Convergence judgment method of adaptive narrowband active filter based on correlation multiplication

Abstract

The invention discloses a convergence judgment method of an adaptive narrowband active filter based on correlation multiplication, relates to an adaptive narrowband active filter, and particularly relates to a DC power grid adaptive narrowband active filter that judges the convergence state based on correlation multiplication. A convergence state judgment method is designed that uses correlation multiplication of a filter output signal and a residual error signal and accumulates and averages. In the case where the amount of calculation does not increase significantly, it is possible to judge the overall convergence of the algorithm and the convergence of any single frequency point without being affected by interference. Compared with the method of calculating the root mean square of the error signal, multiple links of the entire closed-loop system can be monitored.

Description

Technical Field

[0001] The invention relates to an adaptive narrowband active filter, in particular to an adaptive narrowband active filter for a direct current power grid for judging a convergence state based on correlation multiplication. Background Art

[0002] With the rapid development of power electronics technology, all kinds of AC and DC equipment, whether onshore or onboard, are highly dependent on power electronics. However, the nonlinear characteristics of power electronics bring a large number of harmonics to the grid, and the electromagnetic interference (EMI) generated will interfere with weak current systems such as communication systems, aggravating electromagnetic environmental pollution; harmonics will also cause abnormal operation of electrical equipment, endangering the safe and reliable operation of equipment.

[0003] The traditional harmonic suppression method is to reduce the harmonic current by connecting a series-parallel passive power filter (PPF) near the device that generates harmonics. It has the advantages of low cost, easy maintenance, and mature technology, but it can only filter specific harmonics and has gradually failed to meet current needs. In order to make up for this shortcoming, active power filters (APF) have been rapidly developed in recent years.

[0004] The basic idea of ​​active power filter is to compensate harmonics by detecting the voltage and current of the compensation object and generating an output current with equal amplitude and opposite phase to the harmonic current.

[0005] Unlike land-based power grids, ship power grids have limited capacity margins and poor self-regulation capabilities. As ship operating conditions continue to change, frequent changes in high-power loads can lead to significant changes in power grid parameters. Traditional active filters are difficult to cope with such situations where the system transfer function changes dramatically. Adaptive active filters that update parameters in real time have become a research hotspot in recent years. Adaptive active filters autonomously adjust filter parameters through adaptive algorithms to meet the statistical characteristics of signals that change over time.

[0006] According to the principle, the adaptive active filter can be divided into broadband filtering algorithm and narrowband filtering algorithm. The broadband algorithm occupies more hardware resources and increases the cost. The narrowband algorithm can minimize the demand for hardware resources. In fact, the harmonics of the power grid are always periodic and limited in bandwidth. As long as the frequency points with the largest amplitude are suppressed, a better overall suppression effect can be achieved. Therefore, the narrowband method has more engineering practical value. And judging the convergence state of the algorithm is crucial to monitoring the operating state of the device and improving the stability of the overall system. The simplest way to judge whether the algorithm has converged to a steady state or is still in a search state is to monitor the residual error signal and calculate the root mean square of the residual error to determine the residual energy of the load harmonic current. However, the residual error signal is mixed with many frequencies, and it is impossible to judge whether a single frequency point has converged, and it is easily affected by interference. Summary of the invention

[0007] In view of the above defects or improvement needs of the prior art, the present invention proposes a method for judging the convergence state of an adaptive narrowband filter based on correlation multiplication, and designs a method for performing correlation multiplication on an error signal and a single-frequency point filtering output signal, and accumulating and averaging the results to judge whether the current single frequency point and the algorithm as a whole have converged.

[0008] To achieve the above object, the present invention provides a method for determining the convergence state of an adaptive narrowband filter based on correlation multiplication, comprising:

[0009] S1: Collect the harmonic current signal on the load side;

[0010] S2: Low-pass filter the harmonic current signal on the load side to eliminate high-frequency clutter interference;

[0011] S3: amplify the filtered harmonic current signal;

[0012] S4: performing correlation multiplication operations on the amplified load-side residual harmonic current signal and the generated single-frequency output cancellation signal sine waves respectively;

[0013] S5: The result of the correlation multiplication operation is saved in a shift register, and an average operation is performed on each group of several numbers;

[0014] S6: Judge the average operation result. According to the orthogonality of trigonometric functions, it is concluded that signals of different frequencies are orthogonal to each other, and their average operation results are close to zero, while the product integral of signals of the same frequency is not equal to 0. Then, the cancellation of single-frequency point signals is extracted from the composite waveform, and the convergence state of each frequency point is independently judged.

[0015] In some optional embodiments, step S4 comprises:

[0016] S4.1: Perform AD conversion on the analog signal sent from the previous stage to obtain the digital signal error_signal corresponding to the residual harmonic current on the load side;

[0017] S4.2: multiply the output filtered signal of a single frequency point filtered_signal by the digital signal error_signal corresponding to the residual harmonic current on the load side to obtain the real-time calculation result e(n) corresponding to the current frequency point, where n represents the sampling time, n=1,2,3,...,L1, L1 represents the length of the shift register reg1, and the larger L1 is, the higher the reliability of the calculation result;

[0018] S4.3: Use the shift register reg1 to save the e(n) signal and arrange it in time sequence, where reg1 = [e(n), e(n-1), ..., e(n-L1+1)];

[0019] S4.4: Repeat S4.2 and S4.3 until all frequency points are calculated.

[0020] In some optional embodiments, step S5 comprises:

[0021] S5.1: By Perform summation and averaging operations on e(n), where t represents the sampling time, t=1, 2, 3, ..., L2, L2 represents the length of the shift register reg2, and the smaller L2 is, the more sensitive the judgment of the convergence state is;

[0022] S5.2: The calculation result w(t) is saved using the shift register reg2 to form a moving time window queue data, where reg2 = [w(t), w(t-1), ..., w(t-L2+I)].

[0023] In some optional embodiments, step S6 includes:

[0024] S6.1: After all the data stored in the shift register reg2 are refreshed once, the convergence state is judged;

[0025] S6.2: Select the queue data saved in the shift register reg2 for observation. When the head data w(t) and the tail data w(t-L2+1) are both greater than 0 and less than m, the current frequency point is in a convergence state; when the head data w(t) and the tail data w(t-L2+1) are both close to 0, the harmonic cancellation channel corresponding to the current frequency point is not started or the optimal value cannot be found; when the head data w(t) is greater than m, the algorithm has diverged and a signal needs to be sent to restart or shut down the active filter, where m is the judgment threshold.

[0026] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0027] The present invention proposes a convergence state judgment method for an adaptive narrowband active filter based on correlation multiplication, and designs a convergence state judgment method that uses the filter output signal and the residual error signal to perform correlation multiplication and accumulate and average. Without significantly increasing the amount of calculation, it is possible to judge the overall convergence of the algorithm and the convergence of any single frequency point without being affected by interference. Compared with the method of calculating the root mean square of the error signal, multiple links of the entire closed-loop system can be monitored. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of an overall algorithm convergence judgment provided by an embodiment of the present invention;

[0029] Figure 2 This is a detailed diagram of a method for realizing the convergence state judgment of a single frequency point algorithm provided by an embodiment of the present invention;

[0030] Figure 3 It is a structural diagram of a power grid and an active filter used in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0032] The technical points of the present invention include main technologies such as single frequency point correlation multiplication and the like.

[0033] The purpose of the present invention is achieved through the following technical solutions:

[0034] S1: Use current sensors to collect harmonic current signals on the load side. The sensors should work stably to avoid changes in the induced current amplitude, which may lead to misjudgment of the algorithm.

[0035] S2: Use a low-pass filter circuit to filter the harmonic current signal on the load side to eliminate high-frequency clutter interference;

[0036] S3: amplify the filtered harmonic current signal through a preamplifier circuit. If the current sensor itself has a high conversion sensitivity, this step can be skipped.

[0037] S4: performing correlation multiplication operations on the amplified load-side residual harmonic current signal error_signal and the single-frequency output cancellation signal sine wave filtered_signal generated by the algorithm; the cancellation signal filtered_signal has the characteristic of tracking the phase of the harmonic signal;

[0038] S5: The calculation results are saved in a shift register, and the average operation is performed on each group of several numbers;

[0039] S6: Judge the operation result according to the orthogonality of trigonometric functions, that is:

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047] It can be concluded that signals of different frequencies are orthogonal to each other, and their calculation results are close to zero, while the product integral of signals of the same frequency is not equal to 0. Therefore, the cancellation of single-frequency signals can be extracted from the composite waveform, and the convergence state of each frequency point can be independently judged.

[0048] Furthermore, the specific implementation of step S4 is as follows:

[0049] S4.1: Perform AD conversion on the analog signal sent from the previous stage to obtain the digital signal error_signal corresponding to the residual harmonic current on the load side;

[0050] S4.2: Multiply the output filtered signal of the algorithm at a single frequency point, filtered_signal, by the digital signal error_signal corresponding to the residual harmonic current on the load side to obtain the real-time calculation result e(n) corresponding to the current frequency point, which is specifically expressed as:

[0051] e(n)=filtered_signal·error_signal

[0052] Where n represents the sampling time, n = 1, 2, 3, ..., L1. L1 represents the length of the shift register reg1;

[0053] S4.3: Use the shift register reg1 to save the e(n) signal and arrange it in chronological order; the length L1 of the shift register is determined according to the actual situation, and the larger the L1, the higher the reliability of the calculation result. The specific expression is:

[0054] reg1=[e(n), e(n-1),..., e(n-L1+1)]

[0055] S4.4: Repeat S4.2 and S4.3 until all frequency points are calculated.

[0056] Further, such as Figure 1 As shown, the specific implementation of step S5 is as follows:

[0057] S5.1: Perform summation and average operation on e(n), which is specifically expressed as:

[0058]

[0059] Where t represents the sampling time, t = 1, 2, 3, ..., L2. L2 represents the length of the shift register reg2. The smaller L2 is, the more sensitive the judgment of the convergence state is;

[0060] S5.2: The calculation result w(t) is saved using the shift register reg2 to form a moving time window queue data, which is specifically expressed as:

[0061] reg2=[w(t),w(t-1),…,w(t-L2+1)]

[0062] Further, such as Figure 2 As shown, the specific implementation of step S6 is as follows:

[0063] S6.1: After all the data stored in the shift register reg2 are refreshed once, the convergence state is judged;

[0064] S6.2: Select the queue data stored in the shift register reg2 for observation. When the head data w(t) and the tail data w(t-L2+1) are both greater than 0 and less than m, the frequency point is in a convergence state; when the head data w(t) and the tail data w(t-L2+1) are both close to 0, the harmonic cancellation channel corresponding to the frequency point is not started or the optimal value cannot be found; when the head data w(t) is greater than m, the algorithm has diverged and a signal needs to be sent to restart or shut down the active filter, where m is the judgment threshold. Since the values ​​of w(t) during convergence and divergence usually differ by more than ten times, it is easy to select according to actual conditions.

[0065] like Figure 3As shown, this embodiment uses three sine waves to form harmonic current, uses a narrowband LMS algorithm to track and offset the harmonic current, uses a correlation multiplication-based averaging method to judge the convergence state of the algorithm, and compares it with the method of calculating the root mean square of the error signal. Specifically, the following steps are included:

[0066] (1) Build a power grid simulation model, including a harmonic voltage source, an inductive source internal resistance, and a resistive load. The harmonic voltage source is a composite harmonic of 2kHz, 4kHz, and 6kHz sine waves, the inductive source internal resistance is 50μH, and the load is 0.3Ω;

[0067] (2) Build an adaptive active filter that uses a narrowband LMS algorithm to track and control current. The algorithm module uses a function module programmed in MATLAB. The input signal is a single-frequency reference sine wave, cosine wave, and the residual harmonic current of the load, that is, the error signal. The output filter control signal is output to the power grid through a full-bridge inverter controlled by SPWM. The frequency of the triangular wave carrier is 400kHz. The LC filter parameters of the inverter are 20μH and 10μF, and the DC power supply is set to an ideal 20V DC power supply. The current sampling link uses an ideal current sampler instead of the actual current sensor and filter amplification link.

[0068] (3) Build an algorithm convergence state judgment module, build an algorithm convergence state judgment module based on correlation multiplication and averaging, and build an algorithm convergence state judgment module based on root mean square for comparison results;

[0069] (4) Set the 2kHz amplitude of the harmonic voltage source to 20V, the 4kHz and 6kHz amplitudes to 0V, and the harmonic source to single frequency output;

[0070] (4.1) Start the 2kHz offset channel to test whether the convergence status judgment module works normally;

[0071] (4.2) Continue to simulate the situation where the channel is not opened and the algorithm diverges;

[0072] (4.3) Compare the effects of the two algorithms’ convergence state judgment methods. The results are shown in Table 1:

[0073] Table 1 Judgment results of two judgment methods for single frequency harmonics

[0074]

[0075] (5) The 2kHz amplitude of the harmonic voltage source is set to 20V, the 4kHz amplitude is set to 20V, and the 6kHz amplitude is set to 0V. The harmonic source is a dual-frequency output;

[0076] (5.1) Start the 2kHz and 4kHz cancellation channels to simulate the working conditions when the algorithm converges normally;

[0077] (5.2) Only the 2kHz cancellation channel is activated, and the 4kHz cancellation channel is not activated, to simulate the situation where there is interference from other frequencies, or the device is not fully working properly;

