CN111276117A - Active noise control method based on mixed frog-leaping algorithm - Google Patents

Abstract

The invention proposes an active noise control method based on a hybrid frog leaping algorithm, which organically combines a frog group-based collaborative search method with ANC, performs information transmission according to population classification, and has local search and global information mixed search. The strategy can obtain the global optimal solution in the search process, and can obtain better noise reduction effect compared with the existing FxLMS algorithm. In addition, the noise reduction scheme of the present invention does not require modeling of the secondary path, and can cope with the sudden change of the secondary path during the noise control process.

Description

An Active Noise Control Method Based on Hybrid Frog Leaping Algorithm

technical field

The invention relates to the field of active noise control, and proposes an active noise control method based on a hybrid frog leaping algorithm.

Background technique

Noise pollution is a particularly prominent problem in environmental pollution. There are usually two technical means for noise treatment, namely passive noise control and Active Noise Control (ANC) technology. The traditional noise control method belongs to passive noise control, and its main methods include sound absorption treatment, sound insulation treatment, the use of mufflers, vibration isolation, damping and vibration reduction. In general, the passive noise control method has better control effect on medium and high frequency noise, but has poor effect on low frequency noise. Unlike passive control, ANC technology can effectively control low frequency noise. The mechanism of ANC is based on the principle of destructive interference of sound waves. The primary sound source generates the desired signal, and the secondary sound source generates a secondary signal with the same amplitude and opposite phase as the desired signal. The two cancel each other out, thereby reducing noise.

The most commonly used algorithm in ANC is the FxLMS algorithm (Morgan D R. History, applications and subsequent development of the FxLMS algorithm. IEEE Signal Processing Magazine 2013, 30(3): 172-176). In the ANC system, the physical path from the secondary sound source to the error microphone is called the secondary path, and the links that make up the secondary path usually include the sound field, the frequency response of electro-acoustic devices and electronic circuits. In the traditional method, to complete an active control algorithm, an estimate of the transfer function of the secondary path must first be obtained. The FxLMS algorithm usually adopts offline modeling and estimation of secondary paths. At this time, if the secondary path changes, the performance of the ANC algorithm will be degraded. In practice, the secondary path often behaves as a time-varying system, so the noise reduction effect of the traditional FxLMS algorithm is poor. Moreover, FxLMS updates the weight coefficients according to the gradient algorithm, and there is a problem of falling into a local optimal solution.

SUMMARY OF THE INVENTION

In order to solve the problems existing in the prior art, the present invention proposes an ANC method based on the Shuffled Frog Leaping Algorithm (SFLA). SFLA is a collaborative search method based on frog populations. This method transmits information according to population classification. It has a search strategy that combines local search and global information, and can obtain the global optimal solution during the search process. Compared with the existing FxLMS algorithm Can get better noise reduction effect. And the SFLA-based noise reduction scheme does not require secondary path modeling, and can cope with the sudden change of the secondary path during the noise control process.

The technical scheme of the present invention is:

Described a kind of active noise control method based on hybrid frog leap algorithm, is characterized in that: comprises the following steps:

Step 1: The active controller of the active noise control system randomly generates N control filter weight coefficients, the weight coefficient of each control filter acts as a separate frog individual, and the weight coefficients of all N control filters form a frog group , N is the size of the group;

Step 2: The active controller of the active noise control system selects an individual from the current population and enters step 3;

Step 3: The primary sound source of the active noise control system emits noise, the reference sensor picks up the reference signal x(n) as the input of the control filter, and the desired signal d(n) is formed at the error sensor; in the active noise control In the continuous M sampling periods of the system, the active controller uses the individual entering this step as the weight coefficient of the control filter, the reference signal x(n) is filtered by the control filter to obtain the secondary signal y(n), and the secondary signal y(n) is obtained by filtering the reference signal x(n). The signal drives the secondary sound source to generate the cancellation signal s(n); at the error sensor, the desired signal d(n) and the cancellation signal s(n) are superimposed to generate the error signal e(n), e(n) is the active noise control The error signal obtained by the system in each sampling period; calculate the mean square value of the error signal obtained by the active noise control system in the M sampling periods

