CN103533647B - A kind of radio frequency map adaptive location method based on sub-clustering mechanism and robustness regression - Google Patents

Abstract

The present invention relates to a kind of radio frequency map adaptive location method based on sub-clustering mechanism and robustness regression. First the method collects the signal characteristic of multiple reference points and check-node in a scene in off-line training step, sets up static radio frequency map; Then according to the static radio frequency map of this scene, calculate the path loss parameter of each reference point; According to path loss parameter, reference point is carried out to sub-clustering again, locating area is divided into multiple subregions; Finally in the RSSI fingerprint to every sub regions internal reference point and this subregion, the RSSI fingerprint of checkpoint carries out robust linear regression. Online positioning stage, utilizes checkpoint RSSI and the robustness regression coefficient update radio frequency map of timing acquisition, and then the K neighbor algorithm for weighting by this radio frequency map, locates thereby realize. This method is simple, and positioning precision is high, and RSSI randomized jitter and the indoor occupant factors such as interference of walking about that can effectively reduce cause the out-of-date impact that location Calculation is caused of radio frequency map.

Description

A kind of radio frequency map adaptive location method based on sub-clustering mechanism and robustness regression

Technical field

The present invention relates to indoor wireless field of locating technology, particularly a kind of radio frequency based on sub-clustering mechanism and robustness regressionMap adaptive location method.

Background technology

Along with the development of mobile communication and wireless technology, location-based service is subject to paying close attention to more and more widely. PeopleDemand to indoor positioning information grows with each passing day, underground parking, logistic storage, mine, hospital, prison, archaeology scene, exhibitionThe large-scale indoor place such as the Room, museum all needs personnel or article to locate in real time, could realize navigation, monitoring and intelligenceThe functions such as energy management. Indoor orientation method based on radio frequency map is simple owing to having higher positioning precision and computational methods,Become one of topmost indoor wireless positioning method in recent years.

Indoor orientation method based on radio frequency map is mainly by building online positioning stage unknown node and off-line phaseReceived signals fingerprint in vertical reference point radio frequency map mates the position that calculates unknown node. But because indoor environment is subject toThe impact of the factors such as wall blocks, personnel walk about, causes radio frequency map can accurately not characterize the feelings of real time environment received signals fingerprintCondition, thereby cause positioning precision not high. In order to improve the accuracy of indoor positioning, the radio frequency map of before setting up needs timing moreNewly. The indoor orientation method in the past upgrading for radio frequency map just increased check-node simply, then passed through general manyUnit's linear regression method is determined regression coefficient. These algorithms are not considered RSSI signal active loss feature and renewal functionBe subject to the impact of RSSI exceptional value, thus the result obtaining affected by indoor disturbing factor larger.

Therefore,, for the precision problem of the indoor orientation method based on radio frequency map in indoor wireless networks location, proposeA kind of RSSI of minimizing positioning precision is subject to the method for indoor interference effect to become those skilled in the art's technical course urgently to be resolved hurrilyTopic.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, a kind of penetrating based on sub-clustering mechanism and robustness regression is providedFrequently map adaptive location method, the method is conducive to reduce RSSI, and to be subject to that indoor multipath, diffraction, barrier, personnel walk about etc. dryThe position error that the factor of disturbing causes, thus indoor position accuracy improved.

For achieving the above object, technical scheme of the present invention is: a kind of radio frequency ground based on sub-clustering mechanism and robustness regressionFigure adaptive location method, comprises the following steps:

(1) target area is divided into uniform grid, and in target area, places appropriate check-node and beacon jointPoint, the RSSI of the each beaconing nodes receiving in each grid element center point and check-node measurement respectively, builds static RSSI radio frequencyMap;

(2) according to described static RSSI radio frequency map, calculate each reference point, i.e. the path loss of grid element center point ginsengNumber;

(3) according to path loss parameter, reference point is carried out to sub-clustering, locating area is divided into multiple subregions;

(4) the RSSI fingerprint of checkpoint in the RSSI fingerprint of every sub regions internal reference point and this subregion is carried out steadily and surelyLinear regression;

(5) utilize checkpoint RSSI and the robustness regression coefficient update radio frequency map of timing acquisition, and by this radio frequency mapEstimate the position of unknown node for the K neighbor algorithm of weighting.

Further, in described step (1), target area is divided into uniform grid, the school of placing in target areaTest node and be evenly distributed in target area, an each described beaconing nodes, beaconing nodes placed in Si Ge corner, target areaBy mobile node and base station communication.

