CN113515048B - Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model - Google Patents
Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model Download PDFInfo
- Publication number
- CN113515048B CN113515048B CN202110932412.9A CN202110932412A CN113515048B CN 113515048 B CN113515048 B CN 113515048B CN 202110932412 A CN202110932412 A CN 202110932412A CN 113515048 B CN113515048 B CN 113515048B
- Authority
- CN
- China
- Prior art keywords
- elm
- prediction model
- pso
- matrix
- value
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 239000002245 particle Substances 0.000 claims abstract description 52
- 230000003044 adaptive effect Effects 0.000 claims abstract description 38
- 230000006870 function Effects 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 238000005457 optimization Methods 0.000 claims abstract description 19
- 239000011159 matrix material Substances 0.000 claims description 100
- 238000012549 training Methods 0.000 claims description 26
- 230000008859 change Effects 0.000 claims description 19
- 238000012360 testing method Methods 0.000 claims description 16
- 238000010276 construction Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 4
- 230000008520 organization Effects 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 210000002569 neuron Anatomy 0.000 description 12
- 230000000875 corresponding effect Effects 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 3
- 230000005284 excitation Effects 0.000 description 3
- 230000000873 masking effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 description 2
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 210000005069 ears Anatomy 0.000 description 2
- 238000013213 extrapolation Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000005236 sound signal Effects 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001544 dysphoric effect Effects 0.000 description 1
- 210000000883 ear external Anatomy 0.000 description 1
- 210000000959 ear middle Anatomy 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 230000000638 stimulation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for establishing a fuzzy self-adaptive PSO-ELM sound quality prediction model, which comprises the following steps: constructing an individual ELM prediction model through an ELM neural network; and constructing a fuzzy adaptive PSO-ELM prediction model through fuzzy control and a PSO algorithm. The method has the function of neural network regression prediction of the ELM extreme learning machine, can predict the subjective parameters of the sound quality according to the objective parameters of the sound quality, and has higher accuracy; the method has the PSO particle swarm optimization function, and can automatically search the optimal extreme learning machine parameters, so that the accuracy of model prediction is improved; the method has the function of fuzzy control self-adaptive adjustment of inertia factors of the particle swarm algorithm, and can effectively improve the convergence speed of the algorithm and improve the efficiency of the algorithm.
Description
Technical Field
The invention relates to the technical field of sound quality prediction models, in particular to a method for establishing a fuzzy self-adaptive PSO-ELM sound quality prediction model.
Background
With the application of various noise control technologies, the sound pressure level of noise in the vehicle is improved to a certain extent. However, studies have shown that the sound pressure level does not completely reflect the subjective perception of noise, and sometimes sounds with a high sound pressure level sound more pleasant than sounds with a low sound pressure level, for example, 80dB music is more comfortable than 70dB engine noise, and is less likely to cause a psychological reaction of dysphoric fatigue. Based on this phenomenon, researchers have put forward concepts of sound quality in view of the auditory characteristics of the human ears and human psychology.
In the existing sound quality evaluation, the subjective feeling of a person on sound is taken as a judgment standard, and a subjective listening test is carried out by an organization panel by means of a psychoacoustic research method to obtain a sound quality subjective evaluation result of noise. However, subjective evaluation tests have the disadvantages of poor consistency and low repeatability, and a large amount of cost is usually consumed to obtain reliable and statistically significant results, but the results are most intuitive. The objective parameter of the sound quality can be obtained by calculating parameters such as frequency and sound pressure of the sound signal. Therefore, how to efficiently predict the subjective parameters of the sound quality is very important.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and title of the application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention has been made in view of the above and/or other problems with the existing fuzzy adaptive PSO-ELM acoustic quality prediction model building methods.
Therefore, the problem to be solved by the present invention is how to provide a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
In order to solve the technical problems, the invention provides the following technical scheme: a fuzzy self-adaptive PSO-ELM sound quality prediction model establishing method comprises the steps of establishing an individual ELM prediction model through an ELM neural network; and constructing a fuzzy adaptive PSO-ELM prediction model through fuzzy control and a PSO algorithm.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: the method comprises the following steps of constructing an independent ELM prediction model, collecting acoustic signals through acoustic signal detection equipment, wherein the acoustic signals comprise a training set, a testing set and a verification set, processing and calculating the collected signals to obtain objective acoustic quality parameters of the collected acoustic signals, and subjectively evaluating the collected acoustic signals through an organization panel to obtain subjective parameters of the collected acoustic signals; and generating an input matrix X from the acoustic quality objective parameters of the training set, generating an output matrix T from the subjective parameters of the training set, and randomly generating an input layer weight matrix W and a hidden layer threshold value matrix B.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: obtaining a hidden layer output matrix H by using the input matrix X, the output matrix T, the input layer weight matrix W and the hidden layer threshold matrix B;
wherein h is a sigmoid function;
solving an output layer weight matrix beta by using H beta = T;
and (4) completing the construction of the basic structure of the ELM model through an ELM neural network.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model of the present invention, wherein: the construction of the fuzzy adaptive PSO-ELM predictive model includes the steps of,
the PSO algorithm initially randomly generates n groups of particles according to the population scale n, each group of particles forms an input layer weight matrix W and a hidden layer threshold matrix B in the ELM neural network, training set data is input, and each group generates an ELM prediction model;
substituting the test set data into the generated ELM prediction model, solving a predicted value of the annoyance degree of the subjective parameter of the sound quality, comparing the predicted value with the annoyance degree in the test set, and taking the root mean square value of the two as a fitness value to return to a PSO algorithm;
taking the group of particles with the minimum root mean square error from the PSO algorithm as an individual extreme value, comparing the individual extreme value with the group extreme value, and replacing the group extreme value with the group of particles if the individual extreme value is smaller;
generating a new inertia factor w by the PSO algorithm according to the fitness value and the change rate thereof, updating particles according to a formula, generating n groups of new particles with the same scale, and repeating the steps;
and when the iteration reaches the upper limit of the iteration times, stopping the iteration, taking a group of particles of the group extremum to form an input layer weight matrix W and a hidden layer threshold value matrix B in the ELM neural network, and generating a final ELM prediction model by matching with training set data, wherein the model is a fuzzy self-adaptive PSO-ELM prediction model.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: and replacing the verification set data into a fuzzy self-adaptive PSO-ELM prediction model, solving a prediction value of the subjective parameter annoyance degree of the sound quality, comparing the prediction value with the annoyance degree in the verification set, and evaluating the quality of the prediction model by taking the root mean square value of the two.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: the objective parameters of sound quality include loudness, roughness, fluctuation, sharpness, tonality, semantic clarity and A sound pressure level.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: and optimizing an input layer weight matrix W and a threshold matrix B of the extreme learning machine by adopting a particle swarm optimization algorithm, taking the input layer weight matrix W and the threshold matrix B of the extreme learning machine as particles of the particle swarm optimization algorithm, and substituting the obtained error root mean square value into the test set as a fitness function to perform global optimization.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: in the step of generating a new inertia factor w by the PSO algorithm according to the population extremum adaptability value and the change rate thereof, the population extremum adaptability value and the change thereof are used as input variables, and the inertia factor w is updated by a fuzzy control method.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: before the collected acoustic signals are subjected to data input, normalization processing is performed.
As a preferable scheme of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model, the method comprises the following steps: the acoustic signal data is normalized as follows:
the sound quality objective parameters are converted to [0,1] by the following formula,
x＝(X i -X min )/(X max -X min ) (3)
in the formula X i Is an objective parameter value, X, of a certain sound quality max For the corresponding maximum value, X, of the objective parameter min The objective parameter is corresponding to the minimum value.
The method has the advantages that the method has the function of neural network regression prediction of the ELM extreme learning machine, can predict the subjective parameters of the sound quality according to the objective parameters of the sound quality, and has higher accuracy; the method has the group optimization function of the PSO particle swarm optimization, and can automatically find the optimal extreme learning machine parameters, so that the accuracy of model prediction is improved; the method has the function of fuzzy control self-adaptive adjustment of inertia factors of the particle swarm algorithm, and can effectively improve the convergence speed of the algorithm and improve the efficiency of the algorithm.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:
FIG. 1 is an ELM neural network schematic diagram of a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 2 is a diagram of an ELM neural network prediction model construction process of a fuzzy adaptive PSO-ELM acoustic quality prediction model construction method.
FIG. 3 is a flow chart of a particle swarm optimization algorithm of a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 4 is a fuzzy control flow chart of a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 5 is a diagram of a triangular membership function as a population extremum fitness value for a method of constructing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 6 is a membership function graph of delta in change of the population extremum fitness value of the method for establishing the fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 7 is a control rule image of a method for building a fuzzy adaptive PSO-ELM acoustic quality prediction model.
Fig. 8 is a flow chart of a fuzzy adaptive PSO algorithm of a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
Fig. 9 is a diagram of constructing an ELM prediction model in the method of constructing the fuzzy adaptive PSO-ELM acoustic quality prediction model.
FIG. 10 is a diagram of a PSO-ELM prediction model construction process of a fuzzy adaptive PSO-ELM acoustic quality prediction model construction method.
Fig. 11 is a construction diagram of a fuzzy adaptive PSO-ELM prediction model of a method for establishing a fuzzy adaptive PSO-ELM acoustic quality prediction model.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
Example 1
Referring to fig. 1 to 11, a first embodiment of the present invention provides a method for building a fuzzy adaptive PSO-ELM acoustic quality prediction model, where the method for building the fuzzy adaptive PSO-ELM acoustic quality prediction model includes the following steps:
s1, constructing an individual ELM prediction model through an ELM neural network;
s2, constructing a fuzzy self-adaptive PSO-ELM prediction model through fuzzy control and a PSO algorithm;
an Extreme Learning Machine (ELM) is a typical single hidden layer feedforward neural network and consists of an input layer, a hidden layer and an output layer. The extreme learning machine can randomly generate an initial input weight matrix W and a hidden layer threshold matrix B, only the number of the hidden layer neurons is required to be set, and an output weight matrix is obtained by solving an equation set, so that a complete ELM neural network model is constructed, as shown in FIG. 1.
Inputting a matrix X:
in the formula x i Is the input sample.
Input weight matrix W:
in the formula w i The weights of the ith hidden layer neuron and the input layer are obtained.
Hidden layer threshold matrix B:
in the formula b i Threshold for the ith neuron.
The known output matrix T:
T＝[t 1 t Q … t Q ]' Q×m
wherein t is j For the output of the jth output layer neuron:
in the formula, H (X) is an activation function, and a hidden layer output matrix H can be obtained by calculation through an input matrix X, an input weight matrix W and a hidden layer threshold matrix B:
output weight matrix β:
in the formula beta i The weight of the ith neuron and the output layer.
The ELM neural network randomly generates an input weight matrix W and a hidden layer threshold matrix B, the input matrix X and a known output matrix T are used as the input conditions of a known training set, and the ELM can approach the training set without error, so that an output weight matrix beta can be obtained by reverse extrapolation:
Hβ＝T
c is a penalty coefficient, and finally, an output function expression of the ELM can also be obtained:
so far, the construction of the ELM neural network is completed.
In the prediction process by using the ELM neural network, an output matrix Y is required to be solved:
in the formula y i Is the output sample to be found. Comprises an input matrix X, an input weight matrix W, a hidden layer threshold matrix B,a hidden layer output matrix H can be obtained by calculation, and then an output matrix Y to be solved is obtained by solving:
Hβ＝Y
the Particle Swarm Optimization (PSO) is a swarm intelligent optimization algorithm except ant colony algorithm and fish colony algorithm in the field of intelligent computing, and the inspiration of the PSO comes from the research on bird predation.
The particle swarm algorithm firstly randomly generates a group of particles in a feasible solution, each particle represents a potential optimal solution of an optimization problem, and the characteristics of a single particle are described by three indexes of position, speed and fitness value.
The position and the speed of the particle in the solution space are continuously updated, and the motion in the solution space is further realized. Updating the individual positions by tracking the individual extremum Pbest and the group extremum Gbest (the individual extremum Pbest refers to the position with the optimal fitness value calculated in all the positions where a single particle passes, and the group extremum Gbest refers to the position with the optimal fitness value searched by all the particles in the population), calculating the fitness value once when the particles update the positions once, and updating the sizes and the positions of the individual extremum and the group extremum by comparing the fitness value of the new particles with the sizes of the individual extremum and the group extremum. The particle position update and velocity update formulas are as follows:
the particle swarm optimization algorithm flow chart is shown in FIG. 3.
Fuzzy control is a nonlinear control method, which is a control method using the basic ideas and theories of fuzzy mathematics. For a complex system, the dynamic characteristics of the system are difficult to accurately describe due to too many variables, and fuzzy control is a good solution.
The fuzzy control comprises five parts of variable definition, fuzzification processing, rule table establishment, logical reasoning and output defuzzification, and the flow of the fuzzy control constructed by the method is shown in figure 4.
In the fuzzy control, a group extreme value Gbest adaptability value and the delta of the change of the group extreme value Gbest adaptability value are selected as a first input quantity and a second input quantity of a fuzzy controller, and the discourse domain of the group extreme value Gbest adaptability value is [0,0.015 [ ] -Gbest adaptability value]The argument field of delta of change of population extreme value Gbest fitness value is [0,0.009]. And outputting the fuzzy control after defuzzification of the output of the fuzzy controller to obtain an inertia factor w of the particle swarm optimization algorithm. And performing variable fuzzification processing on the group extreme value Gtest fitness value and the change delta of the group extreme value Gtest fitness value which are used as input quantities respectively. And taking three fuzzy subsets of the group extreme value Gbest fitness value. The ranges of the fuzzy subsets of the input group extremum Gbest fitness value are respectively selected as follows: I.C. A 11 (Small positive PS) is [ -0.075,0,0.0075]，I 12 (median PM) is [0,0.0075,0.015]，I 13 (Positive PB) is [0.0075,0.015,0.0225]. Similarly, each range of the fuzzy subset corresponding to the variation delta of the group extreme value Gbest fitness value is respectively selected as follows: I.C. A 21 (Small positive PS) is [ -0.0045,0,0.0045]，I 22 (median PM) is [0.003,0.0045,0.006]，I 23 (Positive PB) is [0.006,0.0075,0.009]. And the output corresponding to the four fuzzy subsets of the output inertia factor w is defined in [0.2,0.8]Respectively as follows: o is 11 (Positive PB) is [0.8,0.65,0.5]、O 12 (median PM) is [0.65,0.5,0.35]、O 13 (Small positive PS) is [0.5,0.35,0.2]And obtaining the corresponding inertia factor w. And then determining a logic judgment rule, and selecting a triangular membership function as a membership function of the group extremum Gtest fitness value and the delta of the change of the group extremum Gtest fitness value, wherein images of the functions are respectively shown in FIGS. 5 and 6.
