CN114065630A - Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm - Google Patents

Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm Download PDF

Info

Publication number
CN114065630A
CN114065630A CN202111364883.0A CN202111364883A CN114065630A CN 114065630 A CN114065630 A CN 114065630A CN 202111364883 A CN202111364883 A CN 202111364883A CN 114065630 A CN114065630 A CN 114065630A
Authority
CN
China
Prior art keywords
sound source
transfer function
genetic algorithm
channel transfer
source power
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111364883.0A
Other languages
Chinese (zh)
Inventor
孙超
张少东
谢磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202111364883.0A priority Critical patent/CN114065630A/en
Publication of CN114065630A publication Critical patent/CN114065630A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/06—Power analysis or power optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/10—Noise analysis or noise optimisation

Abstract

The invention relates to an uncertain parameter focus matching field sound source power estimation method based on a genetic algorithm, which expands a scanning range to all uncertain parameters including environmental parameters and sound source positions, uses an alternative channel transfer function and an included angle cosine value of a received signal frequency domain snapshot as a target function, optimizes the target function by combining the genetic algorithm, and estimates the sound source power by using a channel transfer function output by the genetic algorithm. The FMFPE method can reduce the included angle between the channel transfer function estimated value and the ideal value and enhance the correlation between the channel transfer function estimated value and the ideal value, so that the environment mismatch robustness of the sound source power estimation method is improved. The basic principle and the implementation scheme of the invention are verified by computer numerical simulation, and the result shows that: the FMFPE method effectively reduces the estimation errors of the sound source power of different frequencies; more accurate power estimation results can also be obtained using the FMFPE method when the sound source is at different locations.

Description

Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm
Technical Field
The invention belongs to sound source power estimation, and relates to an uncertain parameter focus matching field sound source power estimation method based on a genetic algorithm.
Background
A matching Field sound source Power Estimation Method (MFPE) describes a shallow sea waveguide environment according to a sound propagation model, estimates a sound source position by a positioning method, estimates a channel transfer function by using environment parameters and sound source position parameters, and obtains the most objective and real sound source radiation Power by using all energy radiated by a sound source (Xiaoing L F, Sun C. an Estimation method of a ship radial noise level based on a Matched Field processing industry, 2014; 39(5):570 and 576.). However, MFPE requires that the environmental parameters and the sound source location parameters are accurately known, and when the environmental parameters are mismatched, the channel transfer function estimation result has a bias, resulting in a large error of the sound source power estimation result.
A method for improving the mismatch robustness of the MFPE environment is not proposed yet, and common methods for improving the mismatch robustness of the matching field localization environment include: 1) the sector focusing method is characterized in that a projection matrix is constructed by extracting main eigenvalues in a received data cross-spectral density matrix, and parts sensitive to environmental mismatch in the cross-spectral density matrix are removed, so that the robustness of a positioning result is improved, but the signal energy cannot be ensured not to be lost; 2) the Minimax method and the influence of the environmental mismatch are all represented by the change of the channel transfer function, and the Minimax method reconstructs the channel transfer function by utilizing the information of the included angle and the upper and lower bounds of the amplitude change range of the channel transfer function so as to improve the robustness of the environmental mismatch. However, the improvement of the robustness of the environmental mismatch of the method is represented by the fact that the worst positioning performance is optimal when the environment is mismatched, and the robustness of the optimal positioning result cannot be ensured.
Under the framework of the requirement of not losing signal energy, optimization of a channel transfer function is the key for improving the robustness of environment mismatch, and the search of the channel transfer function with smaller deviation under the existing prior information is an optimization problem essentially. In fact, only using a priori information of the variation range of the sound source position parameter to optimize the channel transfer function, the improvement of the robustness of the environmental mismatch is limited because the uncertainty of the environmental parameter is ignored. Collins et al introduce a priori information of the environmental parameter variation range to provide an environmental focus positioning method. The method expands a search space, and searches a channel transfer function with small deviation in a sound source position variation range and an uncertain environment parameter variation range, so that the environment mismatch robustness of the positioning method is improved. But this method is now only used for sound source localization and has not been used to solve the sound source power estimation problem. Because the uncertain environment parameters are usually more, the number of the uncertain parameters to be searched is exponentially increased, and the method can be combined with global optimization algorithms such as genetic algorithm and the like to solve the problem.
Disclosure of Invention
Technical problem to be solved
In order to avoid the defects of the prior art, the invention provides an uncertain parameter focus matching Field sound source Power Estimation method (FMFPE) based on a genetic algorithm. The method comprises the steps of obtaining a channel transfer function search range by utilizing a sound source position variation range and an uncertain environment parameter variation range, constructing a target function as an included angle cosine value of an estimated channel transfer function and a received data frequency domain snapshot, finishing an optimization process of the channel transfer function by combining a Genetic Algorithm (GA), and obtaining a sound source power estimation value with high environment mismatch robustness by using the obtained channel transfer function estimation value.
Technical scheme
An uncertain parameter focus matching field sound source power estimation method based on a genetic algorithm is characterized by comprising the following steps:
step 1: carrying out FFT processing on the uniform vertical linear array collected signals to obtain a received signal frequency domain model:
X＝Ha+N
where X is the received signal frequency domain snapshot, H is the ideal channel transfer function, N is the environmental noise, and the noise power is
Figure BDA0003360509940000021
Step 2: constructing an objective function
Figure BDA0003360509940000022
Wherein: hiA transfer function of an alternative channel corresponding to the ith grid point;
step 3, performing HiH (maximum value) of search target function by using genetic algorithmGA：
The specific search process is as follows:
the first step is as follows: initialization, including setting of crossover probability PcAnd the mutation probability PmGenerating a primary population and setting a cycle termination condition;
cross probability PcSelecting the range of 0.40-0.99; probability of variation PmSelecting the range of 0.001-0.1;
initializing a population: dividing grids for the uncertain parameter variation range by adopting binary coding; each uncertain parameter is represented by K-bit binary number, and the variation range of each uncertain parameter is divided into 2K-1 grid, the binary number corresponding to the value of the uncertainty parameter Q being:
Figure BDA0003360509940000031
in the formula, LB is the lower bound of the uncertain parameter variation range, UB is the upper bound of the uncertain parameter variation range, and u is a decimal number corresponding to binary number; considering J uncertain parameters, the total number is 2KJEach grid point, the corresponding uncertain parameter group of each grid point is represented by a J multiplied by K bit binary number; randomly selecting Pop grid points, corresponding to Pop groups of uncertain parameters, substituting the group of uncertain parameters into an alternative channel transfer function H obtained by sound field softwareiThe individual is an individual, and Pop individuals form an initial population;
setting a circulation termination condition: setting a genetic algebra counter n as 0, selecting a determined genetic algebra Gen as an iteration termination condition, wherein the number of grid points required to be calculated in the genetic algorithm searching process is as follows:
Num＝Pop×Gen
the second step is that: individual evaluation, individual fitness of genetic algorithm is determined by objective function
Figure BDA0003360509940000032
Calculating to obtain;
the third step: population evolution
The selection process comprises the following steps: the selection process of 'winning or losing' is completed by adopting two methods of roulette and elite selection;