Further, described static RSSI radio frequency map is at t₀Moment obtains in the following ways: in each grid element centerPoint and check-node gather RSSI value continuous n time, the RSSI value of the each beaconing nodes receiving are averaged, that is:

Wherein, u_ij(t₀) be t₀The RSSI value of n that moment i reference point receives j beaconing nodes on averageValue, r_kj(t₀) be t₀The mean value of the RSSI value of the n that moment k check-node receives a j beaconing nodes, i=(1 ..., N), N is reference point number, k=(1 ..., M), M is check-node number, j=(1 ..., P), P is beaconing nodes number,N is natural number.

Further, in described step (2), the method for calculating the path loss parameter of each reference point is: utilize indoorSignal propagation characteristic model, obtain at reference point l_iPlace records beaconing nodes AP_jPath lossParameter, wherein d_ijBe the distance between i reference point and j beaconing nodes, α is constant under this environment.

Further, in described step (3), the method for reference point being carried out to sub-clustering according to path loss parameter is as follows:

Reference point l_iLocate the vector for path loss parameter that each beaconing nodes is correspondingRepresent,Definition path loss parameter similarity is:

，

Above formula representsWithSimilarity degree, 1≤u≤M, 1≤v≤M;

Each reference point is divided into K bunch by recycling K means clustering algorithm, is expressed as C={c₁,c₂,…,c_k, to a certain fixingK, maximizes following formula:

WhereinIt is classIn mean value, represent the cluster centre of y class,BeBelong to classSample size.

Further, in described step (4), by verification in the RSSI fingerprint of every sub regions internal reference point and this subregionThe method that the RSSI fingerprint of point carries out robust linear regression is: in every sub regions according to t₀Moment static RSSI radio frequency mapIn u_ij(t₀) and r_kj(t₀), utilize sane multiple linear regression to determine t₀The RSS of each reference point of moment and check-nodeRelation between RSS:, adopt iteration weighted least-squares to estimate backReturn coefficient, determine the power w of each point according to the size of regression residuals_i, to reach sane object, the object function of its optimization is:

Further, in described step (5), utilize the checkpoint RSSI vector of timing acquisitionWith i reference position point robustness regression coefficientUpgrade radio frequency map, wherein robustness regression coefficient is f_ijFor:

，1≤i≤N，1≤j≤M

At positioning stage, to each reference position point l_i, utilize the f of this position_ijReceive beaconing nodes with checkpointRSS value r_k=(r_k1,…,r_kj,…,r_kM) to location point l_iThe radio-frequency fingerprint of corresponding each beaconing nodes periodically updates;Suppose that moment t is at reference position point l_iPass through f_ijThe radio-frequency fingerprint that function effect obtains corresponding each beaconing nodes is, then this radio frequency map is estimated to the position of unknown node for the K neighbor algorithm of weighting.

Compared to prior art, beneficial effect of the present invention is: the present invention, by reference to a sub-clustering, passes radiofrequency signalBroadcast the identical or similar reference point of feature and be classified as in same cluster, and in trying to achieve bunch each reference point with check-node in clusterRelation between RSSI, thus make the fingerprint map that utilizes this relation to upgrade have more reliability. In addition, the present invention is by by eachIn subregion internal reference examination point and this subregion, the RSSI fingerprint of checkpoint carries out robust linear regression, weakens checkpoint and gets respectivelyThe impact of the abnormal RSSI value of beaconing nodes, further improves positioning precision. Provided by the present invention based on sub-clustering mechanism and steadyThe strong radio frequency map adaptive location method returning, does not have a large amount of computings, realizes simply, is applicable to very much energy constraint, calculates energyPower is limited, storage resources is limited and the application scenarios of communication capacity wireless network with limited.

Brief description of the drawings

Fig. 1 is the flow chart of the embodiment of the present invention.

Fig. 2 is wireless network node and mesh point deployment schematic diagram in the embodiment of the present invention.

Fig. 3 be the embodiment of the present invention draw in zones of different, signal propagation path loss parameter η characteristic pattern.

Detailed description of the invention

Below in conjunction with drawings and the specific embodiments, the present invention is described in further detail.

The radio frequency map adaptive location method that the present invention is based on sub-clustering mechanism and robustness regression, comprises the following steps:

(1) target area is divided into uniform grid, and in target area, places appropriate check-node and beacon jointPoint, the RSSI of the each beaconing nodes receiving in each grid element center point and check-node measurement respectively, builds static RSSI radio frequencyMap;

(2) according to described static RSSI radio frequency map, calculate each reference point, i.e. the path loss of grid element center point ginsengNumber;

(3) according to path loss parameter, reference point is carried out to sub-clustering, locating area is divided into multiple subregions;

(4) the RSSI fingerprint of checkpoint in the RSSI fingerprint of every sub regions internal reference point and this subregion is carried out steadily and surelyLinear regression;

(5) utilize checkpoint RSSI and the robustness regression coefficient update radio frequency map of timing acquisition, and by this radio frequency mapEstimate the position of unknown node for the K neighbor algorithm of weighting.

In an embodiment of the present invention, in described step (1), target area is divided into uniform grid, described each gridBetween central point, be spaced apart 1.5m, the check-node of placing in target area is evenly distributed in target area, in targetA described beaconing nodes of the each placement in Si Ge corner, region, beaconing nodes is by mobile node and base station communication.

In an embodiment of the present invention, described static RSSI radio frequency map is at t₀Moment obtains in the following ways:Each grid element center point and check-node gather RSSI value continuous n time, the RSSI value of the each beaconing nodes receiving be averaged,That is:

Wherein, u_ij(t₀) be t₀The RSSI value of n that moment i reference point receives j beaconing nodes on averageValue, r_kj(t₀) be t₀The mean value of the RSSI value of the n that moment k check-node receives a j beaconing nodes, i=(1 ..., N), N is reference point number, k=(1 ..., M), M is check-node number, j=(1 ..., P), P is beaconing nodes number,N is natural number.

In an embodiment of the present invention, in described step (2), calculate the method for the path loss parameter of each reference pointFor: utilize indoor signal propagation characteristic model, obtain at reference point l_iPlace records beaconing nodes AP_jPath loss parameter, wherein d_ijBe the distance between i reference point and j beaconing nodes, α isConstant under this environment.

In an embodiment of the present invention, in described step (3), according to path loss parameter, reference point is carried out the side of sub-clusteringMethod is as follows:

Reference point l_iLocate the vector for path loss parameter that each beaconing nodes is correspondingRepresent,For by all reference point sub-clusterings, definition path loss parameter similarity is:

，

Above formula representsWithSimilarity degree, 1≤u≤M, 1≤v≤M;

Each reference point is divided into K bunch by recycling K means clustering algorithm, is expressed as C={c₁,c₂,…,c_k, to a certain fixingK, maximizes following formula:

WhereinIt is classIn mean value, represent the cluster centre of y class,BeBelong to classSample size.

In an embodiment of the present invention, in described step (4), by the RSSI fingerprint of every sub regions internal reference point and this sonIn region, the RSSI fingerprint of checkpoint carries out the method for robust linear regression and is: in every sub regions according to t₀Moment static stateU in RSSI radio frequency map_ij(t₀) and r_kj(t₀), utilize sane multiple linear regression to determine t₀The RSS of each reference point of momentAnd the relation between the RSS of check-node:, adopt iteration weighting minimumTwo take advantage of estimation regression coefficient, determine the power w of each point according to the size of regression residuals_i, to reach sane object, the order of its optimizationScalar functions is:

In an embodiment of the present invention, in described step (5), utilize the checkpoint RSSI vector of timing acquisitionWith i reference position point robustness regression coefficientMoreNew radio frequency map, wherein robustness regression coefficient is f_ijFor:

，1≤i≤N，1≤j≤M

At positioning stage, to each reference position point l_i, utilize the f of this position_ijReceive beaconing nodes with checkpointRSS value r_k=(r_k1,…,r_kj,…,r_kM) to location point l_iThe radio-frequency fingerprint of corresponding each beaconing nodes periodically updates;Suppose that moment t is at reference position point l_iPass through f_ijThe radio-frequency fingerprint that function effect obtains corresponding each beaconing nodes is, then this radio frequency map is estimated to the position of unknown node for the K neighbor algorithm of weighting.

Concrete, as shown in Figure 1, comprise the steps:

Step 1: the deployment of wireless network node in the present embodiment as shown in Figure 2, is divided into target area evenlyGrid, makes the 1.5m that is spaced apart between each grid element center point, places appropriate check-node in target area, verification jointPoint should be evenly distributed in target area, and respectively places a beaconing nodes in Si Ge corner, target area, and beaconing nodes is logicalCross mobile node and base station communication. