CN104935293B - High-power transformer adaptive active method for noise reduction control and control system - Google Patents

Abstract

The invention discloses a high-power transformer self-adaptive active noise reduction control method, comprising: a sensor collects primary noise as a reference signal and transmits it to an adaptive controller, and the self-adaptive controller outputs a control signal according to the reference signal as a secondary signal for driving The loudspeaker emits secondary noise; the primary sound field established by the noise emitted by the power transformer and the secondary sound field established by the secondary noise emitted by the loudspeaker are superimposed, and the superimposed sound pressure is collected by the error sensor and forms an error signal; self-adaptive After receiving the error signal, the controller adjusts the phase and amplitude of the secondary signal by using the LMS algorithm with the change of the convergence coefficient according to the preset objective function until the error signal satisfies the objective function and reaches a steady state. It can speed up the convergence speed of the algorithm and improve the noise reduction effect while ensuring the steady-state error performance, and the choice of the initial value of the convergence coefficient becomes more free.

Description

High Power Transformer Adaptive Active Noise Reduction Control Method and Control System

technical field

The invention relates to the technical field of high-power transformer noise reduction, in particular to a high-power transformer adaptive active noise reduction control method and control system.

Background technique

Transformer noise is the sum of irregular, intermittent, continuous or random mechanical noise and air noise generated by the transformer body and cooling system during the body structure design, type selection layout, installation, and use. The noise generated by transformers widely affects residential areas, commercial centers, light stations, airports, factories and mines, enterprises, hospitals, schools and other places. With the improvement of people's environmental awareness and the restrictions on various types of noise by the environmental protection department, especially due to the continuous expansion of cities and the needs of urban power grid transformation, some substations are sometimes built in commercial and residential areas, so the transformer noise problem is very important. became very prominent.

The adaptive active noise reduction system refers to a method in which the secondary signal is generated by the adaptive controller to make the speaker sound to cancel the noise. Due to the characteristics of the primary sound source and the surrounding environment changing at all times, the adaptive controller must be able to Only by adjusting the amplitude, frequency and phase of the secondary noise can a good noise reduction effect be achieved. The quality of the adaptive controller control method plays a decisive role in the quality of the noise reduction effect. At present, in the traditional control algorithm, each weight coefficient in the iterative process has the same convergence step size, which leads to the convergence speed Contradictory issues with steady-state error performance.

Contents of the invention

The purpose of the present invention is to solve the above problems, to provide a high-power transformer adaptive active noise reduction control method and control system, which can adjust the size of the convergence coefficient by itself, thereby adjusting the search direction and convergence speed of the control method to obtain the best The superiority coefficient and better adaptability, the noise reduction processing effect is more stable, and a good noise reduction effect has been achieved.

In order to achieve the above object, the present invention adopts the following technical solutions:

A high-power transformer adaptive active noise reduction control method, the filter in the adaptive controller adopts an FIR filter, including the following steps:

Step 1, the sensor collects the primary noise as a reference signal x(n) and transmits it to the adaptive controller, and the adaptive controller outputs a control signal as the secondary signal y(n) according to the reference signal to drive the speaker to emit secondary noise;

Step 2, the primary sound field established by the noise emitted by the power transformer and the secondary sound field established by the secondary noise emitted by the loudspeaker are superimposed, the superimposed sound pressure is collected by the error sensor, and an error signal e(n) is formed;

Step 3: After receiving the error signal e(n), the adaptive controller adjusts the phase and amplitude of the secondary signal by using the LMS algorithm with the change of the convergence coefficient according to the preset objective function J(n), until the error signal satisfies The objective function J(n), reaches a steady state.