and (3) a crossing process: randomly distributing the 'offspring' individuals obtained in the selection process, dividing every two individuals into a group, and sequentially judging each group of individuals as follows: generating a random number epsilon within the range of 0-11If epsilon1＜PcIf not, the cross operation is not executed on the group of individuals; the crossing process adopts single-point crossing, a point is randomly generated in an interval [1, J multiplied by K ] to serve as a crossing point, the crossing point divides each individual binary string into two sections, one section of binary string is exchanged to generate a new binary string, and the crossing process is finished;
and (3) mutation process: the following determinations are performed for each individual in turn: generating a random number epsilon within the range of 0-12If epsilon2＜PmIf not, the individual is not subjected to mutation operation; the mutation process adopts single point mutation in the interval of [1, J × K]Randomly generating a point as a variation point, and reversing the binary number of the variation point to complete the variation process;
the fourth step: judging the loop termination, terminating iteration when n is larger than Gen, and outputting the individual with the maximum fitness when iteration is terminated; otherwise, making n equal to n +1 and returning to the second step;
and taking the absolute value of the cosine value of the included angle between the ideal value H of the channel transfer function and the receiving signal frequency domain snapshot as a reference value for comparison, wherein the expression of the reference value REF is as follows:
Figure BDA0003360509940000041
after multiple iterations, the fitness of the genetic algorithm is maximum and is close to a reference value REF;
will be inheritedWhen the algorithm terminates iteration, the channel transfer function corresponding to the individual with the highest fitness is recorded as hGA；
And 4, step 4: the sound source power estimated value obtained by the FMFPE method is as follows:
Figure BDA0003360509940000042
the roulette selection is the random selection of Pop individuals from the "parent" individuals.
The elite selection is directly reserved as the optimal individual of the "parent", and the roulette selection only produces Pop-1 "offspring" individuals.
Advantageous effects
The invention provides an uncertain parameter focus matching field sound source power estimation method based on a genetic algorithm, which expands a scanning range to all uncertain parameters including environmental parameters and sound source positions, uses an alternative channel transfer function and an included angle cosine value of a received signal frequency domain snapshot as a target function, optimizes the target function by combining the genetic algorithm, and estimates the sound source power by using the channel transfer function output by the genetic algorithm. The FMFPE method can reduce the included angle between the channel transfer function estimated value and the ideal value and enhance the correlation between the channel transfer function estimated value and the ideal value, so that the environment mismatch robustness of the sound source power estimation method is improved.
In the present invention: the FMFPE method is presented. And expanding a search range, acquiring a channel transfer function search range by utilizing the sound source position change range and the uncertain environment parameter change range, and taking the channel transfer function search range as a search domain of the channel transfer function estimation problem. And constructing an objective function as an included angle cosine value of the estimated channel transfer function and the received data frequency domain snapshot, finishing the optimizing process of the channel transfer function by combining a genetic algorithm, and estimating the sound source power by using the channel transfer function obtained by optimizing to obtain higher environment mismatch robustness.
The basic principle and the implementation scheme of the invention are verified by computer numerical simulation, and the channel transfer function and the genetic algorithm searching process obtained by the method provided by the invention are provided. The results show that: the FMFPE method effectively reduces the estimation errors of the sound source power of different frequencies; more accurate power estimation results can also be obtained using the FMFPE method when the sound source is at different locations. The robustness of the environmental mismatch of the FMFPE method proposed by the invention in estimating the sound source power is tested by estimating the sound source power under different frequencies and sound source position conditions.
Drawings
FIG. 1 is a schematic diagram of a single-point crossover in a genetic algorithm search;
FIG. 2 is a schematic diagram of single point mutation in a genetic algorithm search;
FIG. 3 is a main flow of the steps involved in the present invention;
FIG. 4 is a flow chart of the present invention for searching a channel transfer function using a genetic algorithm;
FIG. 5 shows a sound velocity profile of a Benchmark standard shallow sea environment in an embodiment;
in the embodiment of fig. 6, the effect of environmental uncertainty on the modulus and angle of the channel transfer function.
In the embodiment of fig. 7, the ROM and cos θ of the channel transfer function searched by the genetic algorithm vary with SNR.
In the embodiment of FIG. 8, f is 10dB SNRaveAnd fmaxThe number of generations varies with the heredity.
In the embodiment of FIG. 9, f is the SNR of 30dBaveAnd fmaxThe number of generations varies with the heredity.
Fig. 10 shows the estimation results of the sound source power of the FMFPE method and the MFPE method at different SNRs in the embodiment.
Fig. 11 shows the estimation results of the sound source power of the FMFPE method and the MFPE method at different frequencies in the embodiment.
Fig. 12 shows the estimation results of the sound source power of the FMFPE method and the MFPE method when the sound source is at different depths in the embodiment.
Fig. 13 shows the estimation result of the sound source power of the FMFPE method and the MFPE method when the distance from the sound source to the receiving matrix is changed in the embodiment.
Detailed Description
The invention will now be further described with reference to the following examples and drawings:
the technical scheme adopted by the invention for solving the existing problems can be divided into the following 2 steps:
1. modeling the channel transfer function optimization problem. Obtaining a channel transfer function search range by utilizing the uncertain parameter variation range, and taking the channel transfer function search range as a search domain of a channel transfer function problem; the contribution of the present invention is constructed as an objective function for the sound source power estimation problem.
2. Solving the optimization problem of the channel transfer function by combining a genetic algorithm to obtain an estimation result of the channel transfer function, and obtaining an estimation value of the sound source power of the FMFPE by using the estimation value of the channel transfer function, namely the contribution of the invention.
The specific content related to the step 1) is as follows:
calculating a channel transfer function by combining a sound field model, wherein sound source position parameters and environment parameters are required to be used, and when the environment is uncertain, the channel transfer function at the position of a real sound source is used, so that deviation is bound to exist; the channel transfer function at the position of the sound source is estimated to have deviation because the scanning area is only the parameter variation range of the sound source position, and real parameters do not exist in the searching area when scanning is carried out under the mismatch environment; the search results can only obtain the channel transfer function with the smaller deviation in a series of mismatched transfer functions. In order to further reduce the deviation of the channel transfer function obtained by searching, the search domain is considered to be expanded, and the scanning is simultaneously carried out in the variation range of the environmental parameters and the variation range of the sound source position parameters. At this time, there are real parameters in the search domain, but due to the limitation of the search precision, the search result is usually an approximate solution of the optimal solution of the channel transfer function theory.
