CN110674996B - Urban traffic noise prediction method - Google Patents
Urban traffic noise prediction method Download PDFInfo
- Publication number
- CN110674996B CN110674996B CN201910925415.2A CN201910925415A CN110674996B CN 110674996 B CN110674996 B CN 110674996B CN 201910925415 A CN201910925415 A CN 201910925415A CN 110674996 B CN110674996 B CN 110674996B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- gamma
- influence factor
- optimal
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Abstract
The invention provides an urban traffic noise prediction method, which comprises the following steps: s1, measuring and recording traffic noise influence factor data on urban roads, establishing a traffic noise influence factor database, S2, determining genetic algorithm parameters, and generating an initial population in a bit string form by encoding whether the influence factors are selected or not; s3, completing the optimal solution search of the influence factor combination by setting the genetic algorithm parameters; s4, obtaining an optimal influence factor combination by using a Gamma-Test nonlinear data analysis method; s5, determining the required quantity of BP neural network model training data through M-Test to obtain a nonlinear traffic noise prediction model with given quality; s6, constructing a BP neural network model by using the obtained optimal influence combination factors, training the BP neural network model by using the data volume obtained in S5, and completing prediction of urban traffic noise, so that the noise prediction has better implication and fault tolerance, and higher fitting accuracy and adaptability.
Description
Technical Field
The invention belongs to the technical field of traffic noise prediction methods, and particularly relates to an urban traffic noise prediction method.
Background
Road traffic is a main source of urban environmental noise, and noise has an important influence on physical and mental health, public health and labor efficiency of people, so that predictive modeling of the environmental noise is very important.
The current traffic noise prediction technology is mainly based on regression analysis, is not enough to describe the variation trend of noise, and has serious defects.
Disclosure of Invention
The present invention provides a method for predicting urban traffic noise, aiming at the deficiencies of the prior art, so as to solve the problem that the current traffic noise prediction technology proposed in the background art is mainly based on regression analysis and is not enough to describe the variation trend of noise.
In order to solve the technical problems, the invention adopts the technical scheme that: a method for predicting urban traffic noise comprises the following steps:
s1, measuring and recording traffic noise influence factor data at an observation point of the urban road, establishing a traffic noise influence factor database by using the influence factor data, and carrying out normalization processing on the influence factor data;
s2, determining genetic algorithm parameters, and generating an initial population in a bit string form by encoding whether the influence factors are selected or not;
S3, completing the optimal solution search of the influence factor combination by setting the genetic algorithm parameters;
s4, based on the optimal solution search of the influence factor combination obtained in S3, selecting the optimal influence factor to be input before calibrating and testing the model by using a Gamma-Test nonlinear data analysis method and taking the minimum Gamma value as a standard, and obtaining the optimal influence factor combination;
s5, combining the optimal influence factors obtained in the S4, and generating a stable asymptote through M-Test to determine the required quantity of the training data of the BP neural network model so as to obtain a nonlinear traffic noise prediction model with given quality;
s6, constructing a BP neural network model by using the optimal influence combination factors obtained by Gamma-Test, and training the BP neural network model by using the data volume obtained by S5 to complete the prediction of urban traffic noise.
Preferably, in S1, the influence factors include the number of large vehicles and the average speed, the number of medium vehicles and the average speed, the number of cars and the average speed, the number of motorcycles and the average speed, the length and the width of the road section, the height of the building around the measuring point, and the distance between the measuring point and the center line, which are 12 influence factors.
Preferably, in S1, the data is normalized by setting the original input data to x 1、x2、x3…xnThen the normalized data is:
wherein: y isi∈[0.1,0.9]。
Preferably, in S2, the genetic algorithm parameters include four parameters of population size, mutation probability, cross probability and gradient fitness.
Preferably, in S4, the Gamma-Test is specifically processed as
Setting the given traffic noise training sample data as follows:
{x1(i),…,xm(i),Leq(i)}={(xi,Leqi)|1≤i≤M}
wherein: x is the number ofiIs the input of the sample or samples and,
Leqiis the equivalent continuous sound pressure level of the sound,
m is the number of samples,
m is the input sample embedding dimension and,
xiincluding pair output LeqiThe factor with predictive effect is input xiAnd an output LeqiThe relationship between can be decomposed as:
Leqi=f(x1,…,xm)+r
in the formula: f is a smooth function, r represents a random amount of noise with a mean of zero and a variance of Var (r).
