CN113688908B - Bluetooth indoor propagation model correction method based on twin support vector regression machine - Google Patents
Bluetooth indoor propagation model correction method based on twin support vector regression machine Download PDFInfo
- Publication number
- CN113688908B CN113688908B CN202110983720.4A CN202110983720A CN113688908B CN 113688908 B CN113688908 B CN 113688908B CN 202110983720 A CN202110983720 A CN 202110983720A CN 113688908 B CN113688908 B CN 113688908B
- Authority
- CN
- China
- Prior art keywords
- sample
- equation
- updated
- shift
- represented
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Complex Calculations (AREA)
Abstract
The invention relates to a Bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine. The method reduces the time for updating the model, improves the real-time performance of correcting the indoor propagation model of the Bluetooth signal, has higher prediction precision, is more suitable for the deployment of terminal equipment with limited memory, and provides an effective method for correcting the indoor propagation model on site by using portable terminal equipment.
Description
Technical Field
The invention relates to the technical field of Bluetooth indoor positioning, in particular to a Bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine.
Background
At present, along with the rapid development of the internet of things and mobile communication technology, the demand for indoor accurate positioning is also increasing. The complexity of the indoor environment causes significant interference and impact on the propagation process of the wireless signals, and thus GPS technology cannot be used for indoor positioning. Based on ranging algorithm and ranging independent algorithm are two common types of indoor positioning algorithm. The method has certain requirements on hardware based on a ranging algorithm but has higher positioning accuracy, and mainly represents RSSI, TOA, TDOA and AOA.
Currently, various RSSI-based indoor positioning algorithms generally use some common indoor propagation models to fit the relationship between the distance and the bluetooth signal strength RSSI, and model parameters in the positioning algorithms are often derived from empirical values or fit to the overall environment. However, factors such as multi-path propagation of radio signals and nonlinear time-varying characteristics introduced by the motion of positioning nodes or surrounding scatterers are main characteristics of a mobile communication channel, and the characteristics are not only main reasons influencing communication quality, but also main factors causing fitting errors of indoor propagation models. Therefore, the interference of factors such as complicated indoor environment layout and narrow space on the RSSI measurement is not fully considered from the empirical value or the model parameter fitting to the whole environment, so that the final positioning accuracy is limited and the method is difficult to adapt to different scenes.
Therefore, in order to improve the accuracy, a signal transmitter and a signal receiver need to be erected for signal simulation test, and collected data is input into corresponding tool software for data processing, model correction and signal prediction. The existing equipment is large in size, the testing equipment and the data processing and analyzing equipment are independent and need to be processed separately, results cannot be obtained in real time in the testing process, and the requirement on the capability of a design engineer is high, so that the working efficiency is low, and a large amount of manpower and material resources need to be consumed.
Disclosure of Invention
Therefore, the technical problem to be solved by the present invention is to overcome the above technical problems, and to provide a bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine, which is based on the online epsilon type twin support vector regression machine and improves the efficiency of bluetooth signal indoor propagation model correction by saving effective information in the training process.
In order to solve the technical problem, the invention provides a bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine, which comprises the following steps:
s10: a signal receiver is used for collecting the Bluetooth signal strength RSSI of a signal transmitter for multiple times at the kth sampling point, and mean value filtering is carried out to obtain a logarithmic distance loss model r of the Bluetooth signal strength RSSI k =RSSI(d k )=RSSI(d 0 )+10wlg(d k /d 0 ) Wherein d is k Denotes the distance of the sample point from the signal transmitter, the subscript k denotes the number, RSSI (d) k ) Representing a distance d to the signal transmitter k Signal receiver of (2) collected Bluetooth signal strength, d 0 Representing the close distance to the signal transmitter, w represents the weight of the model, let d 0 =1m,y k =r k ,x k =10lg(d k ) B = RSSI (1), yielding a linear model y k =wx k + b, training sample is (x) k ,y k );
S20: and comparing the root mean square error of the training value and the predicted value of the y value in the training sample with a set threshold, if the root mean square error of the training value and the predicted value of the y value is greater than or equal to the threshold, utilizing the training sample to correct and update the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression, and going to the step S10, if the root mean square error of the training value and the predicted value of the y value is less than the threshold, stopping correcting and updating the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