Assuming that the signals are collected using a uniform vertical linear array, the received signal frequency domain model can be expressed as:
X＝Ha+N (1)
where X is the received signal frequency domain snapshot, H is the ideal channel transfer function, N is the environmental noise, and the noise power is
Figure BDA0003360509940000071
Definition matrixThe column received Signal-to-Noise Ratio (Signal to Noise Ratio: SNR) is:
Figure BDA0003360509940000072
in order to search for a channel transfer function with a small deviation, it is first necessary to measure the deviation of the channel transfer function. The channel transfer function is a vector whose deviation can be described by the module deviation and the angle. For channel transfer function estimation
Figure BDA0003360509940000073
The deviation of the modulus of the channel transfer function can be expressed by
Figure BDA0003360509940000074
And the ratio of the modulus values of H describe:
Figure BDA0003360509940000075
in the formula, the symbol '| | |' represents a two-norm. The angle between the channel transfer functions may be determined by
Figure BDA0003360509940000076
And the cosine of the angle of H indicates:
Figure BDA0003360509940000077
it can be easily found by observing the formulas (3) and (4): ROM and cos theta are different only in denominator, and compared with numerator, an included angle between channel transfer functions is more sensitive to environment mismatch; therefore, consider using the angle cosine value as an objective function for the optimization problem. In practice, the ideal channel transfer function H cannot be obtained. Due to the fact that the formula (4) is middle
Figure BDA0003360509940000079
And H are both subjected to module value normalization processing, and cos theta does not contain module value information; when the signal-to-noise ratio is larger, the method can ensure thatThe ideal channel transfer function H is replaced by a received signal frequency domain snapshot X.
Usually, a parameter variation range is divided into grids, and uncertain parameter sets corresponding to the grid points are searched in sequence. The channel transfer function has a one-to-one correspondence with the uncertain parameter sets corresponding to the grid points, and the alternative channel transfer function corresponding to the ith grid point is recorded as Hi. The objective function for constructing the FMFPE method is:
Figure BDA0003360509940000078
(5) formula is essentially HiAnd X, within the variation range of the sound source position parameter and the environmental parameter, making f (H)i) The channel transfer function corresponding to the largest parameter set is the estimated channel transfer function with less deviation from the ideal channel transfer function. Due to the presence of environmental noise and the fact that the target function does not take the modulus into account, the search results can only yield a satisfactory solution that deviates from the ideal channel transfer function within an acceptable error range.
The specific content related to the step 2) is as follows:
to obtain a more accurate channel transfer function estimate, it is generally desirable that the grid spacing be small; because uncertain parameters are usually more, the number of grid points to be searched increases exponentially, and the calculation amount requirement of traversal search is too large. The channel transfer function estimation problem in the FMFPE method is essentially an optimization problem, and in the case that the search domain and the objective function are known, an intelligent global optimization algorithm such as a genetic algorithm can be used to complete the search process. The genetic algorithm regards an iterative process as a genetic evolution process, and the basic idea is to simulate a natural genetic mechanism and a biological evolution theory. The unit of genetic algorithm iteration is a 'population', a plurality of 'individuals' are arranged in the population, an uncertain parameter group corresponding to each individual is represented by 'genes', and the competitive advantage of the individual in the population is represented by 'fitness'; the higher the fitness is, the higher the competitive advantage of the individual is; the fitness is usually an objective function of the optimization problem, i.e. calculated by equation (5). And when iteration is terminated, outputting an uncertain parameter group and a channel transfer function corresponding to the individual with the maximum fitness in the population, wherein the deviation of the channel transfer function and an ideal channel transfer function is small, and the environment mismatch robustness can be improved by estimating the sound source power.
The specific search process is as follows:
the first step is as follows: initialization, including setting of crossover probability PcAnd the mutation probability PmGenerating a primary population and setting a cycle termination condition.
Setting the crossover probability PcAnd the mutation probability Pm。PcAnd PmThe selection standard and the recommendation range which respectively influence the searching capability and the ability of jumping out of the extreme point of the genetic algorithm and the cross probability and the variation probability are common are that the cross probability PcThe selection range is larger, and is 0.40-0.99; probability of variation PmAnd the selection range is smaller, and is 0.001-0.1.
And initializing the population. Dividing grids for the uncertain parameter variation range by adopting binary coding; each uncertain parameter is represented by K-bit binary number, and the variation range of each uncertain parameter is divided into 2K-1 grid, the binary number corresponding to the value of the uncertainty parameter Q being:
Figure BDA0003360509940000081
in the formula, LB is the lower bound of the uncertain parameter variation range, UB is the upper bound of the uncertain parameter variation range, and u is the decimal number corresponding to binary number. Considering J uncertain parameters, the total number is 2KJAnd each grid point corresponds to an uncertain parameter group which can be represented by a J multiplied by K bit binary number. Randomly selecting Pop grid points, corresponding to Pop groups of uncertain parameters, substituting the group of uncertain parameters into an alternative channel transfer function H obtained by sound field softwareiI.e. an "individual", the Pop individuals constitute the initial population.
A cycle end condition is set. Setting a genetic generation number counter n as 0, generally selecting a determined genetic generation number Gen as an iteration termination condition, wherein the more uncertain parameters which have great influence on a channel transfer function, the larger the genetic generation number Gen. The number of grid points required to be calculated in the genetic algorithm searching process is as follows:
Num＝Pop×Gen (7)
comparing Num and 2KJTherefore, the genetic algorithm can obviously reduce the number of grid points needing to be searched and quickly converge to an optimal solution.
The second step is that: and (4) evaluating individuals. The individual fitness of the genetic algorithm is calculated by the formula (5).
The third step: and (4) population evolution, including processes of selection, crossing, mutation and the like.