[0078] (5.3) Compare the effects of the two algorithms’ convergence state judgment methods. The results are shown in Table 2:

[0079] Table 2 Judgment results of two judgment methods for dual-frequency harmonics

[0080]

[0081] (6) The 2kHz amplitude of the harmonic voltage source is set to 20V, the 4kHz amplitude is set to 20V, and the 6kHz amplitude is set to 20V. The harmonic source is a three-frequency output;

[0082] (6.1) Start the 2kHz, 4kHz and 6kHz cancellation channels to simulate the working conditions when the algorithm converges normally;

[0083] (6.2) Only the 2kHz and 4kHz cancellation channels are activated, and the 6kHz cancellation channel is not activated, to simulate the situation where there is interference from other frequencies, or the device is not fully working properly;

[0084] (6.3) Compare the effects of the two algorithms’ convergence state judgment methods. The results are shown in Table 3:

[0085] Table 3 Judgment results of three-frequency harmonics by two judgment methods

[0086]

[0087] (7) The technical solution is verified based on the results. The results are shown in Table 4.

[0088] Table 4 Comparison of the effects of two judgment methods

[0089]

[0090] It can be seen from Table 4 that the convergence state judgment method based on correlation multiplication can distinguish the convergence state of a single frequency point, while the convergence state judgment method based on root mean square can only judge the overall convergence state when there are no other interference frequencies. This is because the convergence state judgment method based on root mean square can only calculate the total energy value of the residual error signal and cannot distinguish the frequency, which is applicable to wideband adaptive filters. However, narrowband adaptive filters can only offset a limited number of harmonic frequency points. The energy contained in the unprocessed frequency points will cause the convergence state judgment method based on root mean square to make misjudgments, thereby causing instability of the device. The convergence state judgment method based on correlation multiplication is more valuable in engineering practice.

[0091] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0092] It will be easily understood by those skilled in the art that the above description is only 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 in the protection scope of the present invention.

Claims

1. A method for judging the convergence state of an adaptive narrowband filter based on correlation multiplication, characterized in that: include: S1: Collect the harmonic current signal on the load side; S2: Low-pass filter the harmonic current signal on the load side to eliminate high-frequency clutter interference; S3: amplify the filtered harmonic current signal; S4: performing correlation multiplication operations on the amplified load-side residual harmonic current signal and the generated single-frequency output cancellation signal sine waves respectively; Step S4 includes: S4.1: Perform AD conversion on the analog signal sent from the previous stage to obtain the digital signal error-signal corresponding to the residual harmonic current on the load side; S4.2: multiply the output filtered signal of a single frequency point filtered_signal by the digital signal error_signal corresponding to the residual harmonic current on the load side to obtain the real-time calculation result e(n) corresponding to the current frequency point, where n represents the sampling time, n=1, 2, 3, ..., L1, L1 represents the length of the shift register reg1, and the larger L1 is, the higher the reliability of the calculation result; S4.3: Use the shift register reg1 to store the e(n) signal and arrange it in time sequence, where reg1 = [e(n), e(n-1), ..., e(n-L1+1)]; S4.4: Repeat S4.2 and S4.3 until all frequency points are calculated; S5: The result of the correlation multiplication operation is saved in a shift register, and an average operation is performed on each group of several numbers; Step S5 includes: S5.1: By Perform summation and averaging operations on e(n), where t represents the sampling time, t=1, 2, 3, ..., L2, L2 represents the length of the shift register reg2, and the smaller L2 is, the more sensitive the judgment of the convergence state is; S5.2: The calculation result w(t) is saved using the shift register reg2 to form a moving time window queue data, where reg2 = [w(t), w(t-1), ..., w(t-L2+1)]; S6: judging the average operation result, according to the orthogonality of trigonometric functions, it is concluded that signals of different frequencies are orthogonal to each other, and their average operation result is close to zero, while the product integral of signals of the same frequency is not equal to 0, and then the cancellation of single-frequency point signals is extracted from the composite waveform, and the convergence state of each frequency point is independently judged; Step S6 includes: S6.1: After all the data stored in the shift register reg2 are refreshed once, the convergence state is judged; S6.2: Select the queue data saved in the shift register reg2 for observation. When the head data w(t) and the tail data w(t-L2+1) are both greater than 0 and less than m, the current frequency point is in a convergence state; when the head data w(t) and the tail data w(t-L2+1) are both close to 0, the harmonic cancellation channel corresponding to the current frequency point is not started or the optimal value cannot be found; when the head data w(t) is greater than m, the algorithm has diverged and a signal needs to be sent to restart or shut down the active filter, where m is the judgment threshold.

Images