The output of the secondary signal y(n) at the nth moment is expressed in vector form as follows:

y(n)=X ^T (n)W=W ^T X(n)

where X(n)=[x(n),x(n-1),...,x(n-L+1)] ^T , W=[w ₁ ,w ₂ ,...,w _L ] ^T , w _l (n) is the weight coefficient, L is the length of the filter; the error signal e(n) at the nth moment is:

e(n)=d(n)-y(n)=d(n)-W ^T X(n)

The goal is to minimize the mean square error between the desired signal d(n) and the secondary signal y(n), and the objective function J(n) is obtained from the desired signal d(n) and the secondary signal y(n).

Described objective function J (n) is the mean square error of expected signal d (n) and secondary signal y (n), namely:

J(n)=E[e ² (n)]=E[(d(n)-W ^T X(n)) ² ]

＝E[d ² (n)]+W ^T E[X(n)X ^T (n)]W-2W ^T E[d(n)X(n)]

＝E[d ² (n)]+W ^T RW-2W ^T P

Wherein, R is the autocorrelation matrix of the input signal x(n), R=E[X(n)X ^T (n)]; P is the cross-correlation vector, P=E[d(n)X(n)], When using the LMS algorithm with changing convergence coefficient, the optimal weight vector W ^* is obtained by iterative method. When the optimal weight vector W ^* is taken, the objective function is the smallest.

The iterative function used in the process of obtaining the optimal weight vector W ^* using the iterative method when using the LMS algorithm with the change of the convergence coefficient is:

W(n+1)=W(n)+2μ(n)X(n)e(n)

Among them, μ(n) is a diagonal matrix of L×L:

μ _l (n)(l=0,1,...,L-1) is the convergence coefficient, and the specific value method is as follows:

where α is greater than 1.

The stable state is when the difference between the amplitude h _t of the error signal and the amplitude h _∞ at the steady state moment is within 5%.

The high-power transformer adaptive active noise reduction control system includes a reference sensor, an error sensor and an adaptive controller. The reference sensor is connected to the input end of the adaptive controller through a first preamplifier, and the error sensor is connected through a second The preamplifier is connected to the input end of the adaptive controller, and the output end of the adaptive controller is connected to the loudspeaker through the power amplifier.

The reference sensor and the error sensor use condenser microphones.

Beneficial effects of the present invention:

This control method can speed up the convergence speed of the algorithm while ensuring the steady-state error performance, and improve the noise reduction effect. The choice of values also becomes more free, effectively reducing the primary noise.

Description of drawings

Fig. 1 is the system working schematic diagram of the present invention;

Fig. 2 is a mathematical model diagram of an active noise reduction system;

Figure 3(a) is a simplified diagram of the continuous domain of the active noise reduction system model; Figure 3(b) is a simplified diagram of the discrete domain of the active noise reduction system model;

Fig. 4 is a schematic diagram of an adaptive filtering system;

Fig. 5 is the schematic diagram of FIR filter;

Figure 6(a) is the time-domain waveform diagram of the original noise of the transformer, and Figure 6(b) is the amplitude-frequency characteristic diagram of the original noise of the transformer;

Figure 7(a) is a simulation diagram of residual noise signal after adaptive filtering, and Figure 7(b) is a simulation diagram of amplitude-frequency characteristics after adaptive filtering,

Figure 7(c) is a simulation diagram of the secondary signal after adaptive filtering;

Fig. 8 (a) is the residual noise signal simulation figure of the present invention, and Fig. 8 (b) is the amplitude-frequency characteristic simulation figure of the present invention,

Fig. 8(c) is the learning curve diagram of the present invention, and Fig. 8(d) is the value of the weight coefficient.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, the high-power transformer adaptive active noise reduction control system includes a reference sensor, an error sensor and an adaptive controller, and the reference sensor is connected to the input end of the adaptive controller through the first preamplifier, so The error sensor is connected to the input end of the adaptive controller through the second preamplifier, and the output end of the adaptive controller is connected to the speaker through the power amplifier. The reference sensor and the error sensor use condenser microphones.

The signals collected by the reference sensor and the error sensor are generally weak, so they must be amplified by the first preamplifier and the second preamplifier before they can be used as the input of the adaptive controller. Likewise, the signal output by the adaptive controller is insufficient to drive the speaker, so a power amplifier is needed for proportional amplification.