And (6) selecting the process. The selection process of 'winning or losing' is completed by adopting two methods of roulette and elite selection. The roulette selection is to randomly select Pop individuals from parent individuals, and the probability of each individual being selected is different according to different individual fitness; the probability of individual selection is increased along with the increase of individual fitness, so that the individual competition advantage with high fitness is ensured to be larger, and the 'winning or the elimination' is realized. Because the roulette randomly selects individuals according to probability, the possibility of being eliminated exists although the individual with the highest fitness is selected with the highest probability. Elite selection is the optimal individual of the 'parent' and is directly reserved, the roulette selection only generates Pop-1 'child' individuals, and the Elite selection can be adopted to avoid the optimal individual of the 'parent' from being eliminated.
And (5) a crossing process. The crossover process is the primary method of generating new individuals. Randomly distributing the 'offspring' individuals obtained in the selection process, dividing every two individuals into a group, and sequentially judging each group of individuals as follows: generating a random number epsilon within the range of 0-11If epsilon1＜PcPerforming the crossover operation on the group of individuals, otherwise not performing the crossover operation on the group of individuals. The crossing process adopts single-point crossing, and a single-point crossing schematic diagram is given in fig. 1. Randomly generating a point in the interval [1, J multiplied by K) as an intersection point, dividing each individual binary string into two sections by the intersection point, exchanging one section of the binary string to generate a new binary string, and finishing the intersection process.
And (5) performing mutation process. The mutation process is to generate newThe individual auxiliary method determines the local searching capability of the genetic algorithm. The following determinations are performed for each individual in turn: generating a random number epsilon within the range of 0-12If epsilon2＜PmIf not, the mutation operation is not executed to the individual. The mutation process adopts single point mutation, and fig. 2 shows a schematic diagram of single point mutation. In the interval [1, JXK]And randomly generating a point as a variation point, and reversing the binary number of the variation point to complete the variation process.
The fourth step: and judging the loop termination. When n is larger than Gen, stopping iteration, and outputting the individual with the maximum fitness when iteration is stopped; otherwise, let n be n +1 and return to the second step.
And taking the absolute value of the cosine value of the included angle between the ideal value H of the channel transfer function and the receiving signal frequency domain snapshot as a reference value for comparison, wherein the expression of the reference value REF is as follows:
Figure BDA0003360509940000101
over a number of iterations, the fitness maximum of the genetic algorithm is very close to the reference value REF.
Recording a channel transfer function corresponding to an individual with the highest fitness when the genetic algorithm terminates iteration as HGAAt any frequency point, the sound source power estimated value obtained by the FMFPE method is as follows:
Figure BDA0003360509940000102
the main flow of the present invention is shown in FIG. 3, and the flow of the genetic algorithm is shown in FIG. 4.
Examples of the embodiments
The embodiment of the invention is given by taking a Benchmark standard shallow sea environment sound velocity profile provided by NRL works hop'93 as an example. The implementation example utilizes a computer to carry out numerical simulation, and Krakenc sound field software based on a normal wave method is adopted for sound field calculation, so that the effect of the method provided by the invention is tested.
1) Uncertain environmental parameters in shallow sea
The sound velocity profile of the Benchmark shallow sea environment is shown in fig. 5, the nominal value and the variation range of the uncertain environment parameters are given in the graph, and it is assumed that the uncertain environment parameters are uniformly distributed in the variation range. In the figure, a black solid line represents a nominal value of a sound velocity profile, a linear negative gradient sound velocity profile is formed in seawater, a linear positive gradient sound velocity profile is formed in a sedimentary layer, and the sound velocity of a basement semi-space is consistent with the sound velocity of the lower surface of the sedimentary layer; the dashed lines on the left and right represent the lower and upper bounds of the sound velocity profile, respectively. The absorption coefficient alpha and the density rho of the substrate half space and the deposition layer are kept consistent, and the thickness of the deposition layer is kept unchanged all the time. The depth of the research sea area is 1 m-100 m, and the distance is 1 m-1000 m.
2) Transmitting and receiving transducer parameters
The data was received using a 100-element uniform vertical linear array with an array element spacing of 1m, with the array element depth closest to the sea surface being 1 m. The FMFPE method measures the sound source power of different frequency points one by one, and assumes that the nominal value of the signal power at any frequency point is 100dB, the nominal value of the depth of a sound source to be measured is 50m, and the nominal value of the distance from the sound source to be measured to a receiving array is 500 m. Most of the sound source radiation signals are broadband signals, the frequency change range of the signals is assumed to be 100 Hz-1000 Hz, and the Monte Carlo times adopted by simulation analysis are 1000 times.
3) Simulation analysis of influence of environment uncertainty on channel transfer function module value, included angle and sound source power estimation performance
The signal frequency is 500Hz, assuming that the depth of the sound source to be measured and the distance from the sound source to be measured to the receiving array are both nominal values. Randomly generating 1000 groups of environment parameters within the uncertain environment parameter variation range, and obtaining 1000 estimated channel transfer functions by using a Krakenc normal wave sound field model; and substituting the nominal values of the environmental parameters into a normal wave sound field model to obtain an ideal channel transfer function, and obtaining the ROM and cos theta according to the formulas (3) and (4), as shown in FIG. 6. The ideal values of ROM and cos theta are both 1, and in the result of 1000 Monte Carlo simulations, the mean value of ROM is 0.9923, and the variance is 0.0029; the mean value of cos θ is 0.5905, and the variance is 0.0764; it shows that environmental mismatch has less effect on ROM and more effect on cos θ. The mean value of the sound source power estimation result of the MFPE method is 94.3009dB, the error is 5.6991dB, the variance is 22.6555, and the influence of the visible environment on the sound source power estimation performance is great.
4) Simulation analysis of search performance of genetic algorithms
The number of the considered uncertain parameters is 9, and the considered uncertain parameters comprise 7 uncertain environment parameters and 2 sound source position parameters, wherein the sound source position parameters are the depth of the sound source to be detected and the distance from the sound source to be detected to the receiving matrix respectively. Taking the cross probability as 0.8 and the mutation probability as 0.05; the 9 uncertain parameters are coded by 12-bit binary number, and each uncertain parameter is divided into 212-1 ═ 4095 grids; if the number of individuals Pop is 100 and the generation number Gen is 200, a genetic algorithm needs to calculate 100 × 200 to 2 × 104A grid of points. Under the same conditions, using a traversal search requires calculation 29×12≈3.25×1032And the genetic algorithm greatly reduces the number of grid points to be searched and reduces the calculation amount.
Assuming a signal frequency of 500Hz, white ambient noise, and a noise level of 52dB, the ROM and cos θ of the channel transfer function searched by the genetic algorithm are given as a function of SNR in fig. 7. When SNR is more than 0dB, ROM and cos theta are converged to an ideal value 1 gradually, which shows that the difference between the channel transfer function searched by the genetic algorithm and the ideal channel transfer function is reduced gradually. Assuming that the SNR is 10dB, the fitness is calculated by equation (5), and the fitness reference value REF is 0.8092. f. ofaveAnd fmaxRespectively representing the fitness mean and the fitness maximum, and f is given in fig. 8aveAnd fmaxThe number of generations varies with the heredity. In the first 25 generations, faveAnd fmaxRapidly increase; from generation 13 to 24, fmax0.696 remains unchanged, which indicates that the search process converges to the extreme point and jumps out the extreme point in the 25 th generation; from generation 26 to 200, faveAnd fmaxThe REF is approximated and the fitness output at the end of the iteration is at a maximum of 0.7999. Assuming a SNR of 30dB, fig. 9 gives faveAnd fmaxThe number of generations varies with the heredity. The search process converges to the extreme point in generation 6, and the extreme point still cannot jump out at the end of iteration, and at this time, the number Gen of genetic generations needs to be increased. FMFPE squareThe method estimates the sound source power by using the channel transfer function searched by the genetic algorithm, the mean value is 99.7443dB, the error is 0.2527dB, the variance is 0.2883, and the FMFPE method has high environment mismatch robustness and small sound source power estimation error.
5) Sound source power estimation performance of simulation analysis FMFPE method under different SNR conditions
Assuming that the signal frequency is 500Hz and the SNR varies in the range of-20 dB to 30dB, FIG. 10 shows the source power estimation results of the FMFPE method and the MFPE method at different SNR. As can be seen from the figure, the sound source power estimation error increases with the decrease of SNR, and when the SNR is lower than 5dB, the environmental noise is the main factor influencing the sound source power estimation performance, so that the difference of the environmental mismatch robustness of different sound source power estimation methods is difficult to distinguish; therefore, the following simulation analysis was performed under conditions of large SNR. When the SNR is greater than 5dB, the environmental uncertainty is a major factor affecting the performance of sound source power estimation. Taking the SNR of 20dB as an example, the sound source power estimated value of the MFPE method is 94.3dB, and the error can reach-5.7 dB; the estimated value of sound source power of FMFPE method is 99.75dB, and the error is-0.25 dB. Therefore, the robustness of the FMFPE method for environment mismatch is obviously improved.
6) Sound source power estimation performance of simulation analysis FMFPE method under different frequency conditions
The signal radiated by the actual sound source is generally a broadband signal, and the sound source power estimation performance of the FMFPE method is simulated and analyzed when different frequencies are used. Assuming that the SNR is 20dB and the frequency variation range is 100Hz to 1000Hz, fig. 11 shows the sound source power estimation results of the FMFPE method and the MFPE method at different frequencies. In a low frequency band, the influence of environment uncertainty on the MFPE method is small, the estimation result of the sound source power at the frequency of 100Hz is 99.49dB, and the error is only 0.51 dB; as the frequency increases, the effect of environmental uncertainty on the MFPE method increases rapidly, with an error in the estimation of the source power at a frequency of 1000Hz of about 6.5 dB. In the FMFPE method, in the frequency range of 100 Hz-1000 Hz, the sound source power estimation results are all less than 99dB, and the deviation between the sound source power estimation value and the nominal value is all less than 1 dB; therefore, the FMFPE method can obtain a more accurate sound source power estimation value in a larger frequency range, and the improvement of the environment mismatch robustness is particularly obvious in a higher frequency band.
7) Sound source power estimation performance of simulation analysis FMFPE method under different sound source position conditions
The tested sound source can be at any position in the sea area, and the sound source power estimation performance of the FMFPE method when the sound source is at different positions is simulated and analyzed. Assuming that the SNR is 20dB, the depth variation range of the sound source to be detected is 1 m-100 m, and fig. 12 shows the sound source power estimation results of the FMFPE method and the MFPE method when the sound source is at different depths. When the environment is uncertain, the sound source power estimation result of the MFPE method is changed within the range of 95 dB-97.1 dB, and the minimum error is 2.9 dB. The maximum change of the sound source power estimation result of the FMFPE method is 1.3dB within the range of 98.7 dB-100.1 dB; the sound source depth is within the range of 5 m-100 m, and the error of the sound source power estimation result of the FMFPE method is less than 0.5 dB.
The range of the distance from the measured sound source to the receiving array is 1 m-1000 m, and fig. 13 shows the sound source power estimation results of the FMFPE method and the MFPE method when the distance from the sound source to the receiving array changes. When the environment is uncertain, the estimation error of the sound source power of the MFPE method is rapidly increased along with the increase of the distance from the sound source to the receiving array, and when the distance from the sound source to the receiving array is 1000m, the estimation result of the sound source power of the MFPE method is 95dB, and the error is 5 dB. The distance from the sound source to the receiving array is increased, the deviation between the sound source power estimated value of the FMFPE method and the nominal value is also increased, but the speed is slow; when the sound source is 1000m away from the receiving array, the estimation result of the sound source power of the FMFPE method is 99.06dB, and the error is only 0.94 dB. It can be seen that: for sound sources at different positions, the FMFPE method can effectively reduce the error of the sound source power estimation result and improve the robustness of environment mismatch.
According to the implementation example, the uncertain parameter focus matching field sound source power estimation method based on the genetic algorithm provided by the invention can be considered as expanding the scanning range, constructing the target function and optimizing in the scanning range, so that the deviation of the estimated channel transfer function and the ideal channel transfer function can be effectively reduced, the searched channel transfer function is used, the high-precision sound source power estimation value is successfully obtained, and the environment mismatch robustness is effectively improved.