In step S1, constructing a separate ELM prediction model includes the steps of,
s11: acquiring acoustic signals by acoustic signal detection equipment, wherein the acoustic signals comprise a training set, a testing set and a verification set, processing and calculating the acquired signals to obtain objective parameters of the acoustic quality of the acquired acoustic signals, and subjectively evaluating the acquired acoustic signals by an organization review group to obtain subjective parameters of the acquired acoustic signals;
s12: preprocessing, i.e. normalizing, the acquired acoustic signals before inputting the data into the acoustic quality prediction model
S13: generating an input matrix X from the acoustic quality objective parameters of the training set, generating an output matrix T from the subjective parameters of the training set, randomly generating an input layer weight matrix W and a hidden layer threshold matrix B, and optimizing the input layer weight matrix W and the threshold matrix B of the extreme learning machine by adopting a particle swarm optimization algorithm. And taking an input layer weight matrix W and a threshold matrix B of the extreme learning machine as particles of the particle swarm optimization algorithm, and taking the root mean square value of the error as a fitness function to carry out global optimization.
S14: obtaining a hidden layer output matrix H by using the input matrix X, the output matrix T, the input layer weight matrix W and the hidden layer threshold matrix B;
wherein h is a sigmoid function;
solving an output layer weight matrix beta by using H beta = T;
and (4) completing the construction of the basic structure of the ELM model through an ELM neural network.
In the step S11, the objective parameters of the sound quality comprise loudness, roughness, fluctuation degree, sharpness, tonality, semantic definition and A sound pressure level; wherein,
(1) Loudness
Loudness is a psychoacoustic index of the overall loudness of sound by human ears, and is expressed in sones (song), and the loudness of pure tone with frequency of 1000Hz and sound pressure level of 40dB is defined as 1 song. And the loudness level of a sound is defined as the sound pressure level equal to a 1000Hz pure tone, where L is the loudness level N Is represented by the unitSquare (phon).
The international standard ISO532 specifies a loudness calculation method, which comprises two methods of Stevens and Zwicker, and the ISO532B Zwicker method is suitable for both a diffuse sound field and a free sound field, so that the method is generally selected to calculate the loudness in the research of the sound quality of automobiles. The process is as follows:
determining external ear and middle ear transfer functions;
filtering by using an auditory filter to obtain an excitation level E of each critical frequency band;
calculating the characteristic loudness N' of each critical frequency band according to the obtained excitation level:
in the formula, E TQ For stimulation under the auditory threshold, E 0 Excitation corresponding to reference sound pressure;
and IV, integrating the specific loudness N' in a Bark domain to obtain the total loudness N:
(2) Degree of fluctuation
When two sound signals with different frequencies and different amplitudes are superposed together, a modulation effect is generated, the fluctuation describes that when the modulation frequency is 0.5 to 20Hz, the subjective feeling of a person on slowly modulated sound is a physical quantity reflecting the fluctuation of the sound brightness degree, the unit is vacil, and the fluctuation of the 60dB 1kHz pure sound after being modulated by the frequency of 4Hz and the amplitude of 100 percent is defined as 1vacil. Calculating a model by adopting Zwicker:
in the formula, f mod Is the modulation frequency; Δ L E For masking depth, it is positively correlated with the amount of change in sound.
(3) Roughness of
Roughness is a psychological index describing the human subjective feeling of rapidly modulated sound when the modulation frequency is 20 to 300Hz, and is a feeling of noise, harshness, etc., reflecting the sound, and is expressed in unit of asper. The roughness after the frequency of 70Hz and the amplitude of 100% modulation are defined as 1asper for a pure tone of 1kHz with 60 dB. Using an Aures roughness calculation model:
in the formula f mod Is the modulation frequency; Δ L E For masking depth, which is positively correlated with the amount of sound variation, masking depth Δ L E An increase in roughness results.
(4) Sharpness degree
Sharpness is a psychoacoustic indicator describing the high frequency content of sound, in acum. Defined within a 150Hz bandwidth with a center frequency of 1000Hz, a narrow-band noise of 60dB is defined as 1 accum. The calculating method adopts a Zwicker sharpness calculating model:
wherein k is a weighting coefficient, and is generally 0.11; n is the overall loudness; n' (z) is the specific loudness of the Bark field, number z; g (z) is a function of the weight coefficients in the different Bark domains:
(5) Tone scheduling
Tone scheduling, also known as pure tone, describes the degree of prominence of a pure tone in a sound, which reflects the pure tone heard by a human or a sound within a bandwidth less than a critical frequency band, in tu. A1 kHz pure tone of 60dB is defined as 1tu. Tonality is generally calculated using the method proposed by Terhardt and Aures:
in the formula, W 1 (Δz i ) Is the difference of critical bands of the ith single-frequency component domain; w 2 (f i ) Is the relationship of frequency to the ith single frequency component; w 3 (ΔL i ) Is the sound level surplus effect of the ith single-frequency component.
(6) Semantic clarity
Speech intelligibility is an index describing the intelligibility of speech in noisy environments, expressed in percentage terms. Studies have shown that when the noise is 12dB above the pitch of the speaking voice, the speaking voice is completely inaudible, i.e. AI =0%, and an upper bound noise value UL (f) can be determined; when the noise is 30dB below the upper noise level, the speaking voice is completely heard clearly, i.e. AI =100%, and a lower noise level LL (f) is determined. Thus, there is the following equation:
UL(f)＝H(f)+12
LL(f)＝UL(f)-30
in the formula, H (f) is the sound pressure level of the speech sound.
Because the frequency of the voice of people speaking daily is basically in the range of 200-6000 Hz, a weighting coefficient W (f) is introduced to weight different frequencies, and the value of W (f) is the largest in a middle frequency band. The weighting coefficient is used together with the noise to calculate the speech intelligibility AI, and the calculation formula is as follows:
AI＝∑W(f)D(f)/30
in step S12, the acoustic signal data is normalized as follows:
the sound quality objective parameters are converted to [0,1] by the following formula,
x＝(X i -X min )/(X max -X min ) (3)
in the formula X i Is an objective parameter value, X, of a certain sound quality max For the corresponding maximum value, X, of the objective parameter min The objective parameter is corresponding to the minimum value.
In step S2, constructing the fuzzy adaptive PSO-ELM prediction model comprises the steps of,
s21: the PSO algorithm initially randomly generates n groups of particles according to the population scale n, each group of particles forms an input layer weight matrix W and a hidden layer threshold matrix B in the ELM neural network, training set data is input, and each group generates an ELM prediction model;
s22: substituting the test set data into the generated ELM prediction model, solving a predicted value of the annoyance degree of the subjective parameter of the sound quality, comparing the predicted value with the annoyance degree in the test set, and taking the root mean square value of the two as a fitness value to return to a PSO algorithm;
s23: taking the group of particles with the minimum root mean square error from the PSO algorithm as an individual extreme value, comparing the individual extreme value with the group extreme value, and replacing the group extreme value with the group of particles if the individual extreme value is smaller;
s24: generating a new inertia factor w by the PSO algorithm according to the fitness value and the change rate thereof, updating particles according to a formula, generating n groups of new particles with the same scale, and repeating the steps;
s25: when iteration reaches the upper limit of iteration times, the iteration is stopped, a group of particles of a group extreme value are taken to form an input layer weight matrix W and a hidden layer threshold value matrix B in the ELM neural network, and a final ELM prediction model is generated by matching with training set data and is a fuzzy self-adaptive PSO-ELM prediction model;
s26: and replacing the verification set data into a fuzzy self-adaptive PSO-ELM prediction model, solving a prediction value of the subjective parameter annoyance degree of the sound quality, comparing the prediction value with the annoyance degree in the verification set, and evaluating the quality of the prediction model by taking the root mean square value of the two.
Further, in the step of generating a new inertia factor w by the PSO algorithm according to the fitness value and the change rate thereof, the fitness value and the change rate thereof are used as input variables, and the inertia factor w is updated by a fuzzy control method.
Example 2
Referring to fig. 1 to 11, there is shown a second embodiment of the present invention, which is based on the above embodiment.
The acquisition of acoustic signal data by an acoustic signal detection device is shown in table 1:
TABLE 1 data set collected
The data samples of groups 1-30 are selected as training set, the samples of groups 31-35 are selected as testing set, and the samples of groups 36-40 are selected as verifying set.
Inputting a matrix X:
in the formula x i Is the input sample.
Substituting objective parameters of the training set samples as input samples, wherein the first row data is (x) 11 )-(x n1 ) The first column data is (x) 11 )-(x 1Q ) And so on, the last row of data is (x) 1Q )-(x nQ ) The last column of data is (x) n1 )-(x nQ )。
Table 2 training set sample data
(2) Input weight matrix W:
in the formula w i The weights of the ith hidden layer neuron and the input layer are obtained. The parameters are randomly generated in the initialization process and are calculated as known parameters.
(3) Hidden layer threshold matrix B:
in the formula b i Is the threshold for the ith neuron. And randomly generating in the initialization process, and calculating as a known parameter.
(4) The known output matrix T:
T＝[t 1 t 2 … t Q ]' Q×m
wherein t is j For the output of the jth output layer neuron:
in the formula, h (x) is an activation function, subjective parameters of training set samples are used as a known output matrix, and the first column of data is (t) 1 )-(t Q )。
TABLE 3 subjective annoyance
(5) And calculating to obtain a hidden layer output matrix H by using the input matrix X, the input layer weight matrix W and the hidden layer threshold matrix B:
(6) Output weight matrix β:
in the formula beta i The weight of the ith neuron and the output layer.
The ELM neural network randomly generates an input weight matrix W and a hidden layer threshold matrix B, the input matrix X, a hidden layer output matrix H and a known output matrix T are used as the input conditions of a known training set, and the ELM can approach the training set without error, so that an output weight matrix beta can be obtained by reverse extrapolation:
Hβ＝T
at this point, the construction of the ELM prediction model is completed, as shown in FIG. 9.
The particle swarm algorithm searches a space dimension D, and the space dimension D is calculated by adopting the following formula:
D＝I*H+H
wherein I is the number of input sample neurons I =7; h is the number of neurons in the hidden layer, and H =7;
the particle swarm initialization settings are as follows: t is t max Is 100 times; learning factor c 1 And c 2 1.4 are taken; maximum velocity v of particles max And a minimum velocity v min Are 1 and-1, respectively; maximum position x of particle max And a minimum position x min 1 and-1 respectively.
The fitness function selects a training set sample for reverse generation to obtain a root mean square value of a prediction output value and an actual output value sample of the training set sample as a calculation basis:
in the formula, N is the amount of training samples; m is the number of output sample neurons; y is k A predicted output value for the test set; c. C k To test the set actual output values, the PSO-ELM predictive model is shown in FIG. 10.
Designing a fuzzy knowledge base, namely a fuzzy rule according to experience and a change rule: when the population extreme value Gbest fitness value is larger and the change delta of the population extreme value Gbest fitness value is larger, the inertia factor w is larger; when the population extremum Gbest fitness value is smaller and the change delta of the population extremum Gbest fitness value is smaller, the inertia factor w is smaller. A fuzzy control rule is designed according to the idea that the population extreme value Gbest fitness value is rapidly converged along with the iteration number and is not easy to fall into the local minimum value, as shown in table 4. The image corresponding to the fuzzy control rule table is shown in fig. 7:
TABLE 4 fuzzy control rules Table
Let T be the sampling period, n be the number of sampling points, the input of the fuzzy controller is represented by Gbest (nT) and delta (nT), respectively, and the output is represented by w (nT), which are fuzzified to obtain S (nT), V (nT), and F (nT), respectively, so that the control rule can be represented as follows:
R(nT)＝S(nT)V(nT)F(nT)
defuzzification can be realized according to a logic judgment rule (TS inference), a fuzzy control rule and the corresponding size of each output fuzzy subset. Although the simple PSO algorithm can also achieve the purpose of group optimization, the convergence speed is low, the convergence can be achieved only through multiple iterations, and the calculation efficiency is low. Therefore, on the basis of the PSO algorithm, the fuzzy adaptive control is added, the adaptive change of the inertia factor w in the iterative process is realized, so as to achieve a better calculation effect, the flow chart of the fuzzy adaptive PSO algorithm is shown in FIG. 8, and the fuzzy adaptive PSO-ELM prediction model is shown in FIG. 11.
The prediction process of the built ELM prediction model is as follows:
(1) Inputting a matrix X:
in the formula x i Is the input sample.
Substituting objective parameters of the verification set samples as input samples, wherein the first row data is (x) 11 )-(x n1 ) The first column data is (x) 11 )-(x 1Q ) And so on, the last row of data is (x) 1Q )-(x nQ ) The last column of data is (x) n1 )-(x nQ )：
Table 5 verification set data
(6) From the known conditions: inputting a weight matrix W, a hidden layer threshold matrix B, an input matrix X and a hidden layer output matrix H to obtain an output matrix Y:
Y＝[y 1 y 2 … y 5 ]' 5×1
Y＝Hβ
table 6 output matrix data
This completes the prediction.
It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.
Claims (4)
1. A method for establishing a fuzzy self-adaptive PSO-ELM sound quality prediction model is characterized by comprising the following steps: the method comprises the following steps:
constructing an individual ELM prediction model through an ELM neural network;
constructing a fuzzy adaptive PSO-ELM prediction model through a fuzzy control and PSO algorithm;
constructing a separate ELM predictive model includes the steps of,
acquiring acoustic signals through acoustic signal detection equipment, wherein the acoustic signals comprise a training set, a testing set and a verification set, processing and calculating the acquired signals to obtain objective parameters of the acoustic quality of the acquired acoustic signals, and subjectively evaluating the acquired acoustic signals through an organization evaluation group to obtain subjective parameters of the acquired acoustic signals;
generating an input matrix X from the acoustic quality objective parameters of the training set, generating an output matrix T from the subjective parameters of the training set, and randomly generating an input layer weight matrix W and a hidden layer threshold value matrix B;
obtaining a hidden layer output matrix H by using the input matrix X, the output matrix T, the input layer weight matrix W and the hidden layer threshold matrix B;
wherein h is a sigmoid function;
solving an output layer weight matrix beta by using H beta = T;
the construction of the basic structure of the ELM model is completed through an ELM neural network;
the construction of the fuzzy adaptive PSO-ELM predictive model includes the following steps,
the PSO algorithm initially randomly generates n groups of particles according to the population scale n, each group of particles forms an input layer weight matrix W and a hidden layer threshold matrix B in the ELM neural network, training set data is input, and each group generates an ELM prediction model;
substituting the test set data into the generated ELM prediction model, solving a predicted value of the annoyance degree of the subjective parameter of the sound quality, comparing the annoyance degree with the annoyance degree in the test set, and taking root mean square values of the annoyance degree and the annoyance degree as fitness values to return to a PSO algorithm;
taking the group of particles with the minimum root mean square error from the PSO algorithm as an individual extreme value, comparing the individual extreme value with the group extreme value, and replacing the group extreme value with the group of particles if the individual extreme value is smaller;
generating a new inertia factor w by the PSO algorithm according to the fitness value and the change rate thereof, updating particles according to a formula, generating n groups of new particles with the same scale, and repeating the steps;
when iteration reaches the upper limit of iteration times, the iteration is stopped, a group of particles of a group extreme value are taken to form an input layer weight matrix W and a hidden layer threshold value matrix B in the ELM neural network, and a final ELM prediction model is generated by matching with training set data and is a fuzzy self-adaptive PSO-ELM prediction model;
the verification set data is substituted into a fuzzy self-adaptive PSO-ELM prediction model, the subjective parameter annoyance degree prediction value of the sound quality is solved and is compared with the annoyance degree in the verification set, and the root mean square value of the two is used for evaluating the quality of the prediction model
Optimizing an input layer weight matrix W and a threshold matrix B of the extreme learning machine by adopting a particle swarm optimization algorithm, taking the input layer weight matrix W and the threshold matrix B of the extreme learning machine as particles of the particle swarm optimization algorithm, substituting the obtained error root mean square value into a test set as a fitness function, and carrying out global optimization;
in the step of generating a new inertia factor w by the PSO algorithm according to the fitness value and the change rate thereof, the fitness value and the change rate thereof are used as input variables, and the inertia factor w is updated by a fuzzy control method.
2. The method of building a fuzzy adaptive PSO-ELM acoustic quality prediction model of claim 1, wherein: the sound quality objective parameters comprise loudness, roughness, fluctuation, sharpness, tonality, semantic clarity and A sound pressure level.
3. The method of building a fuzzy adaptive PSO-ELM acoustic quality prediction model of claim 2, wherein: before the collected acoustic signals are subjected to data input, normalization processing is performed.
4. The method of claim 3, wherein the method comprises: the acoustic signal data is normalized as follows:
the sound quality objective parameters are converted to [0,1] by the following formula,
x＝(X i -X min )/(X max -X min ) (3)
in the formula X i Is an objective parameter value, X, of a certain sound quality max Corresponding maximum value, X, for the objective parameter min The objective parameter is corresponding to the minimum value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110932412.9A CN113515048B (en) | 2021-08-13 | 2021-08-13 | Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110932412.9A CN113515048B (en) | 2021-08-13 | 2021-08-13 | Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113515048A CN113515048A (en) | 2021-10-19 |
| CN113515048B true CN113515048B (en) | 2023-04-07 |
Family
ID=78069223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110932412.9A Active CN113515048B (en) | 2021-08-13 | 2021-08-13 | Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113515048B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116125803B (en) * | 2022-12-28 | 2024-06-11 | 淮阴工学院 | Inverter backstepping fuzzy neural network control method based on extreme learning machine |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104634478A (en) * | 2015-03-06 | 2015-05-20 | 沈阳工业大学 | Soft measurement method for burning zone temperature of rotary kiln |
| CN105114242A (en) * | 2015-07-22 | 2015-12-02 | 重庆邮电大学 | Hydro governor parameter optimization method based on fuzzy self-adaptive DFPSO algorithm |
| CN106249256A (en) * | 2016-07-08 | 2016-12-21 | 辽宁工程技术大学 | Real-time GLONASS phase deviation estimation method based on particle swarm optimization algorithm |
| CN108053077A (en) * | 2017-12-28 | 2018-05-18 | 华中科技大学 | A kind of short-term wind speed forecasting method and system based on two type T-S fuzzy models of section |
| CN109947124A (en) * | 2019-04-25 | 2019-06-28 | 南京航空航天大学 | Improve particle swarm algorithm Optimization of Fuzzy PID unmanned helicopter attitude control method |
| CN111355633A (en) * | 2020-02-20 | 2020-06-30 | 安徽理工大学 | Mobile phone internet traffic prediction method in competition venue based on PSO-DELM algorithm |
| CN112133323A (en) * | 2020-09-15 | 2020-12-25 | 青岛科技大学 | Unsupervised classification and supervised modification fusion voice separation method related to spatial structural characteristics |
| WO2021026944A1 (en) * | 2019-08-09 | 2021-02-18 | 东北大学 | Adaptive transmission method for industrial wireless streaming media employing particle swarm and neural network |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102122134A (en) * | 2011-02-14 | 2011-07-13 | 华南理工大学 | Method and system for wastewater treatment of dissolved oxygen control based on fuzzy neural network |
| CN103389155B (en) * | 2013-06-26 | 2015-05-27 | 浙江工业大学 | Digital image generation method of three-dimensional spatial distribution of sound quality objective parameters |
| CN106568501B (en) * | 2016-10-25 | 2020-06-23 | 浙江工业大学 | Near-field detection method for sound quality objective parameters of low-noise product |
| CN108920854A (en) * | 2018-07-11 | 2018-11-30 | 湖南大学 | It is a kind of based on wireless interconnected and noise inline diagnosis harmony method for evaluating quality and system of athe portable client |
| CN109839824A (en) * | 2019-01-24 | 2019-06-04 | 青岛理工大学 | Network control system delay compensation method based on predictive control |
| US11138989B2 (en) * | 2019-03-07 | 2021-10-05 | Adobe Inc. | Sound quality prediction and interface to facilitate high-quality voice recordings |
-
2021
- 2021-08-13 CN CN202110932412.9A patent/CN113515048B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104634478A (en) * | 2015-03-06 | 2015-05-20 | 沈阳工业大学 | Soft measurement method for burning zone temperature of rotary kiln |
| CN105114242A (en) * | 2015-07-22 | 2015-12-02 | 重庆邮电大学 | Hydro governor parameter optimization method based on fuzzy self-adaptive DFPSO algorithm |