It can be clearly seen from Figure 1 that the entire adaptive noise reduction system mainly consists of three loops to complete the entire noise reduction process. The function of the primary loop is to pick up the primary noise as the reference signal of the system, and the secondary loop completes the amplification and output of the secondary signal. , the feedback loop sends the reference signal and the error signal to the adaptive controller for operation to generate a secondary signal.

A high-power transformer adaptive active noise reduction control method, the filter in the adaptive controller adopts an FIR filter, including the following steps:

A high-power transformer adaptive active noise reduction control method, the filter in the adaptive controller adopts an FIR filter, including the following steps:

Step 1, the sensor collects the primary noise as a reference signal x(n) and transmits it to the adaptive controller, and the adaptive controller outputs a control signal as the secondary signal y(n) according to the reference signal to drive the speaker to emit secondary noise;

Step 2, the primary sound field established by the noise emitted by the power transformer and the secondary sound field established by the secondary noise emitted by the loudspeaker are superimposed, the superimposed sound pressure is collected by the error sensor, and an error signal e(n) is formed;

Step 3: After receiving the error signal e(n), the adaptive controller adjusts the phase and amplitude of the secondary signal by using the LMS algorithm with the change of the convergence coefficient according to the preset objective function J(n), until the error signal satisfies The objective function J(n), reaches a steady state.

Assume that the frequency response functions of the adaptive controller, reference signal sensor, and error sensor are W(ω), S ₁ (ω) and S ₂ (ω) respectively, and the frequency response of the secondary sound source P is P(ω). Assuming that the transfer function of the propagation path of the sound wave from the transformer to the reference signal sensor _S1 is H _S1 (ω), the transfer functions from the transformer and the secondary sound source P to the error sensor are H _S2 (ω), H _PS ( ω). Assume that the frequency response of devices such as preamplifiers in the primary loop is N ₁ (ω), the frequency response of devices such as power amplifiers in the secondary loop is N ₂ (ω), and the frequency response of devices such as preamplifiers in the feedback loop is N ₃ (ω). The mathematical model of the entire transformer intelligent active noise reduction system is shown in Figure 2.

In Fig. 2, p(t), x(t) and e(t) are primary signal, reference signal and error signal respectively. The transfer functions of each loop are arranged as follows:

H _r (ω)＝S ₁ (ω)H _S1 (ω)N ₁ (ω) (1)

H _p (ω) = H _ps (ω) (2)

H _S (ω)＝H _S2 (ω)N ₂ (ω)P(ω) (3)

H _f (ω)＝S ₂ (ω)N ₃ (ω) (4)

It can be seen from the above that H _r (ω), H _p (ω), H _s (ω) and H _f (ω) are the transfer functions of the reference loop, primary loop, secondary loop and feedback loop respectively, so Figure 2 can be simplified It is Figure 3(a). For the convenience of subsequent calculations, the simplified diagram can be transformed into a discrete domain, as shown in Figure 3(b), and the transfer functions of the controller and each loop can be written as W(z), _Hr (z), H _p (z), H _s (z), and H _f (z).

It can be clearly seen from Figure 3 that the core of the system is the control algorithm of the adaptive controller when the transfer function of each loop is known.

The adaptive filter is an optimal filter based on the minimum mean square error or the least square method, which can automatically adjust its unit pulse to achieve the best optimization effect. It can be divided into two parts, one part is the filter to complete the filtering task, and the other part is the control algorithm to complete the coefficient adjustment task. Its system principle is shown in Fig. 4.

In Fig. 4, x(n) and y(n) are reference signal and secondary signal respectively, d(n) is expected signal, e(n) is error signal. The adaptive filter can minimize the mean square error between d(n) and y(n), so the mean square error is the objective function of the system, expressed by J(n), that is

J(n)=E[e ² (n)]=E[(d(n)-y(n)) ² ] (5)

The adaptive filter structure is a transverse structure finite-duration impulse response (Finite-duration Impulse Response, FIR) filter made of tap delay, and its tap weight coefficient set is exactly equal to its impulse response. The FIR filter is a filter with a non-recursive structure, which has two characteristics: firstly, it has a linear phase-frequency characteristic, which can ensure that the signal will not be distorted in the passband; secondly, it is an unconditionally stable causal system, without Feedback loop, containing zero only. Its structure is shown in Figure 5, w _l (n) is the weight coefficient, assuming the length of the filter is L, then the output at the nth moment is:

It can be seen from formula (6) that the output signal y(n) is the linear weighted sum of the first L input signals x(n), then the input signal and weight coefficient of the filter are expressed in vector form:

X(n)=[x(n),x(n-1),...,x(n-L+1)] ^T (7)

W＝[w ₁ ,w ₂ ,...,w _L ] ^T (8)

Then, formula (6) can be written as:

y(n)= ^XT (n)W= ^WTX (n) (9)

From this, the error signal at the nth moment can be written:

e(n)=d(n)-y(n)=d(n)-W ^T X(n) (10)

Therefore, the objective function can be transformed as follows:

J(n)=E[e ² (n)]=E[(d(n)-W ^T X(n)) ² ]

＝E[d ² (n)]+W ^T E[X(n)X ^T (n)]W-2W ^T E[d(n)X(n)]

＝E[d ² (n)]+W ^T RW-2W ^T P (11)

In the formula, the autocorrelation matrix of R—input signal x (n), R=E[X (n) X ^T (n)];

P—cross-correlation vector, P=E[d(n)X(n)]

It can be seen from formula (11) that the objective function J(n) is a quadratic function of the weight coefficient vector W, and its shape is a concave surface, so it has a unique minimum value. Therefore, when the gradient of the objective function J(n) is equal to zero, the optimal weight vector W ^* can be obtained. Therefore, let the gradient equation of the objective function J(n) to the weight vector W ^* be zero:

If the matrix R is a full-rank matrix and its inverse matrix R ^-1 exists, the optimal weight vector can be obtained as:

W ^* = R ^-1 P (13)

Although formula (13) gives the optimal weight vector to obtain the minimum objective function, it is necessary to pre-calculate the autocorrelation matrix R and cross-correlation vector P of the input signal x(n), and invert the autocorrelation matrix. However, in actual engineering, it is difficult to directly obtain the autocorrelation matrix R and the cross-correlation matrix P, which must be obtained by calculation. However, when the filter length L is relatively large, it is bound to seriously affect the computational complexity of the autocorrelation matrix R and the cross-correlation matrix P. Therefore, using an iterative method to estimate the optimal weight vector becomes a feasible solution. At present, there are many kinds of adaptive algorithms, most of which are derived from the least mean square error method (LMS) and the least square method (LS).

The adaptive process is the process of gradually approaching the minimum value of the objective function by continuously adjusting the weight vector, and its ultimate goal is to seek the optimal weight vector. The least mean square algorithm can achieve this function. It is a simple and effective iterative algorithm. It does not need to know the autocorrelation matrix in advance, and it does not need to invert the matrix. It only needs to use the steepest descent method to obtain the weight vector. push formula.

The process of the existing LMS algorithm is as follows.

According to the principle of the steepest descent method, the weight vector W(n+1) at the n+1th moment is equal to the weight vector W(n) at the nth moment minus a change proportional to the gradient ▽(n), as shown in the formula (14 ) as shown:

W(n+1)=W(n)-μ▽(n) (14)

In the formula, μ—convergence coefficient, which affects the convergence speed of the algorithm;

▽(n)—the gradient of the objective function J(n) to the weight vector W ^* , namely

In order to improve the real-time performance of the system, the instantaneous value can be used as the estimated value of formula (15), namely:

It can be proved that the estimated gradient vector value is an unbiased estimate of the true gradient vector, then:

But there are still some differences between the two:

In the formula, N(n)—gradient noise.