let the set of training samples be T, which is divided into S 1 、R 1 、E 1 Three sets of where S 1 Set of vectors for boundary support of lower-bound insensitive function, R 1 To reserve a set of support vectors, E 1 A set of vectors is supported for errors.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
t is further divided into S 2 、R 2 、E 2 Three sets of where S 2 Supporting a set of vectors for the boundaries of upper-bound insensitive functions, R 2 To reserve a set of support vectors, E 2 A set of vectors is supported for errors.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
s21: reading training samples (x) k ,y k );
S22: when k =1, c is set 1 、c 2 、c 3 、ε 1 And epsilon 2 A value of (b), wherein c 1 、c 2 Is a penalty constant, c 3 Is a regularization constant, ε 1 、ε 2 Is an insensitive constant; initializing linear model parametersLet b 1 =b 2 =w 1 =w 2 =0, wherein b 1 、b 2 To be offset, w 1 、w 2 Is a weight; then, the initial M is calculated from the formula (1) -1 ;
When k ≠ 1, M is updated by equation (2) -1 When is coming into contact withWhen it is time, V is updated by equation (3) 1 When is coming into contact withWhen the V is updated by the formula (4) 2 ;
Wherein, g k =[1,x k ]I is a unit matrix of corresponding dimension;
M -1 =M -1 -ZZ T /J (2)
wherein, the first and the second end of the pipe are connected with each other,J=1+g k Z, is a row vector g i ,i∈S 1 A composed matrix, Z 1 ′=V 1 Z 1 , Is a row vector g i ,i∈S 2 A composed matrix, Z 2 ′=V 2 Z 2 ,
S23: calculation of α from equations (5) and (6) k And gamma k :
α k =y k -(L T L) -1 L T l 1 (5)
γ k =(L T L) -1 L T l 2 -y k (6)
Wherein alpha is k And gamma k Respectively represent training samples (x) k ,y k ) Lagrange multipliers in the lower bound insensitive function and the upper bound insensitive function,G=[e,X]e is a column vector with the corresponding dimension elements all being 1, X = [ X = 1 ,x 2 ,...,x k-1 ] T ,l 1 =G(ZZ T /J)G T (Y-α),l 2 =G(ZZ T /J)G T (Y+γ),Y=[y 1 ,y 2 ,...,y k-1 ] T Alpha is a group of i I ∈ T, a column vector consisting of γ i I belongs to a column vector consisting of T;
When the time is over, the time is updated by the formula (7) -formula (9) Is formed by alpha i ,i∈S 1 A column vector of components; when in useWhen the time is over, the time is updated by the formula (10) -formula (12) Is formed by gamma i ,i∈S 2 A column vector of components;
u 1 =M -1 G T (Y-α) (7)
H 1 =Y-(Gu 1 -ε 1 e) (8)
u 2 =M -1 G T (Y+γ) (10)
H 2 =(Gu 2 +ε 2 e)-Y (11)
wherein H 1 A column vector H representing the boundary distance components from the training samples in the T set to the lower-bound insensitive function 2 A column vector consisting of the boundary distances from the training samples in the T set to the upper-bound insensitive function,is S 1 A column vector composed of the boundary distances corresponding to the training samples in the set,is S 2 A column vector consisting of boundary distances corresponding to training samples in the set;
s25: recalculating H from equation (7) -equation (8) 1 Then screening the abnormal sample set F 1 ,t 1(i) Is represented by F 1 A target value adjusted by a Lagrange multiplier of an ith sample in the set;
for S 1 Samples in the set, if α i Not more than 0, transferred into F 1 Set and t 1(i) =c 1 (ii) a If α is i ≥c 1 Is moved into F 1 Set and t 1(i) =0, and the inverse matrix V needs to be updated by equation (13) 1 ;
For R 1 Sample in the set, if H 1(i) Less than or equal to 0, transferring into F 1 Set and t 1(i) =c 1 ;
For E 1 Sample in the set, if H 1(i) Not less than 0, shift-in F 1 Set and t 1(i) =0;
For training sample (x) k ,y k ) If H is H 1(k) ≥0,α k If not less than 0, then move into R 1 Gathering; if H 1(k) =0,0<α k <c 1 Then move into S 1 The inverse matrix V is collected and updated by equation (14) 1 (ii) a If H 1(k) ≤0,α k =c 1 Then move into E 1 Gathering; otherwise, when alpha is k Not more than 0, transferred into F 1 Set and t 1(k) =c 1 (ii) a When alpha is k ≥c 1 Is moved into F 1 Set and t 1(k) =0; when 0 < alpha k <c 1 If H is H 1(k) < 0, move into F 1 Set and t 1(k) =c 1 If H is H 1(k) > 0, into F 1 Set and t 1(k) =0;
Wherein the subscript t denotes a shift out of S 1 The samples of the set, the subscript \ tt representing the elements of the t rows and t columns of the deletion matrix, the subscript x representing the full index of the row or column;
s26: recalculating H from equation (10) -equation (11) 2 Then screening the abnormal sample set F 2 ,t 2(i) Is represented by F 2 The target value of the lagrange multiplier adjustment of the ith sample;
for S 2 Samples in the set, if γ i Not more than 0, transferred into F 2 Set and t 2(i) =c 2 (ii) a If gamma is i ≥c 2 Is moved into F 2 Set and t 2(i) =0, and the inverse matrix V needs to be updated by equation (15) 2 ;
For R 2 Samples in the set, if H 2(i) Not more than 0, transferred into F 2 Set and t 2(i) =c 2 ;
For E 2 Samples in the set, if H 2(i) Not less than 0, shift-in F 2 Set and t 2(i) =0;
For sample (x) k ,y k ) If H is H 2(k) ≥0,γ k If not less than 0, then move into R 2 Gathering; if H 2(k) =0,0<γ k <c 2 Then move into S 2 The inverse matrix V is collected and updated by equation (16) 2 (ii) a If H 2(k) ≤0,γ k =c 2 Then move into E 2 Gathering; otherwise, when gamma is k Not more than 0, transferred into F 2 Set and t 2(k) =c 2 (ii) a When gamma is k ≥c 2 Is moved into F 2 Set and t 2(k) =0; when 0 < gamma k <c 2 If H is H 2(k) < 0, move into F 2 Set and t 2(k) =c 2 If H is H 2(k) > 0, into F 2 Set and t 2(k) =0;
Wherein the subscript t denotes a shift out of S 2 A sample of the collection;
s27: from the respective equations (17) and (18)Then, the maximum step length eta is calculated from the formula (19) to the formula (26) max ;
If it is usedThen the corresponding sample is represented by F 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
if it is notThen the corresponding sample is represented by S 1 Set shift into R 1 Set or E 1 Set, determined by Lagrange multiplier of corresponding sample, and update V by equation (13) 1 ;
If it is notThen the corresponding sample is represented by R 1 Set migration S 1 Is aggregated and V is updated by equation (14) 1 ;
If it is notThen the corresponding sample is represented by E 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
wherein h is 1 (i),t 1(i) And alpha i Are each H 1 ,t 1 And the ith element of α; andare respectively F 1 The ith sample in the set is shifted into S 1 Set of R 1 Set sum E 1 Step size of the set;is S 1 Moving the ith sample in the set into R 1 Set or E 1 A step size of the set;is R 1 Set shift into S 1 Step size of the set;is E 1 Set shift into S 1 Step size of the set;andare respectively composed ofAnda column vector of components;
s28: from the respective equations (27) and (28)Then, equation (29) -equation (36) calculates the maximum step η' max ;
If it is notThen the corresponding sample is represented by F 2 Set shift into S 2 Is aggregated and V is updated by equation (16) 2 ;
if it is notThen the corresponding sample is represented by S 2 Set shift into R 2 Set or E 2 Set, determined by the Lagrange multiplier of the corresponding sample, and update V by equation (15) 2 ;
If it is usedThen the corresponding sample is represented by R 2 Set shift into S 2 Is aggregated and V is updated by equation (16) 2 ;
If it is notThen the corresponding sample is represented by E 2 Set shift into S 2 Is collected and V is updated by equation (16) 2 ;
wherein h is 2 (i),t 2(i) And gamma i Are each H 2 ,d 2 And the ith element of γ; andare respectively F 2 The ith sample in the set is shifted into S 2 Set of R 2 Set sum E 2 Step size of the set;is S 2 Moving the ith sample in the set into R 2 Set or E 2 A step size of the set;is R 2 Set shift into S 2 Step size of the set;is E 2 Set shift into S 2 Step size of the set;andare respectively composed of Anda column vector of components;
s29: calculating the solution u of the updated model from equations (7) and (10) 1 And u 2 And store M -1 Parameter u of lower bound insensitive function 1 ,α,V 1 Set R 1 、S 1 、E 1 And the parameter u of the upper bound insensitive function 2 ,γ,V 2 Set R 2 、S 2 、E 2 。
Compared with the prior art, the technical scheme of the invention has the following advantages:
the invention saves the effective information in the training process, gradually and iteratively adjusts the Lagrange multiplier of the training sample in each updating process, so that the Lagrange multiplier meets the KKT condition, the model updating speed in the correction process is improved, the method is more suitable for the deployment of terminal equipment with limited memory, and an effective method is provided for the field correction of the indoor propagation model by using the portable terminal equipment.
Drawings
In order that the present invention may be more readily and clearly understood, reference will now be made in detail to the present invention, examples of which are illustrated in the accompanying drawings.
FIG. 1 is a flow chart of a Bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine.
FIG. 2 is a graph of a predicted result of a Bluetooth signal indoor propagation model based on an online epsilon type twin support vector regression machine.
FIG. 3 is a diagram of predicted output residuals of a Bluetooth signal indoor propagation model based on an online epsilon type twin support vector regression.
Detailed Description
The present invention is further described below in conjunction with the following figures and specific examples so that those skilled in the art may better understand the present invention and practice it, but the examples are not intended to limit the present invention.
Referring to fig. 1, the present invention provides a bluetooth signal indoor propagation model calibration method based on an online epsilon type twin support vector regression, including the following steps:
s10: collecting the Bluetooth signal strength RSSI of the signal transmitter for multiple times by using a signal receiver at the kth sampling point, and carrying out mean value filtering to obtain a logarithmic distance loss model r of the Bluetooth signal strength RSSI k =RSSI(d k )=RSSI(d 0 )+10wlg(d k /d 0 ) Wherein d is k Indicating the distance of the sample point from the signal transmitter, subscript k indicating the number, RSSI (d) k ) Representing a distance d to the signal transmitter k Signal receiver of (2) collected Bluetooth signal strength, d 0 Representing the close distance to the signal transmitter, w represents the weight of the model, let d 0 =1m,y k =r k ,x k =10lg(d k ) B = RSSI (1), yielding a linear model y k =wx k + b, training sample is (x) k ,y k );
S20: and comparing the root mean square error of the training value and the predicted value of the y value in the training sample with a set threshold, if the root mean square error of the training value and the predicted value of the y value is greater than or equal to the threshold, utilizing the training sample to correct and update the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression, and going to the step S10, if the root mean square error of the training value and the predicted value of the y value is less than the threshold, stopping correcting and updating the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
let the set of training samples be T, which is divided into S 1 、R 1 、E 1 Three sets of where S 1 Set of vectors for boundary support of lower-bound insensitive function, R 1 To reserve a set of support vectors, E 1 A set of vectors is supported for errors.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
t is further divided into S 2 、R 2 、E 2 Three sets of where S 2 Supporting a set of vectors for the boundaries of upper-bound insensitive functions, R 2 To reserve a set of support vectors, E 2 A set of vectors is supported for errors.
In one embodiment of the invention, the updating, by using the training sample, the correction of the bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression comprises the following steps:
s21: reading training samples (x) k ,y k );
S22: when k =1, c is set 1 、c 2 、c 3 、ε 1 And ε 2 A value of (b), wherein c 1 、c 2 Is a penalty constant, c 3 Is a regularization constant, ε 1 、ε 2 Is an insensitive constant; initialize the parameters of the linear model, let b 1 =b 2 =w 1 =w 2 =0, wherein b 1 、b 2 To be biased, w 1 、w 2 Is a weight; then, the initial M is calculated from the formula (1) -1 ;
When k ≠ 1, M is updated by equation (2) -1 When is coming into contact withWhen it is time, V is updated by equation (3) 1 When is coming into contact withWhen it is time, V is updated by equation (4) 2 ;
Wherein, g k =[1,x k ]I is a unit matrix of corresponding dimension;
M -1 =M -1 -ZZ T /J (2)
wherein, the first and the second end of the pipe are connected with each other,J=1+g k Z, is a row vector g i ,i∈S 1 A composed matrix, Z 1 ′=V 1 Z 1 , Is a row vector g i ,i∈S 2 A composed matrix, Z 2 ′=V 2 Z 2 ,
S23: calculation of α from equations (5) and (6) k And gamma k :
α k =y k -(L T L) -1 L T l 1 (5)
γ k =(L T L) -1 L T l 2 -y k (6)
Wherein alpha is k And gamma k Respectively represent training samples (x) k ,y k ) Lagrange multipliers in the lower bound insensitive function and the upper bound insensitive function,G=[e,X]e is a column vector with the corresponding dimension elements all being 1, X = [ X = 1 ,x 2 ,...,x k-1 ] T ,l 1 =G(ZZ T /J)G T (Y-α),l 2 =G(ZZ T /J)G T (Y+γ),Y=[y 1 ,y 2 ,...,y k-1 ] T Alpha is a group of i I ∈ T, a column vector consisting of γ i I belongs to a column vector consisting of T;
s24: let alpha = [ alpha ] T ,α k ] T ,γ=[γ T ,γ k ] T ,Y=[Y T ,y k ] T T = T { k }; when in useWhen the time is over, the time is updated by the formula (7) -formula (9) Is formed by alpha i ,i∈S 1 A column vector of components; when in useWhen the time is over, the time is updated by the formula (10) -formula (12) Is formed by gamma i ,i∈S 2 A column vector of components;
u 1 =M -1 G T (Y-α) (7)
H 1 =Y-(Gu 1 -ε 1 e) (8)
u 2 =M -1 G T (Y+γ) (10)
H 2 =(Gu 2 +ε 2 e)-Y (11)
wherein H 1 A column vector H representing the boundary distance components from the training samples in the T set to the lower-bound insensitive function 2 A column vector consisting of the boundary distances from the training samples in the T set to the upper-bound insensitive function,is S 1 A column vector composed of the boundary distances corresponding to the training samples in the set,is S 2 A column vector consisting of boundary distances corresponding to training samples in the set;
s25: recalculating H from equation (7) -equation (8) 1 Then screening an abnormal sample set F 1 ,t 1(i) Is represented by F 1 A target value adjusted by a Lagrange multiplier of an ith sample in the set;
for S 1 Samples in the set, if α i Not more than 0, transferred into F 1 Set and t 1(i) =c 1 (ii) a If α is i ≥c 1 Is moved into F 1 Set and t 1(i) =0, and the inverse matrix V needs to be updated by equation (13) 1 ;
For R 1 Sample in the set, if H 1(i) Not more than 0, transferred into F 1 Set and t 1(i) =c 1 ;
For E 1 Sample in the set, if H 1(i) Not less than 0, shift-in F 1 Set and t 1(i) =0;
For training sample (x) k ,y k ) H if H 1(k) ≥0,α k If not less than 0, then move into R 1 Gathering; if H 1(k) =0,0<α k <c 1 Then move into S 1 Inverse integration and updating by equation (14)Matrix V 1 (ii) a If H 1(k) ≤0,α k =c 1 Then move into E 1 Gathering; otherwise, when alpha is k Not more than 0, transferred into F 1 Set and t 1(k) =c 1 (ii) a When alpha is k ≥c 1 Is moved into F 1 Set and t 1(k) =0; when 0 < alpha k <c 1 If H is H 1(k) < 0, move into F 1 Set and t 1(k) =c 1 If H is H 1(k) > 0, into F 1 Set and t 1(k) =0;
Wherein the subscript t denotes the removal of S 1 The samples of the set, the subscript \ tt representing the elements of the t rows and t columns of the deletion matrix, the subscript x representing the full index of the row or column;
s26: recalculating H from equation (10) -equation (11) 2 Then screening the abnormal sample set F 2 ,t 2(i) Is represented by F 2 The target value of the lagrange multiplier adjustment of the ith sample;
for S 2 Samples in the set, if γ i Not more than 0, transferred into F 2 Set and t 2(i) =c 2 (ii) a If gamma is i ≥c 2 Is moved into F 2 Set and t 2(i) =0, and the inverse matrix V needs to be updated by equation (15) 2 ;
For R 2 Samples in the set, if H 2(i) Not more than 0, transferred into F 2 Set and t 2(i) =c 2 ;
For E 2 In a collectionIf H is a sample of 2(i) Not less than 0, shift-in F 2 Set and t 2(i) =0;
For sample (x) k ,y k ) If H is H 2(k) ≥0,γ k If not less than 0, then move into R 2 Gathering; if H 2(k) =0,0<γ k <c 2 Then move into S 2 The inverse matrix V is collected and updated by equation (16) 2 (ii) a If H 2(k) ≤0,γ k =c 2 Then move into E 2 Gathering; otherwise, when γ k Not more than 0, transferred into F 2 Set and t 2(k) =c 2 (ii) a When gamma is k ≥c 2 Is moved into F 2 Set and t 2(k) =0; when 0 < gamma k <c 2 If H is H 2(k) < 0, move into F 2 Set and t 2(k) =c 2 H if H 2(k) > 0, into F 2 Set and t 2(k) =0;
Wherein the subscript t denotes a shift out of S 2 A sample of the collection;
s27: from the respective equations (17) and (18)Then, the maximum step length eta is calculated from the formula (19) to the formula (26) max ;
If it is notThen the corresponding sample is represented by F 1 Set shift into S 1 Are combined and represented by(14) Update V 1 ;
if it is notThen the corresponding sample is represented by S 1 Set shift into R 1 Set or E 1 Set, determined by the Lagrange multiplier of the corresponding sample, and update V by equation (13) 1 ;
If it is notThen the corresponding sample is represented by R 1 Set migration S 1 Is aggregated and V is updated by equation (14) 1 ;
If it is notThen the corresponding sample is represented by E 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
wherein h is 1 (i),t 1(i) And alpha i Are each H 1 ,t 1 And the ith element of α; andare respectively F 1 The ith sample in the set is shifted into S 1 Set of R 1 Set and E 1 Step size of the set;is S 1 Moving the ith sample in the set into R 1 Set or E 1 Step size of the set;is R 1 Set migration S 1 Step size of the set;is E 1 Set shift into S 1 A step size of the set;andare respectively composed ofAnda column vector of components;
s28: from the respective equations (27) and (28)Then, equation (29) -equation (36) calculates the maximum step η' max ;
If it is usedThen the corresponding sample is represented by F 2 Set shift into S 2 Is aggregated and V is updated by equation (16) 2 ;
if it is notThen the corresponding sample is represented by S 2 Set shift into R 2 Set or E 2 Set, determined by Lagrange multiplier of corresponding sample, and update V by equation (15) 2 ;
If it is notThen the corresponding sample is represented by R 2 Set shift into S 2 Is collected and V is updated by equation (16) 2 ;
If it is usedThen the corresponding sample is represented by E 2 Set shift into S 2 Is aggregated and V is updated by equation (16) 2 ;
wherein h is 2 (i),t 2(i) And gamma i Are each H 2 ,d 2 And the ith element of γ; andare respectively F 2 The ith sample in the set is shifted into S 2 Set of R 2 Set sum E 2 Step size of the set;is S 2 The ith sample in the set is moved into R 2 Set or E 2 Step size of the set;is R 2 Set shift into S 2 Step size of the set;is E 2 Set shift into S 2 Step size of the set;andare respectively composed of Anda column vector of components;
s29: the solution u of the updated model is calculated from equations (7) and (10) 1 And u 2 And save M -1 Parameter u of lower bound insensitive function 1 ,α,V 1 Set R 1 、S 1 、E 1 And the parameter u of the upper bound insensitive function 2 ,γ,V 2 Set R 2 、S 2 、E 2 。
The invention saves the effective information in the training process, gradually and iteratively adjusts the Lagrange multiplier of the training sample in each updating process to ensure that the Lagrange multiplier meets the KKT condition, improves the updating speed of the model in the correction process, is more suitable for the deployment of terminal equipment with limited memory, and provides an effective method for the field correction of the indoor propagation model by using the portable terminal equipment.
The following describes in detail a bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression provided by the present invention with an embodiment.
The effectiveness of the invention is explained by adopting the method for correcting the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression machine and combining a specific example of correcting the Bluetooth signal propagation model between a transmitting device and a receiving device. Fixing a Bluetooth signal transmitting device, then selecting 50 different sampling points from near to far, acquiring 10 RSSI values at each sampling point by using a Bluetooth signal receiving device to carry out mean filtering to obtain the more stable RSSI value of the sampling point, recording the distance between the Bluetooth signal transmitting device and the receiving device, thereby obtaining 50 data samples, and selecting 40 samples as training samples and 10 samples as testing samples. The specific implementation mode is as follows:
initializing various parameters, setting C 1 =C 2 =2 10 ,ε 1 =ε 2 =0.01,C 3 =2 -5 And correcting the indoor propagation model of the Bluetooth signal in an online updating mode, then storing the corrected model parameters, and directly substituting the corrected model parameters into test data to obtain the predicted values of the RSSI (received signal strength indicator) of the Bluetooth signals at different distances.
As can be seen from the attached figures 2 and 3, the RSSI value of the Bluetooth signal strength predicted by the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression machine has better prediction accuracy.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications of the invention may be made without departing from the spirit or scope of the invention.
Claims (1)
1. A Bluetooth signal indoor propagation model correction method based on an online epsilon type twin support vector regression machine is characterized by comprising the following steps:
s10: a signal receiver is used for collecting the Bluetooth signal strength RSSI of a signal transmitter for multiple times at the kth sampling point, and mean value filtering is carried out to obtain a logarithmic distance loss model r of the Bluetooth signal strength RSSI k =RSSI(d k )=RSSI(d 0 )+10wlg(d k /d 0 ) Wherein d is k Denotes the distance of the sample point from the signal transmitter, the subscript k denotes the number, RSSI (d) k ) Representing a distance d to the signal transmitter k Signal receiver of (2) collected Bluetooth signal strength, d 0 Representing the close distance to the signal transmitter, w represents the weight of the model, let d 0 =1m,y k =r k ,x k =10lg(d k ) B = RSSI (1), yielding a linear model y k =wx k + b, training sample is (x) k ,y k );
S20: comparing the root mean square error of the training value and the predicted value of the y value in the training sample with a set threshold, if the root mean square error of the training value and the predicted value of the y value is greater than or equal to the threshold, utilizing the training sample to correct and update the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression machine, and going to the step S10, if the root mean square error of the training value and the predicted value of the y value is less than the threshold, stopping correcting and updating the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression machine;
let the set of training samples be T, which is divided into S 1 、R 1 、E 1 Three sets of where S 1 Set of vectors for boundary support of lower-bound insensitive function, R 1 To reserve a set of support vectors, E 1 Supporting a set of vectors for errors; t is further divided into S 2 、R 2 、E 2 Three sets of where S 2 Supporting a set of vectors for the boundaries of upper-bound insensitive functions, R 2 To reserve a set of support vectors, E 2 Supporting a set of vectors for errors;
the method for correcting and updating the Bluetooth signal indoor propagation model based on the online epsilon type twin support vector regression by using the training samples comprises the following steps:
s21: reading training samples (x) k ,y k );
S22: when k =1, c is set 1 、c 2 、c 3 、ε 1 And ε 2 A value of (b), wherein c 1 、c 2 Is a penalty constant, c 3 Is a regularization constant, ε 1 、ε 2 Is an insensitive constant; initialize the parameters of the linear model, let b 1 =b 2 =w 1 =w 2 =0, wherein b 1 、b 2 To be biased, w 1 、w 2 Is a weight; then, the initial M is calculated from the formula (1) -1 ;
When k ≠ 1, M is updated by equation (2) -1 When is coming into contact withWhen the V is updated by the formula (3) 1 When is coming into contact withWhen the V is updated by the formula (4) 2 ;
Wherein, g k =[1,x k ]I is a unit matrix of the corresponding dimension;
M -1 =M -1 -ZZ T /J (2)
V 1 =V 1 +Z′ 1 Z′ 1 T /J′ 1 (3)
V 2 =V 2 +Z′ 2 Z′ 2 T /J′ 2 (4)
wherein the content of the first and second substances,J=1+g k Z, is a row vector g i ,i∈S 1 Matrix of composition, Z' 1 =V 1 Z 1 , Is a row vector g i ,i∈S 2 Matrix of composition, Z' 2 =V 2 Z 2 ,
S23: byFormula (5) and formula (6) calculate α k And gamma k :
α k =y k -(L T L) -1 L T l 1 (5)
γ k =(L T L) -1 L T l 2 -y k (6)
Wherein alpha is k And gamma k Respectively represent training samples (x) k ,y k ) Lagrange multipliers in the lower bound insensitive function and the upper bound insensitive function,e is a column vector with all 1's of the corresponding dimension elements, X = [ X = 1 ,x 2 ,...,x k-1 ] T ,l 1 =G(ZZ T /J)G T (Y-α),l 2 =G(ZZ T /J)G T (Y+γ),Y=[y 1 ,y 2 ,...,y k-1 ] T Alpha is a group of i I ∈ T, a column vector consisting of γ i I belongs to a column vector consisting of T;
s24: let alpha = [ alpha ] T ,α k ] T ,γ=[γ T ,γ k ] T ,Y=[Y T ,y k ] T T = T { k }; when in useWhen the time is over, the time is updated by the formula (7) -formula (9) Is formed by alpha i ,i∈S 1 A column vector of components; when the temperature is higher than the set temperatureWhen the time is over, the time is updated by the formula (10) -formula (12) Is formed by gamma i ,i∈S 2 A column vector of components;
u 1 =M -1 G T (Y-α) (7)
H 1 =Y-(Gu 1 -ε 1 e) (8)
u 2 =M -1 G T (Y+γ) (10)
H 2 =(Gu 2 +ε 2 e)-Y (11)
wherein H 1 A column vector H composed of boundary distances from the training samples in the T set to the lower-bound insensitive function 2 A column vector consisting of the boundary distances from the training samples in the T set to the upper-bound insensitive function,is S 1 A column vector composed of the boundary distances corresponding to the training samples in the set,is S 2 A column vector consisting of boundary distances corresponding to training samples in the set;
s25: is represented by formula (7) -formula(8) Recalculating H 1 Then screening an abnormal sample set F 1 ,t 1(i) Is represented by F 1 The target value of the Lagrange multiplier adjustment of the ith sample in the set;
for S 1 Samples in the set, if α i Not more than 0, transferred into F 1 Set and t 1(i) =c 1 (ii) a If α is i ≥c 1 Is moved into F 1 Set and t 1(i) =0, and the inverse matrix V needs to be updated by equation (13) 1 ;
For R 1 Samples in the set, if H 1(i) Not more than 0, transferred into F 1 Set and t 1(i) =c 1 ;
For E 1 Samples in the set, if H 1(i) Not less than 0, shift-in F 1 Set and t 1(i) =0;
For training sample (x) k ,y k ) If H is present 1(k) ≥0,α k =0, then move into R 1 Gathering; if H is 1(k) =0,0<α k <c 1 Then move into S 1 The inverse matrix V is collected and updated by equation (14) 1 (ii) a If H is present 1(k) ≤0,α k =c 1 Then move into E 1 Gathering; otherwise, when alpha is k Less than or equal to 0, transferring into F 1 Set and t 1(k) =c 1 (ii) a When alpha is k ≥c 1 Is moved into F 1 Set and t 1(k) =0; when 0 < alpha k <c 1 If H is present 1(k) < 0, move into F 1 Set and t 1(k) =c 1 If H is present 1(k) > 0, transfer into F 1 Set and t 1(k) =0;
Wherein the subscript t denotes a shift out of S 1 The samples of the set, the subscript \ tt representing the elements of the t rows and t columns of the deletion matrix, the subscript x representing the full index of the row or column;
s26: recalculating H from equation (10) -equation (11) 2 Then screening the abnormal sample set F 2 ,t 2(i) Is represented by F 2 The target value of the lagrange multiplier adjustment of the ith sample;
for S 2 Samples in the set, if γ i Not more than 0, transferred into F 2 Set and t 2(i) =c 2 (ii) a If gamma is equal to i ≥c 2 Is moved into F 2 Set and t 2(i) =0, and the inverse matrix V needs to be updated by equation (15) 2 ;
For R 2 Samples in the set, if H 2(i) Not more than 0, transferred into F 2 Set and t 2(i) =c 2 ;
For E 2 Samples in the set, if H 2(i) Not less than 0, shift-in F 2 Set and t 2(i) =0;
For sample (x) k ,y k ) If H is 2(k) ≥0,γ k If not less than 0, then move into R 2 Gathering; if H is present 2(k) =0,0<γ k <c 2 Then move into S 2 The inverse matrix V is collected and updated by equation (16) 2 (ii) a If H is present 2(k) ≤0,γ k =c 2 Then move into E 2 Gathering; otherwise, when γ k Not more than 0, transferred into F 2 Set and t 2(k) =c 2 (ii) a When gamma is equal to k ≥c 2 Is moved into F 2 Set and t 2(k) =0; when 0 < gamma k <c 2 If H is present 2(k) < 0, move into F 2 Set and t 2(k) =c 2 If H is present 2(k) > 0, into F 2 Set and t 2(k) =0;
Wherein the subscript t denotes the removal of S 2 A sample of the collection;
s27: from the respective equations (17) and (18)Then, the maximum step length eta is calculated from the formula (19) to the formula (26) max ;
If it is notThen the corresponding sample is represented by F 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
if it is usedThen the corresponding sample is represented by S 1 Set shift into R 1 Set or E 1 Set, determined by Lagrange multiplier of corresponding sample, and updated by equation (13)V 1 ;
If it is notThen the corresponding sample is represented by R 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
If it is notThen the corresponding sample is represented by E 1 Set shift into S 1 Is aggregated and V is updated by equation (14) 1 ;
wherein h is 1 (i),t 1(i) And alpha i Are each H 1 ,t 1 And the ith element of α; andare respectively F 1 The ith sample in the set is shifted into S 1 Set of R 1 Set sum E 1 A step size of the set;is S 1 Moving the ith sample in the set into R 1 Set or E 1 A step size of the set;is R 1 Set migration S 1 Step size of the set;is E 1 Set shift into S 1 Step size of the set;andare respectively composed ofAnda column vector of components;
s28: from the respective equations (27) and (28)Then, equation (29) -equation (36) calculates the maximum step η' max ;
If it is notThen the corresponding sample is represented by F 2 Set shift into S 2 Is collected and V is updated by equation (16) 2 ;
if it is usedThen the corresponding sample is represented by S 2 Set shift into R 2 Set or E 2 Set, determined by the Lagrange multiplier of the corresponding sample, and update V by equation (15) 2 ;
If it is notThen the corresponding sample is represented by R 2 Set shift into S 2 Is aggregated and V is updated by equation (16) 2 ;
If it is notThen the corresponding sample is represented by E 2 Set shift into S 2 Is collected and V is updated by equation (16) 2 ;
wherein h is 2 (i),t 2(i) And gamma i Are each H 2 ,d 2 And the ith element of γ; andare respectively F 2 The ith sample in the set is shifted into S 2 Set of R 2 Set sum E 2 Step size of the set;is S 2 The ith sample in the set is moved into R 2 Set or E 2 Step size of the set;is R 2 Set shift into S 2 Step size of the set;is E 2 Set shift into S 2 A step size of the set;andare respectively composed of Anda column vector of components;
s29: calculating the solution u of the updated model from equations (7) and (10) 1 And u 2 And store M -1 Parameter u of lower bound insensitive function 1 ,α,V 1 Set R 1 、S 1 、E 1 And the parameter u of the upper bound insensitive function 2 ,γ,V 2 Set R 2 、S 2 、E 2 。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110983720.4A CN113688908B (en) | 2021-08-25 | 2021-08-25 | Bluetooth indoor propagation model correction method based on twin support vector regression machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110983720.4A CN113688908B (en) | 2021-08-25 | 2021-08-25 | Bluetooth indoor propagation model correction method based on twin support vector regression machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113688908A CN113688908A (en) | 2021-11-23 |
CN113688908B true CN113688908B (en) | 2022-11-08 |
Family
ID=78582617
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110983720.4A Active CN113688908B (en) | 2021-08-25 | 2021-08-25 | Bluetooth indoor propagation model correction method based on twin support vector regression machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113688908B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109195104A (en) * | 2018-08-27 | 2019-01-11 | 上海市计量测试技术研究院 | A kind of indoor orientation method combined based on support vector regression and Kalman filtering |
CN109640261A (en) * | 2018-11-22 | 2019-04-16 | 天津理工大学 | A kind of location algorithm of the heterogeneous wireless sensor network based on support vector regression |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102184421A (en) * | 2011-04-22 | 2011-09-14 | 北京航空航天大学 | Training method of support vector regression machine |
CN107333238B (en) * | 2017-07-03 | 2020-06-30 | 杭州电子科技大学 | Indoor fingerprint rapid positioning method based on support vector regression |
CN110134088A (en) * | 2019-05-21 | 2019-08-16 | 浙江大学 | A kind of adaptive quality forecasting procedure based on increment support vector regression |
CN110908361B (en) * | 2019-12-03 | 2022-06-14 | 江南大学 | Fermentation process soft measurement method based on online twin support vector regression |
EP3854887A1 (en) * | 2020-01-23 | 2021-07-28 | Institut Jean Paoli & Irène Calmettes | In vitro method for identifying efficient therapeutic molecules for treating pancreatic ductal adenocarcinoma |
-
2021
- 2021-08-25 CN CN202110983720.4A patent/CN113688908B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109195104A (en) * | 2018-08-27 | 2019-01-11 | 上海市计量测试技术研究院 | A kind of indoor orientation method combined based on support vector regression and Kalman filtering |
CN109640261A (en) * | 2018-11-22 | 2019-04-16 | 天津理工大学 | A kind of location algorithm of the heterogeneous wireless sensor network based on support vector regression |
Also Published As
Publication number | Publication date |
---|---|
CN113688908A (en) | 2021-11-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111327377B (en) | Method, device, equipment and storage medium for field intensity prediction | |
CN100403054C (en) | Method and apparatus providing improved position estimate based on an initial coarse position estimate | |
CN109696698A (en) | Navigator fix prediction technique, device, electronic equipment and storage medium | |
CN110232445B (en) | Cultural relic authenticity identification method based on knowledge distillation | |
CN101572857B (en) | Locating method in wireless LAN and device thereof | |
CN110210100B (en) | High-precision sound velocity prediction method for sediment of seabed sediment | |
CN109362084B (en) | Method, apparatus, device and medium for communication service quality optimization | |
CN105956709B (en) | A kind of modularization support vector machines tide prediction method based on GUI | |
CN108303672A (en) | WLAN indoor positionings error correcting method based on location fingerprint and system | |
CN109874104B (en) | User position positioning method, device, equipment and medium | |
CN103368788A (en) | Information processing device, information processing method, and program | |
CN109981195B (en) | Method and device for processing wireless signal strength | |
CN104619009A (en) | Positioning data sampling period adjustment method and device and mobile terminal | |
CN111698695A (en) | LTE fingerprint type positioning method based on neural network | |
CN108401222B (en) | Positioning method and device | |
CN113688908B (en) | Bluetooth indoor propagation model correction method based on twin support vector regression machine | |
CN109065176B (en) | Blood glucose prediction method, device, terminal and storage medium | |
US10820152B2 (en) | Device diversity correction method for RSS-based precise location tracking | |
CN106717083A (en) | Method for position detection by mobile computing device, and mobile computing device performing same | |
CN112995893A (en) | Fingerprint positioning method, system, server and storage medium | |
CN117236515A (en) | Method for predicting urban street tree breast diameter growth trend, prediction system and electronic equipment | |
CN115543638B (en) | Uncertainty-based edge calculation data collection and analysis method, system and equipment | |
KR102500534B1 (en) | Recurrent neural network based water resource information generating device and method | |
CN113890833B (en) | Network coverage prediction method, device, equipment and storage medium | |
EP2335087A2 (en) | Upload and download of position reference data |
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 | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20221017 Address after: No. 6, Electromechanical First Branch Road, Changshengqiao Town, Economic Development Zone, Nan'an District, Chongqing 401336 Applicant after: Chongqing Tailewei Technology Co.,Ltd. Address before: 1800 No. 214122 Jiangsu city of Wuxi Province Li Lake Avenue Applicant before: Jiangnan University |
|
TA01 | Transfer of patent application right | ||
GR01 | Patent grant | ||
GR01 | Patent grant |