In the formula, J _i represents the mean square value of the error of the ith frog individual, and t _i represents the start sampling time of the frog individual as the control filter weight coefficient;

Step 4: When the M sampling periods corresponding to an individual are over, the active controller selects a new individual from the current population to enter Step 3. After M*N sampling periods, all N samples in the current population are obtained. The fitness corresponding to the individual frog; the smaller the error mean square value corresponding to the individual, the higher the fitness;

Step 5: Divide the individuals in the current population into m sub-populations, and execute the evolution process within each sub-population according to the individual fitness or the mean square value of the error. After all sub-populations complete an evolution, each sub-population is remixed to form a new generation. population, and then return to step 2; when the corresponding minimum error mean square value in the population converges to the set standard, the individual corresponding to the minimum error mean square value is obtained as the weight coefficient of the control filter corresponding to the current secondary path of the active noise control system .

Further, it also includes step 6: when it is detected that the secondary path of the active noise control system changes, return to step 1, and re-initialize the population; wherein the basis for judging the change of the secondary path is: the corresponding to the current generation population. The minimum error mean square value is greater than the minimum error mean square value corresponding to the previous generation population, and the difference is greater than the set value.

Further, in step 5, the individuals in the population are divided into m sub-populations as follows: the individuals in the population are sorted according to the mean square value of error or fitness, and then the individuals are grouped cyclically to generate sub-populations, each sub-population contains N /m individuals; the cyclic grouping refers to: the first individual after sorting enters the first sub-population, the second individual enters the second sub-population, ..., the m-th individual enters the m-th sub-population, The m+1th individual enters the first subpopulation again, and so on, until all individuals enter the subpopulation.

Further, the specific process of performing evolution within each subpopulation in step 5 is as follows:

Step 5.1: Set the maximum number of evolutions allowed in each subpopulation as Y, In represents the current number of evolutions, and the initial value is 0;

Step 5.2: In the In-th evolution, the individual W _b with the best fitness in the sub-population is used to guide the individual W _w with the worst fitness, and the moving distance of the individual W _w with the worst fitness is obtained:

D=Rand()×(W _b -W _w )

In the formula, Rand() represents a random number from 0 to 1, and the new individual obtained after the individual with the worst fitness moves is

W _w '=W _w +D

Step 5.3: If step 5.2 can generate an individual with better fitness, replace the individual W _w with the worst fitness with a new individual, and go to step 5.5, otherwise use W _g to replace W _b , repeat the process of step 5.2, and After the completion of step 5.2, go to step 5.4;

Step 5.4: If an individual with better fitness can be generated, replace the individual W _w with the worst fitness with a new individual, and go to step 5.5, otherwise, randomly generate a new individual to replace the individual W _w with the worst fitness;

Step 5.5: If the current evolution times In is less than the maximum evolution times Y allowed by the subpopulation, go back to step 5.2.

Further, in step 1, if the length of the control filter is L, then a single frog individual in the population is an L-dimensional variable, representing the current position of the individual.

Further, in step 5.2, Rand() is also an L-dimensional variable, and the random numbers of each dimension are independent of each other.

beneficial effect

(1) The SFLA-based ANC method of the present invention can realize global information exchange and local information depth search, avoid falling into local extreme points, so as to search for the global optimum point and obtain the best noise reduction effect.

(2) The iterative process of the weight coefficients of this method does not include filtering the -x signal, so no secondary path modeling is required. In the case of changes in the secondary path, the noise reduction performance of the system is not affected.

Additional aspects and advantages of the present invention will be set forth, in part, from the following description, and in part will be apparent from the following description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of embodiments taken in conjunction with the accompanying drawings, wherein:

Figure 1: Schematic diagram of ANC system based on SFLA;

Figure 2: Weight coefficient iteration flow chart;

Figure 3: Individual evolution steps in subpopulations;

Figure 4: Initial population after permutation;

Figure 5: Minimum mean square error convergence process;

Figure 6: Minimum mean square error convergence process for secondary path changes.

Detailed ways

The ANC system based on the SFLA algorithm is shown in Figure 1. In the figure, x(n) is the reference signal picked up by the reference sensor in the ANC system, which is the input of the control filter. d(n) is the desired signal generated by the noise source at the error sensor in the ANC system. y(n) is the secondary signal calculated by the control filter. After passing through the power amplifier, the secondary sound source is driven, and the cancellation signal s(n) is generated at the error sensor. The desired signal and the cancellation signal are superimposed to form an error signal e(n). P(z) and S(z) represent the primary and secondary pathways, respectively. w _g represents the weight coefficient value of the control filter of the current generation, which is expressed as the position of the frog population of the current generation in SFLA. It can be seen from the figure that the iteration of its weight coefficients does not require the filtering-x signal necessary in the FxLMS algorithm, so secondary path modeling is not required.

The SFLA-based algorithm of the present invention only uses one identical control filter weight coefficient within a certain period of time, the filter weight coefficient is updated generation by generation, and a new population is formed according to the adjustment strategy of the algorithm. The iterative steps of the weight coefficient based on SFLA are shown in Figure 2. In order to be able to adapt to the change of secondary path, no end criterion is set in the algorithm.

Specific steps are as follows:

(1) At the beginning of ANC, the active controller of the ANC system randomly generates N control filter weight coefficients, the weight coefficient of each control filter is called a separate frog individual, and the weight coefficients of all N control filters are is called a frog colony, and N is the size of the colony. The population in the controller is initialized as

w _g0 ={w ₁ ,w ₂ ,w ₃ ,...w _N } (1)

Assuming that the length of the control filter is L, a single frog individual in the population is an L-dimensional variable, representing the current position of the frog.

The primary sound source of the ANC system emits noise, the reference sensor picks up the reference signal x(n) as the input to the control filter, and the desired signal d(n) is formed at the error sensor. In the continuous M sampling period of the ANC system, the active controller selects an individual from the current population as the weight coefficient of the control filter. Then, the secondary signal y(n) is obtained by filtering the reference signal x(n) through the control filter, and the secondary signal drives the secondary sound source to generate the cancellation signal s(n). At the error sensor, the desired signal d(n) is superimposed with the cancellation signal s(n) to generate the error signal e(n), that is,

e(n)=d(n)+s(n) (2)

e(n) is the error signal obtained in each sampling period of the ANC system; for the error signal obtained by the ANC system in M sampling periods, the mean value of the square of the error signal is stored in the controller as the currently selected frog The evaluation standard of the fitness of an individual, the smaller the mean square value of the error corresponding to the individual is, the higher the fitness is. For example, the fitness can be taken as the reciprocal of the mean square value of the error. The specific formula is

In the formula, J _i represents the mean square value of the error of the ith frog individual, and t _i represents the start sampling time of the frog individual as the control filter weight coefficient.

After the M sampling periods, the active controller selects another frog individual in the group as the weight coefficient of the control filter. It should be noted that for different individuals, the corresponding sampling time length M should be the same. In this way, the fitness of different individuals can be correctly evaluated. In this way, after M*N sampling periods, the fitness of all N frog individuals in the group is obtained.

In each generation of the population, all N individuals are used as the weights of the control filter to generate secondary signals. In the respective M sampling periods, different individuals in the population are selected as the control filter weight coefficients, and the mean square value of the corresponding error signal is recorded, thereby obtaining the fitness values of N individuals. The fitness of the ith individual is used to measure the performance of the ith individual, that is, the smaller the mean square error value of the ith individual, the higher the fitness, and the better the individual position.

(2) Classify the entire population, for example, N individuals can be sorted in descending order according to their fitness, and then the frogs are grouped cyclically to generate sub-populations. Divide the population into m sub-populations: W ₁ , W ₂ ,...W _m , then each sub-population contains N/m frogs. After sorting, the first frog enters the first sub-population, the second frog enters the second sub-population, ..., the m-th frog enters the m-th sub-population, and the m+1-th frog enters the first sub-population again. subpopulations, and so on, until all individuals enter the subpopulation. The number m of subpopulations needs to be selected reasonably. If the value of m is too large, there will be fewer individuals in each sub-population, which will reduce the exchange of information within the sub-population, and the advantages of local search will be lost.

(3) After the population is grouped, the evolution process is performed within each sub-population, and the individual positions in the sub-population are improved through evolution. Set the maximum number of evolutions allowed in each subpopulation as Y, In represents the current number of evolutions, and the initial value is 0. In each subpopulation, the best-performing frog is denoted by W _b and the worst-performing frog is denoted by W _w . Denote the best performing frog in the entire population by W _g . In each evolution, the best performing frog W _b in the current subpopulation is used to guide the worst performing frog W _w . The individual evolution steps in the subpopulation are shown in Figure 3. The specific operations are as follows:

① Adjust the position of the worst performing frog. Use the best-performing frog W _b in the subpopulation to guide the worst-performing frog W _w . The distance the worst frog can move is

D=Rand()×(W _b -W _w ) (4)

where Rand() represents a random number from 0 to 1. The new position of the worst performing frog after moving is

W _w '=W _w +D (5)

②If step ① can produce a better solution (that is, use _Ww ' as the control filter weight coefficient, calculate the mean square value of the error in M sampling periods of the ANC system, if the mean square value of the error is better than the corresponding value of _Ww Error mean square value), then replace the worst frog W _w with the frog in the new position, go to step 4, otherwise use W _g to replace W _b , repeat the process of step ①, and go to step ③ after step ① is completed;

③ If a better solution can be generated, replace the frog W _w with the worst performance with the frog in the new position, and go to step ④, otherwise, randomly generate a new individual to replace the frog W _w with the worst performance;

④ If the current number of evolutions In is less than the maximum number of iterations allowed by the subpopulation, go back to step ①.

(4) After each subpopulation completes one evolution, the subpopulations are remixed to form a new generation of populations, and then step (2) is started again.

Under the condition that the secondary path remains unchanged, the frog population gradually converges to the best position, that is, the minimum mean square error is obtained, and the best noise reduction effect will be obtained at this time. If the secondary pathway changes suddenly, the population cannot adapt to the change and remains at a local minimum because after convergence all individuals have the same value and randomness is lost. Therefore, if a change in the secondary pathway is detected, the population needs to be re-initialized. The basis for judging the change of the secondary pathway is: if the fitness value of the best frog W _g in the entire population of the current generation is less than the best fitness value of the previous generation, that is, the minimum mean square error of the current generation is greater than the minimum value of the previous generation. If the mean square error is greater than the set value, it means that the current system has changed. At this time, the controller judges that the secondary path has changed, then reinitializes the population and starts the evolution process.

The embodiments of the present invention are described in detail below, and the embodiments are exemplary and intended to explain the present invention, but should not be construed as a limitation of the present invention.

(1) The primary path used in the computer simulation is P(z)=z ^-6 +0.5z ^-7 -0.3z ^-8 +0.6z ^-9 , and the secondary path is S(z)=z ^-3 +1.5 z ^-4 -0.5z ^-5 . Primary noise is white noise with zero mean. Set the length of the control filter to L=20, and the frog individual is a 20-dimensional variable. The selection of SFLA parameters is: frog group size N = 40, the number of sub-populations m = 4, and the maximum allowed evolution times in the sub-population Y = 10.

(2) The active controller randomly generates 40 groups of control filter weight coefficients, and the length of each group is 20. In the current sampling set consisting of M consecutive sampling periods, the controller selects an individual as the control filter. The reference signal x(n) is used as the input of the controller, then the secondary signal y(n) is obtained by filtering the reference signal x(n) through the control filter. At the error sensor, the expected signal d(n) and the cancellation signal s (n) The superposition produces an error signal e(n). Set the sampling length M=100 for evaluating individual fitness. Over this sample length, store the mean of the squared error signal in the controller

All individuals are used as weights of the adaptive filter to generate secondary signals. In different sampling sets, different individuals in the population are selected as filter weight coefficients and the mean square value of the corresponding error signal is recorded, thereby obtaining N mean square errors and evaluating the fitness values of N individuals.

(3) Classify the entire population, and arrange the individuals in descending order according to their fitness. The sorted population is shown in Figure 4. The frogs are then grouped cyclically to generate 4 subpopulations. After sorting, the first frog enters the first sub-population, the second frog enters the second sub-population, ..., the 40th frog enters the fourth sub-population.

(4) After the sub-population is allocated, the evolution process is carried out in each sub-population in turn, and the position of the frog with the worst performance is adjusted. In each evolution, the frog W _b with the best performance in the current sub-population is used to guide the frog with the worst performance. Frog W _w . In the process of local search, taking the same Rand value for each dimension of the variable will limit the randomness of each variable changing in the optimal direction, so the step moving distance formula of the low-fitness frog is improved. Randomly generate different Rand values for each dimension variable, then the moving distance of the worst frog is

In the formula, the superscript j represents the jth dimension component of the parameter.

The maximum number of evolutions allowed within each subpopulation is 10. After the evolution of the four subpopulations is completed, the subpopulations are remixed to form a new generation of populations. The process of group evolution is then repeated for a new generation of populations.

(5) Under the condition that the secondary pathway remains unchanged, the population evolves steadily, and all individuals gradually converge to the best position. The convergence process of the minimum mean square error of each generation is shown in Figure 5. It can be seen from the figure that in the initial stage of the algorithm, the convergence speed is very fast, and the minimum mean square error in the population decreases rapidly. After 50 generations of the population, the convergence process is gradually stable. At this time, the frogs in the population are relatively close, and gradually approach the optimal solution.

(6) The SFLA-based ANC method does not require secondary pathway modeling, so its performance is immune to changes in secondary pathways. In order to verify this feature, when the population evolves to 100 generations, the secondary pathway of the system is changed from the initial secondary pathway to S(z)=z ^-2 +1.5z ^-3 -z ^-4 . At this time, the minimum mean square error of the generation will increase instantaneously. When the controller detects that the current secondary path has changed, it re-initializes the population. The convergence process of the minimum mean square error is shown in Figure 6. It can be seen from the figure that the reinitialized population can converge to the optimal solution. Then the ANC method based on SFLA can adapt to the change of the secondary path of the system and ensure the noise reduction effect.

Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and those of ordinary skill in the art will not depart from the principles and spirit of the present invention Variations, modifications, substitutions, and alterations to the above-described embodiments are possible within the scope of the present invention without departing from the scope of the present invention.

Claims (6)

1. An active noise control method based on a mixed frog-leaping algorithm is characterized in that: the method comprises the following steps:

step 1: an active controller of the active noise control system randomly generates N control filter weight coefficients, the weight coefficient of each control filter is used as a single frog individual, the weight coefficients of all N control filters form a frog group, and N is the scale of the group;

step 2: an active controller of the active noise control system selects a certain individual from the current population and enters step 3;

and step 3: the primary sound source of the active noise control system emits noise, the reference sensor picks up the reference signal x (n) as the input of the control filter, and simultaneously forms the expected signal d (n) at the error sensor; in continuous M sampling periods of an active noise control system, an active controller takes an individual entering the active noise control system as a weight coefficient of a control filter, a reference signal x (n) is filtered by the control filter to obtain a secondary signal y (n), and the secondary signal drives a secondary sound source to generate a cancellation signal s (n); at the error sensor, the expected signal d (n) is superposed with the offset signal s (n) to generate an error signal e (n), wherein e (n) is the error signal obtained by the active noise control system in each sampling period; calculating the mean square value of error signals obtained by the active noise control system in the M sampling periods

In the formula J_iRepresents the mean square error value, t, of the ith frog_iA start sampling time representing the frog individual as a control filter weight coefficient;

and 4, step 4: after M sampling periods corresponding to one individual are finished, the active controller selects a new individual from the current population again and enters step 3, and after M × N sampling periods, the fitness corresponding to all N frog individuals in the current population is obtained; wherein, the smaller the mean square error value corresponding to the individual is, the higher the fitness is;

and 5: dividing individuals in the current population into m sub-populations, executing an evolution process in each sub-population according to individual fitness or an error mean square value, after all the sub-populations finish one-time evolution, remixing each sub-population to form a new generation of population, and then returning to the step 2; and when the corresponding minimum mean square error value in the population converges to a set standard, obtaining an individual corresponding to the minimum mean square error value as a weight coefficient of a corresponding control filter of the current secondary channel of the active noise control system.

2. The active noise control method based on the mixed frog-leaping algorithm according to claim 1, characterized in that: further comprising the step 6: when detecting that the secondary path of the active noise control system changes, returning to the step 1, and initializing the population again; the basis for judging the change of the secondary path is as follows: the minimum error mean square value corresponding to the current generation population is larger than the minimum error mean square value corresponding to the previous generation population, and the difference value is larger than a set value.

3. The active noise control method based on the mixed frog-leaping algorithm according to the claim 1 or 2, characterized in that: in step 5, the manner of dividing the individuals in the population into m sub-populations is as follows: sequencing the individuals in the population according to the mean square error value or fitness, and then circularly grouping the individuals to generate sub-populations, wherein each sub-population comprises N/m individuals; the cyclic grouping refers to: after sorting, the 1 st individual enters a first sub-population, the 2 nd individual enters a second sub-population, …, the mth individual enters a mth sub-population, the m +1 th individual enters the first sub-population, and so on until all individuals enter the sub-populations.

4. The active noise control method based on the mixed frog-leaping algorithm according to claim 3, characterized in that: the specific process of performing evolution inside each sub-population in step 5 is as follows:

step 5.1: setting the allowed maximum evolutionary times In each sub-population as Y, wherein In represents the current evolutionary times and the initial value is 0;

step 5.2: in the In-evolution, the individual W with the best fitness within the sub-population was used_bDirecting the individuals W with the worst fitness_wTo obtain the individual W with the worst fitness_wHas a moving distance of

D＝Rand()×(W_b-W_w)

In the formula, Rand () represents a random number from 0 to 1, and a new individual obtained after the individual with the worst fitness moves is

W'_w＝W_w+D

Step 5.3: if step 5.2 is able to produce a better fitness individual, the newly-fit individual is substituted for the least fitness individual W_wAnd go to step 5.5, otherwise utilize W_gInstead of W_bRepeating the process of step 5.2, and entering step 5.4 after step 5.2 is completed;

step 5.4: if a more well-adapted individual can be generated, the newly adapted individual W is replaced with the less well-adapted individual W_wAnd step 5.5 is carried out, otherwise, a new individual is randomly generated to replace the individual W with the worst fitness_w；

Step 5.5: and if the current evolution number In is less than the maximum evolution number Y allowed by the sub-population, returning to the step 5.2.

5. The active noise control method based on the mixed frog-leaping algorithm according to claim 4, characterized in that: in step 1, if the length of the control filter is L, a single frog individual in the population is an L-dimensional variable, which represents the current position of the individual.

6. The active noise control method based on the mixed frog-leaping algorithm according to claim 5, characterized in that: in step 5.2, Rand () is also an L-dimensional variable, and random numbers of each dimension are independent of each other.

Images