Gradient noise arises because only a limited number of inputs are used during each iteration, thus causing some bias in the gradient estimate.

then use Instead of ▽(n) in formula (14), we can get:

The above method of deriving the weight vector with an iterative algorithm is the existing minimum mean square error algorithm.

In the existing LMS algorithm, each weight coefficient has the same convergence step in the iterative process, which leads to the contradiction between the convergence speed and the steady-state error performance. In response to this problem, this patent improves the traditional LMS algorithm, so that the algorithm can adjust the size of the convergence coefficient by itself, thereby adjusting the search direction and convergence speed of the algorithm to obtain the optimal weight coefficient and better adaptability.

Make the following changes to formula (19):

W(n+1)=W(n)+2μ(n)X(n)e(n) (21)

In the formula, μ(n)—diagonal matrix of L×L:

In order to ensure the convergence of the adaptive algorithm, the value range of μ _l (n) (l=0,1,...,L-1) must be limited between the maximum value and the minimum value, that is, the formula ( 9). Under this condition, its value can be divided into three situations. 1. If the sign of x(n-1)e(n) does not change after multiple consecutive moments, then increase μ _l (n) by α times; 2. If x(n-1)e(n) passes through After several consecutive moments, the sign changes, then decrease μ _l (n) by α times; other conditions remain unchanged. The following formula can be obtained:

α is greater than 1.

In order to verify the effect of the adaptive filtering algorithm in the active noise reduction process, the present invention also conducts experiments on the existing algorithm. The original noise signal of the transformer is used as the input signal of the adaptive filter, as shown in Figure 6(a)-Figure 6(b), and then the output signal of the adaptive filter is superimposed with the noise signal of the transformer to simulate the error signal. The signal is the signal after noise reduction, and an adaptive filter is introduced as the objective function. The adaptive filter length is set to 256 orders, and the convergence coefficient is 0.05. Figure 7(a)-Figure 7(c) respectively shows the error signal and its amplitude-frequency characteristics and the output signal of the adaptive filter.

In order to understand the convergence speed of the adaptive algorithm, the present invention defines that when the amplitude _h of the error signal at a certain moment is within 5% of the amplitude h _∞ of the steady-state moment, it is considered that the algorithm reaches a steady state at this moment, that is:

Comparing Figure 6, it can be seen that the error signal reaches a steady state after 295ms, and the amplitude drops significantly. Comparing the amplitude-frequency characteristics, it can be seen that the amplitude at 100Hz and its multiplied frequency drops from above 0.02 to below 0.006, which has an obvious noise reduction effect, especially at 300Hz. Comparing the secondary signal and the primary signal, it can be found that the phase difference is almost 180°, so the amplitude of the error signal will be significantly reduced after the two are superimposed.

The control method of the present invention is also simulated and verified, assuming that the initial value of the convergence coefficient μ _l (l=0,1,...,L-1) is 0.05, the length of the filter is set to 256 orders, and the value of α is 2, The results shown in Figure 8(a)-Figure 8(d) can be obtained.

From Figure 8(a)-Figure 8(d), it can be seen that the residual noise reached a stable state at about 300ms, and the amplitude finally converged at about 0.015, achieving a good noise reduction effect; the learning curve shows the amplitude of the residual noise From the perspective of amplitude-frequency characteristics, the amplitude of the main frequency components of the primary noise dropped below 0.0016, especially the 300Hz component in the original primary noise dropped most obviously; the final value of the weight coefficient fluctuated around zero.

The improved adaptive algorithm can accelerate the convergence of the algorithm to a certain extent and improve the noise reduction effect while ensuring the steady-state error performance. To sum up, the improved adaptive algorithm solves the problem that the traditional algorithm cannot balance the convergence speed and steady-state error performance to a certain extent, and the selection of the initial value of the convergence coefficient is also more convenient.

Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims (4)

1. high-power transformer adaptive active method for noise reduction control, adaptive controller median filter uses FIR filter, It is characterized in, comprises the following steps：

Step 1, sensor acquisition primary noise, which is used as, to be transferred to reference to signal x (n) in adaptive controller, self adaptive control Device exports a control signal according to this reference signal and sends secondary noise as secondary signal y (n) drive the speakers；

Step 2, the primary sound field that the noise that power transformer is sent is established are established with the secondary noise that loud speaker is sent Secondary sound field generates superposition, collects the acoustic pressure after superposition by error pick-up, and forms error signal e (n)；

After adaptive controller receives error signal e (n), convergence system is utilized according to default object function J (n) for step 3 The LMS algorithm of variation is counted to adjust the phase and amplitude of secondary signal, continues to meet object function J (n) to error signal, reach Stable state；

Secondary signal y (n) outputs at the n-th moment are represented with vector form：

Y (n)=X^T(n) W=W^TX(n)

Wherein X (n)=[x (n), x (n-1) ..., x (n-L+1)]^T, W=[w₁,w₂,...,w_L]^T, w_l(n) it is weight coefficient, L is The length of wave filter；The error signal e (n) at the n-th moment is：

E (n)=d (n)-y (n)=d (n)-W^TX(n)

Can make the minimum target of the mean square error between desired signal d (n) and secondary signal y (n), by desired signal d (n) and Secondary signal y (n) obtains object function J (n)；

The object function J (n) is the mean square error of desired signal d (n) and secondary signal y (n), i.e.,：

J (n)=E [e²(n)]=E [(d (n)-W^TX(n))²]

=E [d²(n)]+W^TE[X(n)X^T(n)]W-2W^TE[d(n)X(n)]

=E [d²(n)]+W^TRW-2W^TP

Wherein, R be input signal x (n) autocorrelation matrix, R=E [X (n) X^T(n)]；P is cross-correlation vector, P=E [d (n) X (n)] optimal weight vector W is obtained using iterative method during the LMS algorithm, changed using convergence coefficient^*, take optimal weight vector W^*When, mesh Scalar functions are minimum；

Optimal weight vector W is obtained using iterative method during the LMS algorithm changed using convergence coefficient^*Used iteration letter in the process Number is：

W (n+1)=+ 2 μ (n) X (n) e (n) of W (n)

Wherein, μ (n) is the diagonal matrix of L × L：

<mrow> <mi>&amp;mu;</mi> <mrow> <mo>(</mo> <mi>n</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <msub> <mi>&amp;mu;</mi> <mn>0</mn> </msub> <mrow> <mo>(</mo> <mi>n</mi> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mn>0</mn> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <msub> <mi>&amp;mu;</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>n</mi> <mo>)</mo> </mrow> </mrow> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mn>0</mn> </mtd> </mtr> <mtr> <mtd> <mo>...</mo> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mo>...</mo> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mn>0</mn> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mrow> <msub> <mi>&amp;mu;</mi> <mrow> <mi>L</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>n</mi> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>

μ_l(n) it is convergence coefficient, l=0,1 ..., L-1；Specifically obtaining value method is：

Wherein α is more than 1.

2. high-power transformer adaptive active method for noise reduction control as described in claim 1, it is characterized in that, the stable state For the amplitude h of error signal_tWith stable state moment amplitude h_∞When within difference 5%.

3. high-power transformer adaptive active method for noise reduction control as described in claim 1, it is characterized in that, using high-power change The control system of depressor adaptive active method for noise reduction control includes reference sensor, error pick-up and adaptive controller, The reference sensor connects the input terminal of adaptive controller by the first preamplifier, and the error pick-up passes through the Two preamplifiers connect the input terminal of adaptive controller, and the output terminal of the adaptive controller is connected by power amplifier Connect loud speaker.

4. high-power transformer adaptive active method for noise reduction control as claimed in claim 3, it is characterized in that, it is described with reference to sensing Device and error pick-up use condenser microphone.