| CN106249256A (en) * | 2016-07-08 | 2016-12-21 | 辽宁工程技术大学 | Real-time GLONASS phase deviation estimation method based on particle swarm optimization algorithm |
| CN108053077A (en) * | 2017-12-28 | 2018-05-18 | 华中科技大学 | A kind of short-term wind speed forecasting method and system based on two type T-S fuzzy models of section |
| CN109947124A (en) * | 2019-04-25 | 2019-06-28 | 南京航空航天大学 | Improve particle swarm algorithm Optimization of Fuzzy PID unmanned helicopter attitude control method |
| WO2021026944A1 (en) * | 2019-08-09 | 2021-02-18 | 东北大学 | Adaptive transmission method for industrial wireless streaming media employing particle swarm and neural network |
| CN111355633A (en) * | 2020-02-20 | 2020-06-30 | 安徽理工大学 | Mobile phone internet traffic prediction method in competition venue based on PSO-DELM algorithm |
| CN112133323A (en) * | 2020-09-15 | 2020-12-25 | 青岛科技大学 | Unsupervised classification and supervised modification fusion voice separation method related to spatial structural characteristics |
Non-Patent Citations (3)
| Title |
|---|
| Jie Zhang等.Robust Extreme Learning Machine for Modeling with Unknown Noise.《Preprint submitted to Journal of the Franklin Institute》.2020,第9885- 9908页. * |
| Sound quality prediction of vehicle interior noise and mathematical modeling using a back propagation neural network (BPNN) based on particle swarm optimization (PSO);Enlai Zhang等;《Measurement Science and Technology》;20151210;第1-9页 * |
| 黄海波 ; 李人宪 ; 黄晓蓉 ; 杨明亮 ; 丁渭平 ; .基于Adaboost算法的车内噪声声品质预测.汽车工程.2016,第38卷(第09期),第1120-1125页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113515048A (en) | 2021-10-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xing et al. | Sound quality recognition using optimal wavelet-packet transform and artificial neural network methods | |
| Johnson-Laird et al. | On musical dissonance | |
| Cartwright et al. | Social-EQ: Crowdsourcing an Equalization Descriptor Map. | |
| Yang et al. | Psychoacoustical evaluation of natural and urban sounds in soundscapes | |
| RU2403626C2 (en) | Base frequency detecting speech analyser, speech analysis method and speech analysis program | |
| CN112006697B (en) | Voice signal-based gradient lifting decision tree depression degree recognition system | |
| Ma et al. | Sound quality evaluation of noise of hub permanent-magnet synchronous motors for electric vehicles | |
| Jin et al. | Evaluation and modeling of automotive transmission whine noise quality based on MFCC and CNN | |
| CN113515048B (en) | Method for establishing fuzzy self-adaptive PSO-ELM sound quality prediction model | |
| CN110827857A (en) | Speech emotion recognition method based on spectral features and ELM | |
| CN108597540A (en) | A kind of speech-emotion recognition method based on variation mode decomposition and extreme learning machine | |
| Garnier et al. | Characterisation of voice quality in Western lyrical singing: From teachers' judgements to acoustic descriptions | |
| Marjieh et al. | Timbral effects on consonance disentangle psychoacoustic mechanisms and suggest perceptual origins for musical scales | |
| Manjula et al. | Adaptive optimization based neural network for classification of stuttered speech | |
| Gaspar et al. | Psychoacoustics of in-car switch buttons: from feelings to engineering parameters | |
| Gao et al. | Interior sound quality evaluation model of heavy commercial vehicles | |
| Lin et al. | Acoustic recognition method in low SNR based on human ear bionics | |
| Peng et al. | Speech emotion recognition of merged features based on improved convolutional neural network | |
| Petersen et al. | Evaluating emotionalizing effects of active sound designs | |
| CN113313397A (en) | Sound quality satisfaction degree grading and limit value determining method | |
| Hamadicharef et al. | Intelligent and perceptual-based approach to musical instruments sound design | |
| Marjieh et al. | Timbral effects on consonance illuminate psychoacoustics of music evolution | |
| Sadeghian et al. | The use of artificial neural networks to predict tonal sound annoyance based on noise metrics and psychoacoustics parameters | |
| Chen et al. | Speech fatigue detection based on deep learning | |
| Chang et al. | A masking-threshold-adapted weighting filter for excitation search |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |