formula of the notch filter. It fully considers the time delay problem caused by the digital operation and signal processing links inside the AEF, and utilizes the principle of secondary path compensation of the LMS algorithm to embed time delay parameters in the algorithm to compensate for the influence of external time delay.

[0020] 3. The digital active EMI filter design method based on online time delay compensation proposed in this invention does not require known impedance or transfer function information of the inverter or AEF, and can realize time delay compensation online with strong portability. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of differential mode electromagnetic interference compensation for an inverter with digital AEF added according to an embodiment of the present invention; Figure 2 This is a block diagram of the filter structure based on the LMS algorithm in an embodiment of the present invention; Figure 3 The input signal frequency in this embodiment of the invention is ω At that time, the filter structure block diagram based on the Fx-LMS algorithm; Figure 4 This is a schematic diagram of AEF parameter update without time delay in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the AEF parameter update when there is a time delay in an embodiment of the present invention; Figure 6 This is a schematic diagram of the feasible region for the filter update direction in an embodiment of the present invention; Figure 7 This is a schematic diagram of the AEF parameter update direction for online delay matching in an embodiment of the present invention; Figure 8 This is a filter block diagram of the notch-type LMS algorithm that takes time delay into account in an embodiment of the present invention; Figure 9 This is a time-domain performance verification of the proposed online delay matching AEF in a single-frequency interference scenario according to an embodiment of the present invention; Figure 10 This is a filter block diagram based on the multi-channel notch LMS algorithm in an embodiment of the present invention; Figure 11 This is a performance verification of the proposed online delay matching AEF in the frequency domain under multi-frequency interference scenarios in this embodiment of the invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Example 1 Introduction to the Differential Mode Path and Internal Components of the Digital AEF in Inverter of This Invention Inverters are typical power electronic devices and also a major source of electromagnetic interference. Figure 1 The diagram illustrates the circuit structure after incorporating a digital AEF (Automatic Electronic Array) and the flow of differential-mode electromagnetic interference (EMI). Differential-mode interference is the sudden change in current on the positive and negative buses when power devices switch, forming a current loop between phases through the DC support capacitor and input cables. The LISN (Line Impedance Stabilization Network) isolates interference introduced by the DC power supply and simulates the grid impedance to provide a reference for measuring the differential-mode interference generated by the target inverter.

[0024] To compensate for differential-mode interference current, the digital AEF (Automatic External Filter) needs to sample the interference current in the interference loop, generate a corresponding compensation current through signal processing circuitry, and finally inject it back into the interference loop to achieve interference current compensation. Specifically, the digital AEF uses a Rogowski coil for interference sampling to obtain the frequency and amplitude of differential-mode interference in the power loop. Subsequently, the collected signal is processed by a conditioning circuit, converted into a digital signal by an ADC, and input to the signal processing circuit as an error signal for the algorithm. At the same time, the algorithm generates a reference signal for the corresponding frequency. Finally, after processing by the LMS (Low Mode Simulation) algorithm, the signal output from the signal processing circuit drives the inverter circuit to generate a signal with opposite phase and equal amplitude to the interference, which is then output to the power loop by the signal injection circuit to achieve interference compensation.

[0025] The signal processing circuit includes a DSP (Digital Signal Processor), FPGA, PWM waveform control circuit, and CAN communication circuit. Its main functions are differential-mode signal processing, PWM signal output, cancellation signal generation, closed-loop control, and external communication and protection. The core of the signal processing circuit is the FPGA, which is the hardware foundation for implementing the adaptive filtering algorithm. Its performance directly determines the interference suppression performance of the AEF (Adaptive Filtering Efficiency). Furthermore, considering that the amplitude of the differential-mode current is usually large in high-power scenarios, a bridge inverter circuit can be used to implement the power amplifier in the DAC. Using an inverter circuit as the power amplifier has the advantages of high controllability and strong output capability, but it requires a high switching frequency (usually more than ten times the highest frequency of the target interference). Therefore, a GaN HEMT can be used as the power switch.

[0026] Example 2 LMS Algorithm and its Delay Issues The digital AEF proposed in this invention is implemented based on the LMS algorithm. The LMS algorithm is a widely used adaptive filtering algorithm that continuously optimizes the filter weight coefficients by minimizing the mean square error (MSE) between the desired and actual outputs. The LMS algorithm is an iterative method based on gradient descent, where the filter weight coefficients can be dynamically adjusted and only require simple addition and multiplication operations, resulting in high computational efficiency. The filter structure based on the LMS algorithm is shown below. Figure 2 As shown; where, x ( n () is the reference signal. y ( n () represents the filter output signal. e ( n The signal is the error signal after cancellation and synthesis. d ( n ) represents the desired signal (i.e., the target output). μThis is the step size parameter of the algorithm. The LMS algorithm is based on the mathematical principle of gradient descent. Considering the application of the algorithm in practical physical systems, instantaneous values ​​can be used instead of mathematical expectations, significantly reducing the algorithm's complexity. The obtained instantaneous gradient is an unbiased estimate of the true gradient. Therefore, when the reference signal... x ( n ) and error signal e ( n During time-domain alignment, the algorithm will update the filter parameters along the steepest descent direction. W ( n As shown in equation (1), the error is ultimately reduced. e ( n (Minimum)

[0027] (1) However, due to the complex signal propagation path design in the physical system, error signals... e ( n There is a time delay, making it impossible to synchronize with the reference signal. x ( n Aligning the gradient in the time domain means that the algorithm's instantaneous gradient will no longer be an unbiased estimate of the true gradient. This causes the LMS algorithm to stop updating the filter parameters along the gradient direction, thus failing to correct the error. e ( n To minimize this, the Filtered-x LMS (Fx-LMS) algorithm can be used. This algorithm introduces secondary path modeling on top of the LMS algorithm to correct the instantaneous gradient, ensuring the effectiveness and stability of the LMS algorithm. After adopting the Fx-LMS algorithm, equation (1) can be transformed into equation (2), where... x’ ( n ) represents the reference signal that has been corrected after considering secondary paths.

[0028] (2) The Fx-LMS algorithm block diagram is as follows: Figure 3 As shown; in Figure 3 In the middle, when the input signal frequency is ω In this case, each step can be represented by a complex number: its primary path P ( z ) represents P ω Filter weight matrix W ( z ) represents W ω ( n Secondary paths S ( z ) represents S ω The estimation of the secondary path is expressed asS’ ω For a single fixed frequency point ω The Fx-LMS algorithm can be expressed as: (3) In the formula To represent conjugate complex numbers, P x ( ω ) indicates that the frequency is ω The power spectral amplitude of the reference signal at that time. If the secondary path S’ ω The estimate has no error, that is S’ ω = S ω Then, equation (3) can be expressed as: (4) When the Fx-LMS algorithm runs for a sufficiently long time, i.e. n As we approach infinity, the weights can be considered to no longer be updated, and we have: (5) Substituting equation (5) into equation (4), we get: (6) This indicates that, under ideal conditions, when the secondary path estimation has no error, the frequency... ω Lower filter weight coefficients W ω It will eventually converge to P ω / S ω Error signal e ( n The differential-mode EMI is effectively suppressed when the Fx-LMS algorithm reaches its minimum. Since all parameters of the Fx-LMS algorithm are complex numbers, its iterative process can be graphically represented in the complex plane, as shown below. Figure 4 and Figure 5 As shown. Among them, W ω ( n () represents the weight coefficient at the current time. W ω ( n+1 ) represents the weight coefficient at the next moment. From equation (4), it can be seen that the weight parameter changes from... W ω ( n )towards W ω ( n+1 Distance traveled L for: (7) Figure 4 This demonstrates the case where the secondary path estimation has no error, therefore W ω ( n It will eventually converge to P ω / S ω To ensure that the weight coefficients eventually converge, from W ω ( n )arrive W ω ( n+1 The weight update distance should be less than twice the distance from the final convergence point. W ω ( n The convergence condition is obtained by considering the distance between the two points: (8) The step size coefficient can be obtained. μ The range of values ​​for: (9) However, in real-world systems, secondary paths are often quite complex and of high order, inevitably leading to errors in their identification. S’ ω ≠ S ω The error is mainly manifested as time delay in the time domain. t d In the complex plane, this is represented by the phase angle difference between the actual weight update path and the ideal path. φ ω , satisfying equation (10), where, T The frequency is ω The corresponding period, φ ω It is limited to a range of 0° to 360°.

[0029] (10) Furthermore, the error is also reflected in the magnitude, specifically in the weight update amplitude. The true secondary path is... S ω Secondary paths with identification errors S’ ( ω This can be represented as: (11) In the formula, A ω Indicates in ω The secondary path identification amplitude error at a given frequency. When there is an identification error in the secondary path, the weight update formula (4) can be expressed as: (12) At this point, the filter parameters are updated as follows: Figure 5 As shown, W ω ( n No longer along Figure 4 The path converges to P ω / S ω , but exists φ ω The phase error. (By) W ω ( n )arrive W ω ( n+1 The weight update distance is: (13) Depend on Figure 5 It can be seen that when -90° < φ ω <90° and satisfies: (14) Right now: (15) W ω ( n+1 )arrive P ω / S ω The distance is less than W ω ( n )arrive P ω / S ω The algorithm converges when the distance is equal to the given distance. Conversely, when | φ ω When |≥90°, the adaptive algorithm fails to converge, and AEF becomes unstable, such as Figure 6 As shown. Therefore, the convergence condition of the algorithm is obtained: (16) Without considering the convergence rate, only the step size coefficient... μ If it is small enough, it will always satisfy the constraint condition of equation (16); however, the time delay present in AEF is very likely to cause | φ ω |≥90°, causing the algorithm to fail to converge. Therefore, time delay is a key factor constraining the performance and stability of the LMS algorithm, and time delay must be compensated to meet | φ ω|<90°, thereby improving the interference suppression performance and stability of digital AEF.

[0030] Example 3 Design of Notch-Type LMS Algorithm Based on Online Delay Matching Time delay causes the iteration direction of the LMS algorithm to differ from the ideal iteration direction. φ ω This leads to a weakening or even instability of the algorithm's performance. To address the problem caused by time delay, this invention proposes an online delay matching LMS algorithm, whose weight update formula is shown in equation (17): (17) in, φ a This is the phase shift preset in the secondary path estimate of the Fx-LMS algorithm. From Figure 7 It can be seen that when φ a Taking 0°, 90°, 180°, and 270° respectively corresponds to update directions 1, 2, 3, and 4; when the weight update direction falls within the feasible region, i.e., when... φ a The algorithm converges when the angle is 0° and 90°. Simultaneously, combined with... Figure 4 and Figure 5 Analysis shows that when the step size μ Enough hours φ ω and φ a The smaller the difference, the better. W ω ( n The closer it is to the ideal value P ω / S ω In other words, the stronger the interference suppression capability, the better. The algorithm proposed in this invention can modify the filter parameters online during the filter parameter update process. φ a It automatically finds the direction that satisfies the convergence condition and has the best suppression performance.

[0031] To ensure the proper functioning of the above algorithm, it is also necessary to accurately collect interference frequencies. ω The inverter generates a compensation signal at a corresponding frequency. Noise generated by the inverter exhibits periodicity, with its frequency domain characteristics resembling isolated line-like spectral lines, meaning that several frequency points on the spectrum have large interference amplitudes. For noise sources with these characteristics, it is not necessary to actually measure the reference signal; instead, only the frequency of the interference is measured, and a waveform with the same frequency is artificially generated as the reference signal for the LMS algorithm. x ( nThis method is called the notch filter LMS algorithm, such as... Figure 8 As shown. After measuring the interference frequency, a reference sine wave of the same frequency is generated by a sine wave generator. x sin ( n ), and after phase shifting it by 90°, the corresponding cosine wave is obtained. x cos ( n Therefore, it is only necessary to... x sin ( n By performing a phase shift, the update direction in the above algorithm can be adjusted, that is: (18) in, x ksin ’ ( n )and x kcos ’ ( n ) respectively represent the first k The phase-shifted reference signal for each channel T s Represents the sampling time. ω k Representing the k The frequency of each channel, φ ak Representing the k The built-in phase shift of each channel. At this time, x’ sin ( n )and x sin ( n The ratio of ) is the secondary path estimate corresponding to this algorithm. S’ ω Therefore, only the reference signal of the notch-type LMS filter needs to be considered. x sin ( n By performing phase shifting, the estimation of secondary paths can be automatically corrected, thereby achieving online latency matching. Figure 9 The effectiveness of the proposed algorithm in suppressing single-frequency interference is demonstrated, and its iterative process can be divided into the following three stages: a) Initialization iteration phase, including sub-phase ①: No LMS weight parameters are updated, iteration time is N sampling periods. T N Calculate the interference during this period e ( n The sum of all the terms and the average of the terms are denoted as . S min ; b) Iterative direction search phase, including sub-phases ②-⑤: In each sub-phase, an update direction is selected (determining a...). φ a (value), iterate in the same way. T N After a certain time, calculate the interference during that period. e ( n The sum of all the terms and the average of the terms are denoted as . S o ;like S o < S min ,but S min = S o ,on the contrary S min The process remains unchanged until the search in all directions is completed, at which point the final result is recorded. S min Corresponding φ a During the update process... e ( n+1 )>2 e ( n If the algorithm fails to converge, the sub-stage can be considered non-convergent, and the sub-stage should be terminated directly. Figure 9 As shown in ④; c) During the normal operation phase of the filter, including sub-phase ⑥: the optimal update direction determined in the previous phases (i.e., the actual phase shift caused by the AEF delay). φ ω Closest built-in phase shift φ a Continue iterating until a steady state is reached.

[0032] from Figure 9 As can be seen, after the filter reaches steady state, the peak-to-peak value of the interference is only 1.2A, which is 99.85% lower than when the filter parameters were not updated, demonstrating significant suppression performance.

[0033] Meanwhile, for differential-mode EMI, the interference noise is often not a single frequency point, but rather manifests in the frequency domain as several large-amplitude interference frequencies superimposed on the noise floor. Therefore, active EMI filters should not be designed for a single frequency point, but rather for several large-amplitude interference frequencies within a certain frequency band. Parallel connection of multiple notch filter LMS algorithms can achieve interference noise suppression at multiple frequencies, extending the suppression bandwidth of the active EMI filter, such as... Figure 10As shown. Because the FPGA employs a hardware parallel architecture, it allows multiple computing units to execute tasks simultaneously, providing the hardware foundation for the implementation of the multi-channel LMS algorithm. The active EMI filter and various components within the inverter have non-flat frequency responses; therefore, the phase shift corresponding to the time delay at each frequency point... φ ωj ( j =1,2,…, k , k The number of frequency points varies. Figure 9 The direction search process shown needs to be performed separately for each channel to ultimately obtain... φ aj ( j =1,2,…, k , k (Number of frequency points). Finally, k The outputs of the individual frequency filters are superimposed to obtain the total output: (19) Therefore, parallel connection of multiple channels can superimpose and merge multiple single-channel algorithms to achieve parallel suppression at multiple frequencies. Figure 11 The spectral results before and after interference suppression, measured by an EMI receiver, are presented. The proposed digital active EMI filter design method can achieve interference suppression of up to 31.7 dB in the 5 kHz–20 kHz range. This algorithm utilizes a reference signal. x ( n The feature of built-in phase shift to change the iteration direction of filter parameters allows for online matching of the time delay caused by the internal components of AEF without the need for known transfer function and impedance information. This ensures the convergence of the LMS algorithm and the interference suppression performance of the digital AEF, and has strong portability.