To calculate Var (r), the input data x are first calculatediAverage distance from its k-th nearest neighbor:
in the formula: i is more than or equal to 1 and less than or equal to M, k is more than or equal to 1 and less than or equal to p
p is the number of adjacent points, and is usually 10-50;
xN[i,k]is xiThe kth nearest neighbor of (1);
the average distance corresponding to the output value is:
in the formula: l iseqiDenotes xiA corresponding output;
LeqNdenotes xiX of the k-th nearest neighbor of (2)N[i,k]Outputting;
for p points (delta)M(k),γM(k) ) a one-dimensional linear regression model is constructed in combination and fitted using the least squares method:
γ=Aδ+Γ
when δ → 0, γ → Var (r), the intercept with the vertical axis is the value of the statistic Γ, an approximation of Var (r);
Gamma-Test provides the optimal MSE estimation of the continuous variable to the unknown smooth function and can realize the modeling technology, the estimation result is applicable to the subsequent smooth function model constructed by any specific technology, and the input combination corresponding to the minimum value Gamma is the optimal input data combination;
the gradient A may provide useful information on the complexity of the study system, while V may be usedratioNormalized result Γ, VratioGiving an invariant noise estimate of the model's Γ value between scales 0 and 1, which expresses goodness of fit to the smooth function class output data with bounded derivatives, i.e., predictability of a given output using available inputs;
Vratiois defined as:
wherein, VratioRepresents the variance of the output y if Γ, VratioWith smaller values, a better quality smooth model can be built.
Preferably, V is increased with the number M of samplesratioAnd gamma tends to a fixed value if M is VratioAnd the gamma value approaches to the number of input data required by a fixed value, which indicates that at least M data are required to construct a model with mean square error of gamma;
if the value of M increases, the V obtained by M-TestratioAnd the gamma value sequence does not tend to be stable, which indicates that the noise of the input and the output of the model is too large, some characteristic factors influencing the output result are possibly missed or the input factors contain characteristic factors which are useless for prediction, and the established model is not smooth and can determine how many samples M are needed to obtain a model with given quality by generating a stable asymptote through M-Test, so that the overfitting of the model when MSE is smaller than Var (r) in a training stage is avoided.
Preferably, in S6, the BP neural network model is composed of 8 input neurons, a hidden layer including 13 neurons, and an output layer including one neuron.
Compared with the prior art, the invention has the following advantages:
the method measures and records traffic noise influence factor data through observation points of urban roads, discusses the applicability of the Gamma-Test in the aspect of providing input data guidance before the traffic noise model is developed through the Gamma-Test on the data of the urban traffic noise influence factor, analyzes model noise under different input parameter combinations through a GT method, and is used for determining and constructing the optimal combination of the traffic noise prediction model influence factor; the optimal combination factor obtained by the GT is combined with a BP neural network model to establish a nonlinear prediction system, the relation between a sound field and a sound source is described by using the nonlinear relation by using the connection weight among the neurons of the neural network, and a TNP model more related to the urban environment is established, so that the aims of improving the prediction precision of urban traffic noise and reducing the workload of model development are fulfilled.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a diagram of an optimal input data combination according to an embodiment 2 of the present invention, in which the minimum Γ is a combination corresponding to the value Var (r);
FIG. 3 shows that M-Test and minimum V are performed in eight combinations in example 2 of the present inventionratioAnd Γ corresponds to the result graph;
FIG. 4 is a graph of the verification of the best performance of the training process of model 1 in embodiment 2 of the present invention;
fig. 5 is a correlation diagram between the predicted value and the measured value of the model 1 in embodiment 2 of the present invention;
fig. 6 is a graph of the test results of the trained noise model for four roads, which are measured by randomly extracting 30 groups of data from the four roads in embodiment 2 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Embodiment 1, as shown in fig. 1, the present invention provides a technical solution: a method for predicting urban traffic noise comprises the following steps:
and S1, measuring and recording traffic noise influence factor data at the observation points of the urban road, wherein the measurement points are arranged in sections during road measurement, 30 measurement points are arranged in total, each point is measured every 5 seconds, and the measurement time is 10 minutes. Taking the average value as the actual traffic noise measurement value of each road segment, simultaneously recording the traffic flow and the average speed of each vehicle, considering from the sound source intensity model and the noise propagation model according to the actual traffic condition of the Kaifeng city, and particularly dividing the influence factors into the following steps when establishing a database: the number and average speed of large vehicles, the number and average speed of medium vehicles, the number and average speed of cars, the number and average speed of motorcycles, the length and width of road sections, the height of buildings around measuring points and the distance between the measuring points and a central line are all 12 factors, and a traffic noise influence factor database is established by utilizing the influence factor data,
Then normalizing the influence factor data, specifically setting the original input data as x1、x2、x3…xnThen the normalized data is:
wherein: y isi∈[0.1,0.9];
S2, determining genetic algorithm parameters by setting the genetic algorithm parameters, including four parameters of population size, variation probability, cross probability and gradient fitness, specifically setting the parameters as shown in the following table, and generating an initial population in a bit string form by selecting or not coding the influence factors;
the combination of '1' and '0' is used as a mask when the genetic algorithm is used for optimizing and searching, the '1' represents whether the factor is selected, the genetic algorithm optimizing and searching finds out 8 ways of the optimal influence factor combination, and the GT operand and the model development period are greatly reduced.
S3, completing the optimal solution search of the influence factor combination by setting the genetic algorithm parameters;
eight combination modes of model input factors obtained after genetic algorithm optimizing search are used for calculating a statistic gamma corresponding to the model input factors, and the statistic gamma is specifically shown in the following table:
respectively computing (delta) for eight combinationsM(k),γM(k) P is 10 near-point values, the 10 values are fitted by the least square method, the intercept between the regression equation obtained after fitting and the y-axis, i.e. Γ is an approximate value of var (r), and the combination with the minimum Γ being the value of var (r) is the optimal input data combination, specifically as shown in fig. 2, in this embodiment Γ is 0.0241, V (r) · ratio=0.0961。
S4, based on the optimal solution search of the influence factor combination obtained in S3, selecting the optimal influence factor to be input before calibrating and testing the model by using a Gamma-Test nonlinear data analysis method and taking the minimum Gamma value as a standard, and obtaining the optimal influence factor combination;
the Gamma-Test comprises the specific process of
Setting the given traffic noise training sample data as follows:
{x1(i),…,xm(i),Leq(i)}={(xi,Leqi)|1≤i≤M}
wherein: x is the number ofiIs the input of the sample or samples and,
Leqiis the equivalent continuous sound pressure level of the sound,
m is the number of samples,
m is the input sample embedding dimension and,
xiincluding pair output LeqiThe factor with predictive effect is input xiAnd an output LeqiThe relationship between can be decomposed as:
Leqi=f(x1,…,xm)+r
in the formula: f is a smooth function, r represents a random amount of noise with a mean of zero and a variance of Var (r).
To calculate Var (r), the input data x are first calculatediAverage distance from its k-th nearest neighbor:
in the formula: i is more than or equal to 1 and less than or equal to M, k is more than or equal to 1 and less than or equal to p
p is the number of adjacent points, and is usually 10-50;
xN[i,k]is xiThe kth nearest neighbor of (1);
the average distance corresponding to the output value is:
in the formula: l iseqiDenotes xiA corresponding output;
LeqNdenotes xiX of the k-th nearest neighbor of (2)N[i,k]Outputting;
for p points (delta)M(k),γM(k) ) a one-dimensional linear regression model is constructed in combination and fitted using the least squares method:
γ=Aδ+Γ
when δ → 0, γ → Var (r), the intercept with the vertical axis is the value of the statistic Γ, an approximation of Var (r);
Gamma-Test provides the optimal MSE estimation of the continuous variable to the unknown smooth function and can realize the modeling technology, the estimation result is applicable to the subsequent smooth function model constructed by any specific technology, and the input combination corresponding to the minimum value Gamma is the optimal input data combination;
the gradient A may provide useful information on the complexity of the study system, while V may be usedratioNormalization result Γ, VratioGiving an invariant noise estimate of the model's Γ value between scales 0 and 1, which expresses goodness of fit to the smooth function class output data with bounded derivatives, i.e., predictability of a given output using available inputs;
Vratiois defined as:
wherein, VratioRepresents the variance of the output y if Γ, VratioWith smaller values, a better quality smooth model can be built.
S5, generating a stable asymptote through the M-Test to determine the required quantity of training data by the optimal influence factor combination obtained in S4, generating a stable asymptote through the M-Test to determine the required quantity of the training data of the BP neural network model by the optimal influence factor combination obtained in S4, and obtaining the nonlinear traffic noise prediction model with given quality; specifically, V is the number of samples M ratioAnd gamma tends to a constant value if M is VratioAnd the gamma value approaches to the number of input data required by a fixed value, which indicates that at least M data are required to construct a model with mean square error of gamma;
if the value of M increases, V obtained by M-TestratioAnd the gamma value sequence does not tend to be stable, which indicates that the noise of the input and the output of the model is too large, some characteristic factors influencing the output result are possibly missed or the input factors contain characteristic factors which are useless for prediction, and the established model is not smooth and can determine how many samples M are needed to obtain a model with given quality by generating a stable asymptote through M-Test, so that the overfitting of the model when MSE is less than Var (r) in a training stage is avoided
Determining the proper training data M is significant for improving the prediction accuracy of the model, and the minimum Gamma and V are used in the embodimentratioAs an optimization target, M-Test is performed, and each time Γ is continuously calculated, the input data amount is increased by a small step until all data is used or the statistic converges to a fixed value. M-Test and minimum V are respectively carried out on the eight combination modesratioThe result of correspondence with Γ is shown in fig. 3, and it is understood from fig. 3 that V is 1000 after the input data amount M is input ratioAnd the gamma value sequence tends to be stable, which shows that the matching degree of the output result and the prediction result is higher when the data length is used for model training, and the model has stronger prediction capability.
S6, constructing a BP neural network model by using the optimal influence combination factors obtained by Gamma-Test, training the BP neural network model by using the data volume obtained by S5, and completing the prediction of urban traffic noise, wherein the BP neural network model is composed of 8 input neurons, a hidden layer containing 13 neurons and an output layer containing one neuron, thereby determining the noise model with the highest precision and completing the prediction of urban traffic noise.
The hidden layer activation function of the BP neural network model used in the present embodiment uses a hyperbolic tangent function
If the network response is not the best match with the expected response, the weight needs to be updated and learnt again by back propagation of the network, wherein the gradient descent-based search method is a rapid optimization technology for updating the weight of the artificial neural network, and the LM algorithm is used in the embodiment to calculate the minimum value of the objective function. Experiments prove that the method is faster than other optimization algorithms and is less prone to falling into local minimum values.
The weight value updating formula is as follows:
wk+1=wk+[JT(w)J(w)+μI]-1JT(w)e(w)
in the formula: i is an identity matrix;
μ is a user-defined learning rate;
j (w) is a Jacobian matrix;
e (w) is a residual vector.
s1, measuring and recording traffic noise influence factor data at observation points of urban roads, wherein the influence factors comprise 12 influence factors including the number and average speed of large vehicles, the number and average speed of medium vehicles, the number and average speed of cars, the number and average speed of motorcycles, the length and width of road sections, the height of buildings around measuring points and the distance between the measuring points and a central line, establishing a traffic noise influence factor database by using the influence factor data, and recording the influence factor data of the traffic noisePerforming normalization processing on the subdata, specifically setting the original input data as x1、x2、x3…xnThen the normalized data is:
wherein: y isi∈[0.1,0.9];
S2, determining genetic algorithm parameters, and generating an initial population in a bit string form by encoding whether the influence factors are selected or not;
s3, completing the optimal solution search of the influence factor combination by setting the genetic algorithm parameters;
s4, based on the optimal solution search of the influence factor combination obtained in S3, selecting the optimal influence factor to be input before calibrating and testing the model by using a Gamma-Test nonlinear data analysis method and taking the minimum Gamma value as a standard, and obtaining the optimal influence factor combination;
The specific Gamma-Test process is that the given traffic noise training sample data is set as follows:
{x1(i),…,xm(i),Leq(i)}={(xi,Leqi)|1≤i≤M}
wherein: x is the number ofiIs the input of the sample or samples and,
Leqiis the equivalent continuous sound pressure level of the sound,
m is the number of samples,
m is the input sample embedding dimension and,
xiincluding pair output LeqiThe factor with predictive effect is input xiAnd an output LeqiThe relationship between can be decomposed as:
Leqi=f(x1,…,xm)+r
in the formula: f is a smooth function, r represents a random amount of noise with a mean of zero and a variance of Var (r).
To calculate Var (r), the input data x are first calculatediAverage distance from its k-th nearest neighbor:
in the formula: i is more than or equal to 1 and less than or equal to M, k is more than or equal to 1 and less than or equal to p
p is the number of adjacent points, and is usually 10-50;
xN[i,k]is xiThe kth nearest neighbor of (1);
the average distance corresponding to the output value is:
in the formula: l iseqiDenotes xiA corresponding output;
LeqNdenotes xiX of the k-th nearest neighbor of (2)N[i,k]Outputting;
for p points (delta)M(k),γM(k) ) a one-dimensional linear regression model is constructed in combination and fitted using the least squares method:
γ=Aδ+Γ
when δ → 0, γ → Var (r), the intercept with the vertical axis is the value of the statistic Γ, an approximation of Var (r);
Gamma-Test provides the optimal MSE estimation of the continuous variable to the unknown smooth function and can realize the modeling technology, the estimation result is applicable to the subsequent smooth function model constructed by any specific technology, and the input combination corresponding to the minimum value Gamma is the optimal input data combination;
The gradient A may provide useful information on the complexity of the study system, while V may be usedratioNormalized result Γ, VratioGiven a constant noise estimate of the Γ value of the model between scales 0 and 1, it expresses the goodness of fit for smooth function-like output data with bounded derivatives, i.e. the predictability of a given output using the available inputs;
Vratiois defined as follows:
wherein, VratioRepresents the variance of the output y if Γ, VratioWith smaller values, a better quality smooth model can be built.
S5, combining the optimal influence factors obtained in the S4, and generating a stable asymptote through M-Test to determine the required quantity of the training data of the BP neural network model so as to obtain a nonlinear traffic noise prediction model with given quality; specifically, V is the number of samples MratioAnd gamma tends to a fixed value if M is VratioAnd the gamma value approaches to the number of input data required by a fixed value, which indicates that at least M data are required to construct a model with mean square error of gamma;
if the value of M increases, the V obtained by M-TestratioAnd the gamma value sequence does not tend to be stable, which indicates that the noise of the input and the output of the model is too large, some characteristic factors influencing the output result are possibly missed or the input factors contain characteristic factors which are useless for prediction, and the established model is not smooth and can determine how many samples M are needed to obtain a model with given quality by generating a stable asymptote through M-Test, so that the overfitting of the model when MSE is smaller than Var (r) in a training stage is avoided.
S6, constructing a BP neural network model by using the optimal influence combination factors obtained by Gamma-Test, training the BP neural network model by using the data volume obtained by S5 to complete the prediction of urban traffic noise, listing the objective standard values of 8 models corresponding to eight influence factor combinations in the following table,
as can be seen from the table, the coefficient of determination R of model 12Large value of gamma, VratioAnd the gradient is small, and the output result of the model established by using the combination parameters determined by the model 1 is most likely to define the variability of the traffic noise;
meanwhile, a BP neural network model is compared and analyzed with the noise model established in S5, the BP neural network model is composed of 8 input neurons, a hidden layer containing 13 neurons and an output layer containing one neuron, and therefore the noise model with the highest precision is determined, and prediction of urban traffic noise is completed.
The hidden layer activation function of the BP neural network model used in the present embodiment uses a hyperbolic tangent function
If the network response is not the best match with the expected response, the weight needs to be updated and learnt again by back propagation of the network, wherein the gradient descent-based search method is a rapid optimization technology for updating the weight of the artificial neural network, and the LM algorithm is used in the embodiment to calculate the minimum value of the objective function. Experiments prove that the method is faster than other optimization algorithms and is less prone to falling into local minimum values.
The weight value updating formula is as follows:
wk+1=wk+[JT(w)J(w)+μI]-1JT(w)e(w)
in the formula: i is an identity matrix; μ is a user-defined learning rate; j (w) is a Jacobian matrix; e (w) is a residual vector.
In order to Test the effect of Gamma-Test and the selection of the optimal combination factor, in the embodiment, two models of all input factors and the optimal input factor combination are respectively established according to model 1 and model 8 traffic noise influence factors, and according to the modeling requirement of a neural network, 1-1000 groups of data are used for network training in the embodiment; 1001-1150 group data are used for network verification; 1151-1300 groups of data are used for network test; the rest is used as the test data of the trained network. The MSE should be set to be a small enough quantity (for example: 0.0241) in the fitting criterion in the model calibration stage, and is set to be 1 in the implementation, and the number of layers of the ANN used by the model 8 is proved to be optimal when the number of the layers is 12 multiplied by 20 multiplied by 1; model 1 is optimal when the number of layers is 8 × 13 × 1. FIG. 4 is a training process best performance validation of model 1; fig. 5 reflects the correlation between the predicted value and the measured value of model 1.
Common decision coefficient R for predicting performance of BP neural network model2And mean square error, the decision coefficient reflecting the closeness of the relationship between the variables, R2The larger and closer to 1, the higher the interpretation of the independent variable on the dependent variable, the denser the observation points are near the regression line, and the better the parameter fit. Model 1 training phase: r 2The value is approximately equal to 0.93t, and the MSE is approximately equal to 1.20 e-3; a verification stage: r2The MSE is approximately equal to 0.934 and 1.22 e-3; test phase, R20.911 and MSE 1.51e-3, R for model 1 and model 8 at each stage is shown in the table below2And a value of the mean square error,
the result shows that the model 1 is superior to the model 8 in performance and has better traffic noise prediction accuracy.
In order to verify the accuracy degree of the prediction result of the model 1, 30 groups of data of each of four roads are randomly extracted from the measurement data to test the trained noise model 1, the result is shown in figure 6, the absolute value of the error is 0.15-2.4 dB, and the noise model has good adaptability.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (6)
1. A method for predicting urban traffic noise is characterized by comprising the following steps:
s1, measuring and recording traffic noise influence factor data at an observation point of the urban road, establishing a traffic noise influence factor database by using the influence factor data, and carrying out normalization processing on the influence factor data;
s2, determining genetic algorithm parameters, and generating an initial population in a bit string form by coding whether the influence factors are selected or not;
s3, completing the optimal solution search of the influence factor combination by setting the genetic algorithm parameters;
s4, based on the optimal solution search of the influence factor combination obtained in S3, selecting the optimal influence factor to be input before calibrating and testing the model by using a Gamma-Test nonlinear data analysis method and taking the minimum Gamma value as a standard, and obtaining the optimal influence factor combination;
s5, combining the optimal influence factors obtained in the S4, and generating a stable asymptote through M-Test to determine the required quantity of the training data of the BP neural network model so as to obtain a nonlinear traffic noise prediction model with given quality;
S6, constructing a BP neural network model by using the optimal influence combination factors obtained by Gamma-Test, and training the BP neural network model by using the data volume obtained by S5 to complete the prediction of urban traffic noise;
s4, based on the optimal solution search of the influence factor combination obtained in S3, selecting the optimal influence factor to be input before calibrating and testing the model by using a Gamma-Test nonlinear data analysis method and taking the minimum Gamma value as a standard, and obtaining the optimal influence factor combination;
the Gamma-Test comprises the specific process of
Given traffic noise training sample data is:
{x1(i),…,xm(i),Leq(i)}={(xi,Leqi)|1≤i≤M}
wherein: x is the number ofiIs the input of the sample or samples and,
Leqiis the equivalent continuous sound pressure level of the sound,
m is the number of samples,
m is the input sample embedding dimension and,
xiincluding pair output LeqiThe factor with predictive effect is input xiAnd an output LeqiThe relationship between can be decomposed as:
Leqi=f(x1,…,xm)+r
in the formula: f is a smooth function, r represents a random amount of noise with a mean value of zero and a variance of var (r);
to calculate Var (r), the input data x are first calculatediAverage distance from its k-th nearest neighbor:
in the formula: i is more than or equal to 1 and less than or equal to M, k is more than or equal to 1 and less than or equal to p
p is the number of adjacent points, and is usually 10-50;
xN[i,k]is xiThe kth nearest neighbor of (1);
the average distance corresponding to the output value is:
In the formula: l is a radical of an alcoholeqiRepresents xiA corresponding output;
LeqNdenotes xiX of the k-th nearest neighbor of (2)N[i,k]Outputting;
for p points (delta)M(k),γM(k) ) a one-dimensional linear regression model is constructed in combination and fitted using the least squares method:
γ=Aδ+Γ
when δ → 0, γ → Var (r), the intercept with the vertical axis is the value of the statistic Γ, an approximation of Var (r);
Gamma-Test provides the optimal MSE estimation of the continuous variable to the unknown smooth function and can realize the modeling technology, the estimation result is applicable to the subsequent smooth function model constructed by any specific technology, and the input combination corresponding to the minimum value Gamma is the optimal input data combination;
the gradient A may provide useful information on the complexity of the study system, while V may be usedratioNormalization result Γ, VratioGiving an invariant noise estimate of the model's Γ value between scales 0 and 1, which expresses goodness of fit to the smooth function class output data with bounded derivatives, i.e., predictability of a given output using available inputs;
Vratiois defined as:
wherein, VratioRepresents the variance of the output y if Γ, VratioWith smaller values, a better quality smooth model can be built.
2. The method of claim 1, wherein in step S1, the influence factors include total 12 influence factors including the number of large vehicles and the average speed, the number of medium vehicles and the average speed, the number of cars and the average speed, the number of motorcycles and the average speed, the length and width of the road, the height of the building around the measuring point and the distance between the measuring point and the center line.
4. The method according to claim 1, wherein in S2, the genetic algorithm parameters include four parameters of population size, variation probability, cross probability and gradient fitness.
5. The method of claim 1, wherein V is the number of samples M that increasesratioAnd gamma tends to a constant value if M is VratioAnd the gamma value approaches to the number of input data required by a fixed value, which indicates that at least M data are required to construct a model with mean square error of gamma;
if the value of M increases, the V obtained by M-TestratioAnd the gamma value sequence does not tend to be stable, which indicates that the noise of the input and the output of the model is too large, some characteristic factors influencing the output result are possibly missed or the input factors contain characteristic factors which are useless for prediction, at the moment, the established model is not smooth, and a stable asymptote is generated through M-Test to determine how many samples M are needed to obtain a model with given quality, so that the overfitting of the model when MSE is smaller than Var (r) in a training stage is avoided.
6. The method of claim 1, wherein in S6, the BP neural network model comprises 8 input neurons, a hidden layer comprising 13 neurons, and an output layer comprising a neuron.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910925415.2A CN110674996B (en) | 2019-09-27 | 2019-09-27 | Urban traffic noise prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910925415.2A CN110674996B (en) | 2019-09-27 | 2019-09-27 | Urban traffic noise prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110674996A CN110674996A (en) | 2020-01-10 |
CN110674996B true CN110674996B (en) | 2022-06-10 |
Family
ID=69079571
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910925415.2A Active CN110674996B (en) | 2019-09-27 | 2019-09-27 | Urban traffic noise prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110674996B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113657490B (en) * | 2021-08-16 | 2022-05-31 | 沭阳县源美装饰材料有限公司 | Door and window silence detection method based on artificial intelligence |
CN114973657A (en) * | 2022-05-12 | 2022-08-30 | 中南大学 | Urban traffic noise pollution analysis and evaluation method based on trajectory data |
CN115859765B (en) * | 2022-09-29 | 2023-12-08 | 中山大学 | Urban expansion prediction method, device, equipment and storage medium |
CN116320042B (en) * | 2023-05-16 | 2023-08-04 | 陕西思极科技有限公司 | Internet of things terminal monitoring control system for edge calculation |
CN116453541B (en) * | 2023-06-16 | 2023-09-19 | 中山大学 | Sound source intensity prediction method and device, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104636801A (en) * | 2013-11-08 | 2015-05-20 | 国家电网公司 | Transmission line audible noise prediction method based on BP neural network optimization |
CN107180273A (en) * | 2017-05-09 | 2017-09-19 | 国网内蒙古东部电力有限公司电力科学研究院 | A kind of transformer station's factory outside noise prediction and evaluation method based on big data statistical analysis |
CN107730054A (en) * | 2017-11-15 | 2018-02-23 | 西南石油大学 | A kind of Gas Load combination forecasting method based on support vector regression |
CN108428012A (en) * | 2018-03-12 | 2018-08-21 | 株洲联诚集团控股股份有限公司 | A kind of fan noise prediction technique of optimization neural network |
CN105760658B (en) * | 2016-02-03 | 2018-09-28 | 华东交通大学 | A kind of bullet train noise prediction method of Interval neural networks |
-
2019
