CN104935293A - Large power transformer self-adaptive active noise reduction control method and control system - Google Patents

Abstract

The invention discloses a large power transformer self-adaptive active noise reduction control method. The method comprises the steps that a sensor collects primary noise as a reference signal and transmits the reference signal to a self-adaptive controller, and the self-adaptive controller outputs, according to the reference signal, a control signal to be used as a secondary signal so as to drive a loudspeaker to give out secondary noise; a primary sound field established by the noise made by a power transformer and a secondary sound field established by secondary noise made by the loudspeaker overlap, and an error sensor collects the overlapped sound pressure and forms an error signal; and the self-adaptive controller receives the error signal, and then adjusts, according to a preset objective function, the phase and amplitude of the secondary signal by use of a least mean square (LMS) algorithm for converging coefficient variation till the error signal satisfies the objective function and achieves a stable state. Through adoption of the method, while the steady state error performance is guaranteed, the converging speed of the algorithm is improved, and the noise reduction effect can be improved; and the initial value of the convergence coefficient can be selected more freely.

Description

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

Technical field

The present invention relates to high-power transformer noise reduction technology field, particularly relate to a kind of high-power transformer adaptive active method for noise reduction control and control system.

Background technology

Transformer noise is by body construction design, type selecting layout, installation, use procedure, the machinery noise that is irregular, that cause intermittently, continuously or at random that transformer body and cooling system produce and airborne noise summation.The noise that transformer produces extensively affects the place such as residential quarters, commercial center, light station, airport, factories and miness, enterprise, hospital, school.Along with the raising of people's environmental consciousness and environmental administration are to the restriction of each noise like, particularly due to the continuous expansion in city and the demand of city network transformation, some transformer stations will be built in shopping centre and residential block sometimes, so transformer noise problem is with regard to extremely highlighting of becoming.

Adaptive active noise reduction system refers to that producing secondary signal by adaptive controller computing makes loud speaker sounding to offset a kind of method of noise, due to the characteristic that elementary sound source and surrounding environment moment change, adaptive controller must adjust the amplitude of secondary noise, frequency and phase place the moment, just can obtain good noise reduction.The quality of quality to noise reduction of adaptive controller control method plays decisive role, in current traditional control algorithm, each weight coefficient in an iterative process, its convergence step-length is all identical, which results in the contradictory problems between convergence rate and steady-state error performance.

Summary of the invention

Object of the present invention is exactly to solve the problem, a kind of high-power transformer adaptive active method for noise reduction control and control system are provided, can from the size of Row sum-equal matrix convergence coefficient, thus the direction of search of adjustment control method and convergence rate, to obtain optimum weight coefficient and better adaptivity, noise reduction process effect is more stable, achieves good noise reduction.

To achieve these goals, the present invention adopts following technical scheme:

High-power transformer adaptive active method for noise reduction control, adaptive controller median filter adopts FIR filter, comprises the following steps:

Step one, transducer gathers primary noise as being delivered in adaptive controller with reference to signal x (n), and adaptive controller exports a control signal according to this reference signal and drives loud speaker to send secondary noise as secondary signal y (n);

Step 2, the secondary sound field that the noise that power transformer the sends primary sound field set up and the secondary noise that loud speaker sends are set up produces and superposes, and collects the acoustic pressure after superposition, and form error signal e (n) by error pick-up;

Step 3, after adaptive controller receives error signal e (n), phase place and the amplitude of secondary signal is adjusted according to target function J (n) the LMS algorithm that utilizes convergence coefficient to change preset, continue to error signal and meet target function J (n), reach stable state.

Secondary signal y (n) output in the n-th moment represents with vector form and is specially:

y(n)＝X ^T(n)W＝W ^TX(n)

Wherein X (n)=[x (n), x (n-1) ..., x (n-L+1)] ^t, W=[w ₁, w ₂..., w _l] ^t, w _ln () is weight coefficient, L is the length of filter; The error signal e (n) in the n-th moment is:

e(n)＝d(n)-y(n)＝d(n)-W ^TX(n)

The mean square error between desired signal d (n) and secondary signal y (n) can be made minimum be target, obtain target function J (n) by desired signal d (n) and secondary signal y (n).

Described target function J (n) is desired signal d (n) and the mean square error of secondary signal y (n), that is:

J(n)＝E[e ²(n)]＝E[(d(n)-W ^TX(n)) ²]

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

＝E[d ²(n)]+W ^TRW-2W ^TP

Wherein, R is the autocorrelation matrix of input signal x (n), R=E [X (n) X ^t(n)]; P is cross-correlation vector, P=E [d (n) X (n)], adopts iterative method to obtain optimum weight vector W during the LMS algorithm utilizing convergence coefficient to change ^*, get optimum weight vector W ^*time, target function is minimum.

Iterative method is adopted to obtain optimum weight vector W during the LMS algorithm utilizing convergence coefficient to change ^*the iteration function adopted in process is:

W(n+1)＝W(n)+2μ(n)X(n)e(n)

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

$\mathrm{\μ}\left(n\right)=\left[\begin{array}{cccc}{\mathrm{\μ}}_0\left(n\right)& 0& ...& 0\\ 0& {\mathrm{\μ}}_1\left(n\right)& ...& 0\\ ...& ...& ...& ...\\ 0& 0& ...& {\mathrm{\μ}}_{L-1}\left(n\right)\end{array}\right]$

μ _l(n) (l=0,1 ..., L-1) be convergence coefficient, concrete obtaining value method is:

Wherein α is greater than 1.

Described stable state is the amplitude h of error signal _twith stable state moment amplitude h _∞time within difference 5%.

High-power transformer adaptive active noise-reduction control system, comprise reference sensor, error pick-up and adaptive controller, described reference sensor connects the input of adaptive controller by the first preamplifier, described error pick-up connects the input of adaptive controller by the second preamplifier, and the output of described adaptive controller connects loud speaker by power amplifier.

Described reference sensor and error pick-up adopt condenser microphone.

Beneficial effect of the present invention:

This control method can accelerate convergence of algorithm speed when ensureing steady-state error performance, and improve noise reduction, alleviate the contradiction that traditional algorithm cannot take into account convergence rate and steady-state error performance simultaneously to a certain extent, and the selection of convergence coefficient initial value also becomes more free, effectively reduces primary noise.

Accompanying drawing explanation

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

Fig. 2 is active noise reduction system Mathematical Modeling figure;

Fig. 3 (a) is active noise reduction system model continuous domain reduced graph; Fig. 3 (b) is active noise reduction system model discrete domain reduced graph;

Fig. 4 is Avaptive filtering system schematic diagram;

Fig. 5 is FIR filter schematic diagram;

Fig. 6 (a) is transformer raw noise time domain beamformer, and Fig. 6 (b) is transformer raw noise amplitude-frequency characteristic figure;

Fig. 7 (a) for residual noise signal simulation figure, Fig. 7 (b) after adaptive-filtering be amplitude-frequency characteristic Simulation figure after adaptive-filtering,

Fig. 7 (c) is secondary signal analogous diagram after adaptive-filtering;

Fig. 8 (a) for residual noise signal simulation figure, Fig. 8 (b) of the present invention be amplitude-frequency characteristic analogous diagram of the present invention,

Fig. 8 (c) for learning curve figure of the present invention, Fig. 8 (d) be weight coefficient value.

Embodiment

Below in conjunction with accompanying drawing and embodiment, the invention will be further described.

As shown in Figure 1, high-power transformer adaptive active noise-reduction control system, comprise reference sensor, error pick-up and adaptive controller, described reference sensor connects the input of adaptive controller by the first preamplifier, described error pick-up connects the input of adaptive controller by the second preamplifier, and the output of described adaptive controller connects loud speaker by power amplifier.Described reference sensor and error pick-up adopt condenser microphone.

The signal that reference sensor and error pick-up collect is general more weak, therefore must could as the input of adaptive controller through the amplification of the first preamplifier and the second preamplifier.Equally, the signal that adaptive controller exports is not enough to drive loud speaker sounding, and therefore required power amplifier carries out scale amplifying.

Whole adaptive noise reduction system obviously can be found out primarily of three loops to complete whole noise reduction process by Fig. 1, the function of primary return is the reference signal of pickup primary noise as system, secondary loop completes amplification and the output services of secondary signal, and feedback loop then Reference Signal and error signal is delivered to adaptive controller computing generation secondary signal.

High-power transformer adaptive active method for noise reduction control, adaptive controller median filter adopts FIR filter, comprises the following steps:

High-power transformer adaptive active method for noise reduction control, adaptive controller median filter adopts FIR filter, comprises the following steps:

Step one, transducer gathers primary noise as being delivered in adaptive controller with reference to signal x (n), and adaptive controller exports a control signal according to this reference signal and drives loud speaker to send secondary noise as secondary signal y (n);

Step 2, the secondary sound field that the noise that power transformer the sends primary sound field set up and the secondary noise that loud speaker sends are set up produces and superposes, and collects the acoustic pressure after superposition, and form error signal e (n) by error pick-up;

Step 3, after adaptive controller receives error signal e (n), phase place and the amplitude of secondary signal is adjusted according to target function J (n) the LMS algorithm that utilizes convergence coefficient to change preset, continue to error signal and meet target function J (n), reach stable state.

Suppose that the frequency response function of adaptive controller, reference signal transducer, error pick-up is respectively W (ω), S ₁(ω) and S ₂(ω), the frequency response of secondary sound source P is P (ω).Suppose that sound wave propagates into reference signal transducer S from transformer ₁the transfer function of propagation path be H _s1(ω), H is respectively from the transfer function transformer and secondary sound source P to error pick-up _s2(ω), H _pS(ω).To suppose in primary return that the device frequency responses such as preamplifier are N ₁(ω), the response of the device frequency such as secondary loop intermediate power amplifier is N ₂(ω) frequency response of the devices such as the preamplifier, in feedback loop is N ₃(ω).The Mathematical Modeling of whole intelligent transformer active noise reduction system as shown in Figure 2.

In Fig. 2, p (t), x (t) and e (t) are respectively primary signal, reference signal and error signal.The transfer function in each loop is done following arrangement:

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)

As from the foregoing, H _r(ω), H _p(ω), H _s(ω) and H _f(ω) be respectively the transfer function of reference loop, primary return, secondary loop and feedback loop, therefore Fig. 2 can be reduced to Fig. 3 (a).In order to the convenience of subsequent calculations, this reduced graph can be converted into discrete domain, as shown in Fig. 3 (b), correspondingly the transfer function in controller and each loop can be designated as W (z), H _r(z), H _p(z), H _s(z) and H _f(z).

Can obviously be found out by Fig. 3, when known each return transfer function, the core of this system is the control algolithm of adaptive controller.

Sef-adapting filter is a kind of optimum filter being criterion with least mean-square error or least square method, can automatically regulate its unit pulse to reach optimum optimization effect.It can be divided into two parts, a part be filter to complete filter task, another part is that control algolithm is to complete the adjustment task of coefficient.Its system principle as shown in Figure 4.

In Fig. 4, x (n) and y (n) is respectively reference signal and secondary signal, and d (n) is desired signal, and e (n) is then error signal.Sef-adapting filter can make the mean square error between d (n) and y (n) minimum, and therefore this mean square error is the target function of system, represents with J (n), namely

J(n)＝E[e ²(n)]＝E[(d(n)-y(n)) ²] (5)

Sef-adapting filter structure is that the transversary utilizing tap time delay to make has limit for length's impulse response (Finite-durationImpulse Response, FIR) filter, and its tap weighting coefficients collection just in time equals its impulse response.FIR filter is a kind of filter of onrecurrent structure, has two features: first it has linear phase-frequency characteristic, can ensure that signal distortion can not occur in passband; Secondly it is the causal system of a unconditional stability, does not have feedback loop, only comprises zero point.Its structure as shown in Figure 5, w _ln () is weight coefficient, suppose that the length of filter is L, then the output in the n-th moment is:

$y\left(n\right)=\underset{l=1}{\overset{L}{\Σ}}{w}_l\left(n\right)x(n-l+1)---\left(6\right)$

As can be seen from formula (6), output signal y (n) is the linear weighted function sum of front L input signal x (n), so represents input signal and the weight coefficient of filter respectively with vector form:

X(n)＝[x(n),x(n-1),...,x(n-L+1)] ^T(7)

W＝[w ₁,w ₂,...,w _L] ^T(8)

So, formula (6) just can be write as:

y(n)＝X ^T(n)W＝W ^TX(n) (9)

The error signal in the n-th moment can be write out thus:

e(n)＝d(n)-y(n)＝d(n)-W ^TX(n) (10)

So, can target function be done as down conversion:

J(n)＝E[e ²(n)]＝E[(d(n)-W ^TX(n)) ²]

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

＝E[d ²(n)]+W ^TRW-2W ^TP (11)

In 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)]

As can be seen from formula (11), target function J (n) is the Quadratic Function Optimization of weight coefficient vector W, and so its shape is concave Iy curved, therefore has unique minimum.So, when target function J (n) gradient equals zero, optimum weight vector W just can be obtained ^*.So, make target function J (n) to weight vector W ^*gradient equations be zero:

$\frac{\∂J\left(n\right)}{\∂W}=2RW-2P=0---\left(12\right)$

If matrix R is full rank battle array, and its inverse matrix R ^-1exist, then can try to achieve best weights vector is:

W ^*＝R ^-1P (13)

Although formula (13) gives the optimum weight vector obtaining minimum target function, autocorrelation matrix R and the cross-correlation vector P of input signal x (n) must be precomputed, and autocorrelation matrix is inverted.But in Practical Project, being difficult to directly obtain autocorrelation matrix R and cross-correlation matrix P, must obtaining by calculating.But when filter length L is larger, autocorrelation matrix R and cross-correlation matrix P computation complexity will certainly be had a strong impact on.Therefore, the method for iteration is used to estimate that optimum weight vector becomes a feasible program.At present, adaptive algorithm has had a variety of, and great majority are derived by LMSE method (LMS) and least square method (LS),

Adaptive process is exactly the process by successively regulating weight vector to approach target function minimum value gradually, and its final purpose seeks optimum weight vector.And least mean square algorithm just can realize this function, it is simply a kind of and effective iterative algorithm, without the need to knowing autocorrelation matrix in advance, also need not, to matrix inversion, only need to adopt steepest descent method to obtain the recurrence formula of weight vector.

The process of existing LMS algorithm is as hereafter.

According to steepest descent method principle, the weight vector W (n+1) in the (n+1)th moment equals weight vector W (n) in the n-th moment and deducts the variable quantity being proportional to gradient ▽ (n), as shown in formula (14):

W(n+1)＝W(n)-μ▽(n) (14)

In formula, μ-convergence coefficient, affects convergence of algorithm speed;

▽ (n)-target function J (n) is to weight vector W ^*gradient, namely

$\▿\left(n\right)=\frac{\∂J\left(n\right)}{\∂W}=E\left[\begin{array}{c}\frac{\∂e^2\left(n\right)}{\∂w_0\left(n\right)}\\ ...\\ \frac{\∂e^2\left(n\right)}{\∂w_{N-1}\left(n\right)}\end{array}\right]=-2E\[e\left(n\right)X\left(n\right)\]---\left(15\right)$

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

$\widehat{\▿}\left(n\right)=\frac{\∂e^2\left(n\right)}{\∂W}=-2e\left(n\right)X\left(n\right)---\left(16\right)$

Can prove, estimate that gradient vector value is the unbiased esti-mator of true gradient vector, then have:

${}_{E\[\widehat{\▿}\left(n\right)\]=E\[\▿\left(n\right)\]---\left(17\right)}$

But both still have certain difference:

$\widehat{\▿}\left(n\right)=\▿\left(n\right)+N\left(n\right)\text{---}\left(18\right)$

In formula, N (n)-gradient noise.

The appearance of gradient noise is because in each iterative process, only uses limited input quantity, therefore causes Gradient estimates value to there is certain deviation.

So, use the ▽ (n) in formula (14) is replaced to obtain:

$\begin{array}{c}W(n+1)=W\left(n\right)-\mathrm{\μ}\widehat{\▿}\left(n\right)\\ \text{=W}\left(n\right)+\text{2}\mathrm{\μ}e\left(n\right)X\left(n\right)\end{array}---\left(19\right)$

Be exactly existing least-mean-square error algorithm by the derive method of weight vector of iterative algorithm above.

In existing LMS algorithm, each weight coefficient in an iterative process, and its convergence step-length is all identical, which results in the contradictory problems between convergence rate and steady-state error performance.For this problem, this patent makes improvement to traditional LMS algorithm, enables algorithm from the size of Row sum-equal matrix convergence coefficient, thus the direction of search of adjustment algorithm and convergence rate, to obtain optimum weight coefficient and better adaptivity.

Formula (19) is changed as shown below:

W(n+1)＝W(n)+2μ(n)X(n)e(n) (21)

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

$\mathrm{\μ}\left(n\right)=\left[\begin{array}{cccc}{\mathrm{\μ}}_0\left(n\right)& 0& ...& 0\\ 0& {\mathrm{\μ}}_1\left(n\right)& ...& 0\\ ...& ...& ...& ...\\ 0& 0& ...& {\mathrm{\μ}}_{L-1}\left(n\right)\end{array}\right]$

In order to ensure the convergence of adaptive algorithm, must by μ _l(n) (l=0,1 ..., L-1) span limit between a minimum value and a maximum value, namely must meet formula (9).Under this condition, its value just can be divided into three kinds of situations.If 1 x (n-1) e (n) is after continuous multiple moment, symbol does not change, so by μ _ln () increases α doubly; If 2 x (n-1) e (n) is after continuous multiple moment, symbol changes, so by μ _ln () reduces α doubly; Other situation is then constant.Following formula can be obtained:

α is greater than 1.

In order to verify the effect of adaptive filter algorithm in active noise reduction process, the present invention also tests existing algorithm.Using the input signal of the original noise of transformer as sef-adapting filter, as shown in Fig. 6 (a)-Fig. 6 (b), then by the output signal of sef-adapting filter and the superimposed analog error signal of the noise signal of transformer.This signal is the signal after noise reduction, and introduces sef-adapting filter as target function.Sef-adapting filter length is set to 256 rank, and convergence coefficient is 0.05.Fig. 7 (a)-Fig. 7 (c) sets forth the output signal of error signal and amplitude-frequency characteristic and sef-adapting filter.

For understanding the convergence rate of adaptive algorithm, the present invention's definition is as the amplitude h of certain moment error signal _twith stable state moment amplitude h _∞time within difference 5%, then think and reach stable state at this moment algorithm, that is:

$\frac{h_t-h_{\mathrm{\∞}}}{h_{\mathrm{\∞}}}\×100\%\≤5\%---\left(20\right)$

Comparison diagram 6 can be found out, error signal reaches stable state after 295ms, and amplitude significantly decreases.Contrast amplitude-frequency characteristic can be found out, the amplitude at 100Hz and frequency multiplication place thereof drops to less than 0.006 by more than 0.02, has obvious noise reduction, and especially 300Hz place declines the most obvious.Contrast secondary signal and primary signal can find, its phase place almost differs 180 °, and therefore after both superpositions, the amplitude of error signal can obviously reduce.

Also simulating, verifying is carried out to control method of the present invention, supposes convergence coefficient μ _l(l=0,1 ..., L-1) initial value gets 0.05, and the length of filter is set to 256 rank, and α value is 2, can obtain as Fig. 8 (a)-Fig. 8 (d) result.

As can be seen from Fig. 8 (a)-Fig. 8 (d), residual noise reaches stable state at about 300ms, and amplitude is finally converged in about 0.015, achieves good noise reduction; The amplitude of learning curve display residual noise have dropped 39dB; From amplitude-frequency characteristic, primary noise main frequency composition amplitude drops to less than 0.0016, and in especially former primary noise, 300Hz component declines the most obvious; The final value of weight coefficient fluctuates at zero crossings.

Adaptive algorithm after improvement can make algorithm accelerating ated test to a certain extent when ensureing steady-state error performance, and improves noise reduction.In sum, the adaptive algorithm after improvement to some extent solves the problem that traditional algorithm cannot take into account convergence rate and steady-state error performance, and choosing of convergence coefficient initial value is also convenient.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.

Claims (7)

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

Step one, transducer gathers primary noise as being delivered in adaptive controller with reference to signal x (n), and adaptive controller exports a control signal according to this reference signal and drives loud speaker to send secondary noise as secondary signal y (n);

Step 2, the secondary sound field that the noise that power transformer the sends primary sound field set up and the secondary noise that loud speaker sends are set up produces and superposes, and collects the acoustic pressure after superposition, and form error signal e (n) by error pick-up;

Step 3, after adaptive controller receives error signal e (n), phase place and the amplitude of secondary signal is adjusted according to target function J (n) the LMS algorithm that utilizes convergence coefficient to change preset, continue to error signal and meet target function J (n), reach stable state.

2. high-power transformer adaptive active method for noise reduction control as claimed in claim 1, is characterized in that, secondary signal y (n) output in the n-th moment represents with vector form and is specially:

y(n)＝X ^T(n)W＝W ^TX(n)

Wherein X (n)=[x (n), x (n-1) ..., x (n-L+1)] ^t, W=[w ₁, w ₂..., w _l] ^t, w _ln () is weight coefficient, L is the length of filter; The error signal e (n) in the n-th moment is:

e(n)＝d(n)-y(n)＝d(n)-W ^TX(n)

The mean square error between desired signal d (n) and secondary signal y (n) can be made minimum be target, obtain target function J (n) by desired signal d (n) and secondary signal y (n).

3. high-power transformer adaptive active method for noise reduction control as claimed in claim 2, is characterized in that,

Described target function J (n) is desired signal d (n) and the mean square error of secondary signal y (n), that is:

J(n)＝E[e ²(n)]＝E[(d(n)-W ^TX(n)) ²]

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

＝E[d ²(n)]+W ^TRW-2W ^TP

Wherein, R is the autocorrelation matrix of input signal x (n), R=E [X (n) X ^t(n)]; P is cross-correlation vector, P=E [d (n) X (n)], adopts iterative method to obtain optimum weight vector W during the LMS algorithm utilizing convergence coefficient to change ^*, get optimum weight vector W ^*time, target function is minimum.

4. high-power transformer adaptive active method for noise reduction control as claimed in claim 3, is characterized in that, adopts iterative method to obtain optimum weight vector W during the LMS algorithm utilizing convergence coefficient to change ^*the iteration function adopted in process is:

W(n+1)＝W(n)+2μ(n)X(n)e(n)

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

$\mathrm{\μ}\left(n\right)=\left[\begin{array}{cccc}{\mathrm{\μ}}_0\left(n\right)& 0& \mathrm{...}& 0\\ 0& {\mathrm{\μ}}_1\left(n\right)& \mathrm{...}& 0\\ \mathrm{...}& \mathrm{...}& \mathrm{...}& \mathrm{...}\\ 0& 0& \mathrm{...}& {\mathrm{\μ}}_{L-1}\left(n\right)\end{array}\right]$

μ _l(n) (l=0,1 ..., L-1) be convergence coefficient, concrete obtaining value method is:

Wherein α is greater than 1.

5. high-power transformer adaptive active method for noise reduction control as claimed in claim 1, it is characterized in that, described stable state is the amplitude h of error signal _twith stable state moment amplitude h _∞time within difference 5%.

6. one kind adopts the control system of high-power transformer adaptive active method for noise reduction control described in claims 1, it is characterized in that, comprise reference sensor, error pick-up and adaptive controller, described reference sensor connects the input of adaptive controller by the first preamplifier, described error pick-up connects the input of adaptive controller by the second preamplifier, and the output of described adaptive controller connects loud speaker by power amplifier.

7. adopt a control system for high-power transformer adaptive active method for noise reduction control described in claims 6, it is characterized in that, described reference sensor and error pick-up adopt condenser microphone.