[0034] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A design method for a digital active EMI filter based on online time delay compensation, characterized in that, include: The interference current signal of the power electronic converter is collected, and frequency domain analysis is performed on the interference current signal to extract multiple interference main frequency points; A multi-channel notch filter is constructed based on multiple interference main frequency points using the least mean square algorithm, so that each channel corresponds to one interference main frequency point; For each main interference frequency, the interference frequency of each main interference frequency is obtained, and the interference frequency is used to generate a waveform signal of the same frequency as the reference signal of the notch filter. The phase of the reference signal for each channel is adjusted online to align the reference signal with the error signal in the time domain, thereby changing the iteration direction of the notch filter weights so that the actual iteration direction of the weights converges to the ideal iteration direction. The adjusted reference signals of each channel are combined as compensation signals and output to the interference current loop, thereby achieving multi-frequency interference suppression; The construction of a multi-channel notch filter using the least mean square algorithm includes: Based on the least mean square algorithm, a secondary path modeling method is introduced to obtain the weight update formula for the notch filter. in, P x ( ω ) is the frequency ω The power spectral amplitude of the reference signal at that time. μ The step size parameter is the step size parameter of the least mean square algorithm. A ω In order to be in ω Secondary path identification amplitude error at a given frequency. P ω This is the primary path. S ω For secondary paths, φ a For the reference signal phase, j The number of interference main frequency points; Generating a waveform signal of the same frequency as the interference frequency includes: After obtaining the interference frequency, a reference sine wave signal of the same frequency is generated by a sine wave generator, and then the phase is shifted by 90° to obtain a cosine wave signal corresponding to the sine wave signal. The expression for the waveform signal is: in, x ksin '( n )and x kcos '( n ) respectively represent the first k The phase-shifted reference signal for each channel T s represent Sampling time, ω k Representing the k The frequency of each channel, φ ak Representing the k The built-in phase shift of each channel; only the phase shift of each channel needs to be adjusted. x sin ( n Phase shifting can be used to update the iteration direction of the notch filter weights.

2. The design method for a digital active EMI filter based on online time delay compensation according to claim 1, characterized in that, Adjust the phase of the reference signal for each channel online. φ a Changing the iteration direction of the notch filter weights involves the following steps: During the initialization phase, the notch filter weight parameters are not updated, and the iteration continues. N Time of each sampling period T N ,calculate T N Interference signals within a time period e ( n The sum of all the terms and the average of the terms are denoted as . S min ; During the iterative direction search phase, a phase is selected. φ a The same iteration T N After a period of time, calculate T N Interference signals within e ( n The sum of all the terms and the average of the terms are denoted as . S o ;like S o < S min ,but S min = S o ,on the contrary S min The process remains unchanged until the search in all directions is completed, at which point the final result is recorded. S min Corresponding φ a ; During the normal operation of the filter, the selected recording... φ a And continue iterating until a steady state is reached.

3. The design method for a digital active EMI filter based on online time delay compensation according to claim 2, characterized in that, If the interference signal continues to increase in a certain update direction, it is determined that the direction does not converge, and the update is terminated immediately.

4. The design method for a digital active EMI filter based on online time delay compensation according to claim 1, characterized in that, The frequency domain analysis uses the Fast Fourier Transform algorithm to extract the main interference frequency points.

5. The design method for a digital active EMI filter based on online time delay compensation according to claim 1, characterized in that, The output signals of the multi-channel notch filter are combined by direct superposition.

6. A digital active EMI filter device obtained by the design method of a digital active EMI filter based on online time delay compensation as described in any one of claims 1-5.

Images