- 2019-09-27 CN CN201910925415.2A patent/CN110674996B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104636801A (en) * | 2013-11-08 | 2015-05-20 | 国家电网公司 | Transmission line audible noise prediction method based on BP neural network optimization |
CN105760658B (en) * | 2016-02-03 | 2018-09-28 | 华东交通大学 | A kind of bullet train noise prediction method of Interval neural networks |
CN107180273A (en) * | 2017-05-09 | 2017-09-19 | 国网内蒙古东部电力有限公司电力科学研究院 | A kind of transformer station's factory outside noise prediction and evaluation method based on big data statistical analysis |
CN107730054A (en) * | 2017-11-15 | 2018-02-23 | 西南石油大学 | A kind of Gas Load combination forecasting method based on support vector regression |
CN108428012A (en) * | 2018-03-12 | 2018-08-21 | 株洲联诚集团控股股份有限公司 | A kind of fan noise prediction technique of optimization neural network |
Non-Patent Citations (5)
Title |
---|
Gamma Test 噪声估计的Kalman 神经网络在动态工业过程建模中的应用;李太福;《机械工程学报》;20140930;第50卷(第18期);第29-35页 * |
Smooth regression to estimate effective porosity using seismic attributes;Ursula Iturrarán-Viveros;《Journal of Applied Geophysics》;20111231;第76卷;第1-12页 * |
基于VRGIS的城市噪声三维分析模型及可视化评价研究;黄宝香;《中国博士学位论文全文数据库 (工程科技Ⅰ辑)》;20150715(第7期);全文 * |
基于数据挖掘的机场噪声预测方法的研究;尤华;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20130415(第04期);全文 * |
组合预测方法及其应用研究;马涛;《中国博士学位论文全文数据库 (信息科技辑)》;20180115(第1期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN110674996A (en) | 2020-01-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110674996B (en) | Urban traffic noise prediction method | |
CN110097755B (en) | Highway traffic flow state identification method based on deep neural network | |
CN108647583B (en) | Face recognition algorithm training method based on multi-target learning | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
CN109934269B (en) | Open set identification method and device for electromagnetic signals | |
CN109360604B (en) | Ovarian cancer molecular typing prediction system | |
CN111400540B (en) | Singing voice detection method based on extrusion and excitation residual error network | |
CN108766464B (en) | Digital audio tampering automatic detection method based on power grid frequency fluctuation super vector | |
CN112085254A (en) | Prediction method and model based on multi-fractal cooperative measurement gating cycle unit | |
CN107609588A (en) | A kind of disturbances in patients with Parkinson disease UPDRS score Forecasting Methodologies based on voice signal | |
CN113140254A (en) | Meta-learning drug-target interaction prediction system and prediction method | |
CN112612820A (en) | Data processing method and device, computer readable storage medium and processor | |
CN114398611A (en) | Bimodal identity authentication method, device and storage medium | |
CN112289391B (en) | Anode aluminum foil performance prediction system based on machine learning | |
CN113392958B (en) | Parameter optimization and application method and system of fuzzy neural network FNN | |
CN115936773A (en) | Internet financial black product identification method and system | |
CN112115754A (en) | Short-term traffic flow prediction model based on firework differential evolution hybrid algorithm-extreme learning machine | |
CN116452904A (en) | Image aesthetic quality determination method | |
CN112382382B (en) | Cost-sensitive integrated learning classification method and system | |
CN114997366A (en) | Protein structure model quality evaluation method based on graph neural network | |
CN114821322A (en) | Small sample remote sensing image classification method and system based on attention mechanism | |
CN113536760A (en) | Rumor-rumor article matching method and system guided by introduction sentences and rumor-avoiding pattern sentences | |
CN110852178A (en) | Piano music score difficulty identification method based on decision tree lifting | |
CN113035363B (en) | Probability density weighted genetic metabolic disease screening data mixed sampling method | |
CN112529403B (en) | Method for determining construction land area influence factor weight value by using neural network algorithm |
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 |