Claims (3)

1. An uncertain parameter focus matching field sound source power estimation method based on a genetic algorithm is characterized by comprising the following steps:
step 1: carrying out FFT processing on the uniform vertical linear array collected signals to obtain a received signal frequency domain model:
X＝Ha+N
where X is the received signal frequency domain snapshot, H is the ideal channel transfer function, N is the environmental noise, and the noise power is
Figure FDA0003360509930000011
Step 2: constructing an objective function
Figure FDA0003360509930000012
Wherein: hiA transfer function of an alternative channel corresponding to the ith grid point;
step 3, performing HiH (maximum value) of search target function by using genetic algorithmGA：
The specific search process is as follows:
the first step is as follows: initialization, including setting of crossover probability PcAnd the mutation probability PmGenerating a primary population and setting a cycle termination condition;
cross probability PcSelecting the range of 0.40-0.99; probability of variation PmSelecting the range of 0.001-0.1;
initializing a population: dividing grids for the uncertain parameter variation range by adopting binary coding; each uncertain parameter is represented by K-bit binary number, and the variation range of each uncertain parameter is divided into 2K-1 grid, the binary number corresponding to the value of the uncertainty parameter Q being:
Figure FDA0003360509930000013
wherein LB is the lower bound of the range of uncertain parameter variation, UB isThe upper bound of the parameter variation range is not determined, and u is a decimal number corresponding to the binary number; considering J uncertain parameters, the total number is 2KJEach grid point, the corresponding uncertain parameter group of each grid point is represented by a J multiplied by K bit binary number; randomly selecting Pop grid points, corresponding to Pop groups of uncertain parameters, substituting the group of uncertain parameters into an alternative channel transfer function H obtained by sound field softwareiThe individual is an individual, and Pop individuals form an initial population;
setting a circulation termination condition: setting a genetic algebra counter n as 0, selecting a determined genetic algebra Gen as an iteration termination condition, wherein the number of grid points required to be calculated in the genetic algorithm searching process is as follows:
Num＝Pop×Gen
the second step is that: individual evaluation, individual fitness of genetic algorithm is determined by objective function
Figure FDA0003360509930000021
Calculating to obtain;
the third step: population evolution
The selection process comprises the following steps: the selection process of 'winning or losing' is completed by adopting two methods of roulette and elite selection;
and (3) a crossing process: randomly distributing the 'offspring' individuals obtained in the selection process, dividing every two individuals into a group, and sequentially judging each group of individuals as follows: generating a random number epsilon within the range of 0-11If epsilon1＜PcIf not, the cross operation is not executed on the group of individuals; the crossing process adopts single-point crossing, a point is randomly generated in an interval [1, J multiplied by K ] to serve as a crossing point, the crossing point divides each individual binary string into two sections, one section of binary string is exchanged to generate a new binary string, and the crossing process is finished;
and (3) mutation process: the following determinations are performed for each individual in turn: generating a random number epsilon within the range of 0-12If epsilon2＜PmIf not, the individual is not subjected to mutation operation; the mutation process adopts single point mutation in the interval of [1, J × K]Internal randomGenerating a point as a variation point, and reversing the binary number of the variation point to complete the variation process;
the fourth step: judging the loop termination, terminating iteration when n is larger than Gen, and outputting the individual with the maximum fitness when iteration is terminated; otherwise, making n equal to n +1 and returning to the second step;
and taking the absolute value of the cosine value of the included angle between the ideal value H of the channel transfer function and the receiving signal frequency domain snapshot as a reference value for comparison, wherein the expression of the reference value REF is as follows:
Figure FDA0003360509930000022
after multiple iterations, the fitness of the genetic algorithm is maximum and is close to a reference value REF;
recording a channel transfer function corresponding to an individual with the highest fitness when the genetic algorithm terminates iteration as HGA；
And 4, step 4: the sound source power estimated value obtained by the FMFPE method is as follows:
Figure FDA0003360509930000023
2. the uncertain parameter focus matching field sound source power estimation method based on genetic algorithm as claimed in claim 1, wherein: the roulette selection is the random selection of Pop individuals from the "parent" individuals.
3. The uncertain parameter focus matching field sound source power estimation method based on genetic algorithm as claimed in claim 1, wherein: the elite selection is directly reserved as the optimal individual of the "parent", and the roulette selection only produces Pop-1 "offspring" individuals.
CN202111364883.0A 2021-11-17 2021-11-17 Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm Pending CN114065630A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111364883.0A CN114065630A (en) 2021-11-17 2021-11-17 Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111364883.0A CN114065630A (en) 2021-11-17 2021-11-17 Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm

Publications (1)

Publication Number Publication Date
CN114065630A true CN114065630A (en) 2022-02-18

Family

ID=80277971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111364883.0A Pending CN114065630A (en) 2021-11-17 2021-11-17 Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm

Country Status (1)

Country Link
CN (1) CN114065630A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117148273A (en) * 2023-10-30 2023-12-01 西北工业大学青岛研究院 Underwater sound source positioning method for matching probability measure

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109539884A (en) * 2018-11-21 2019-03-29 南京长峰航天电子科技有限公司 A kind of Vector Target-missing Quantity method for parameter estimation based on GA
CN111988256A (en) * 2020-07-27 2020-11-24 南京信息工程大学 Underwater acoustic channel estimation method based on improved adaptive genetic algorithm

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109539884A (en) * 2018-11-21 2019-03-29 南京长峰航天电子科技有限公司 A kind of Vector Target-missing Quantity method for parameter estimation based on GA
CN111988256A (en) * 2020-07-27 2020-11-24 南京信息工程大学 Underwater acoustic channel estimation method based on improved adaptive genetic algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李倩倩等: "不确定海洋环境中基于贝叶斯理论的多声源定位算法", 《海洋学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117148273A (en) * 2023-10-30 2023-12-01 西北工业大学青岛研究院 Underwater sound source positioning method for matching probability measure
CN117148273B (en) * 2023-10-30 2024-02-02 西北工业大学青岛研究院 Underwater sound source positioning method for matching probability measure

Similar Documents

Publication Publication Date Title
CN109993280B (en) Underwater sound source positioning method based on deep learning
CN107202976B (en) Low-complexity distributed microphone array sound source positioning system
WO2005098647A2 (en) Genetic algorithm based selection of neural network ensemble for processing well logging data
CN113392961B (en) Method for extracting mesoscale eddy track stable sequence and predicting cyclic neural network
Agarwal et al. Statistical field estimation for complex coastal regions and archipelagos
CN110940949A (en) Quantum penguin search mechanism-based co-prime array DOA estimation method in strong impact noise environment
CN114065630A (en) Uncertain parameter focus matching field sound source power estimation method based on genetic algorithm
Heaney et al. Validation of genetic algorithm-based optimal sampling for ocean data assimilation
Blanding et al. Directed subspace search ML-PDA with application to active sonar tracking
CN116578660A (en) Sparse data-based electromagnetic target situation map construction method
Hosseini Nejad Takhti et al. Classification of marine mammals using the trained multilayer perceptron neural network with the whale algorithm developed with the fuzzy system
CN112305502A (en) Water surface and underwater sound source binary discrimination method based on array invariants
CN111914488A (en) Data regional hydrological parameter calibration method based on antagonistic neural network
CN113221651B (en) Submarine sediment classification method using acoustic propagation data and unsupervised machine learning
CN116205364A (en) Forced oscillation detection method for jointly improving VMD and kurtosis method
US10871585B2 (en) System for processing seismic data based upon linear optimization and related methods
TANG et al. Application of LVQ neural network combined with the genetic algorithm in acoustic seafloor classification
CN111914487A (en) Data-free regional hydrological parameter calibration method based on antagonistic neural network
CN115047408B (en) Underwater multi-sound-source positioning method based on single-layer large convolution kernel neural network
Zhang et al. Estimation of acoustic transfer function by using genetic algorithm
CN111505582B (en) Hydrophone array positioning method based on differential evolution algorithm
CN116011172B (en) Heuristic method for correcting numerical sound field error based on sound scattering theory and storage medium thereof
Hu et al. Spectral level prediction model of ocean ambient noise based on GA-LM-BP neural network
CN113720441B (en) Substation perimeter noise distribution reproduction method
CN116338574B (en) Sparse Bayesian learning underwater sound source positioning method based on matched beam

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
AD01 Patent right deemed abandoned
AD01 Patent right deemed abandoned

Effective date of abandoning: 20230331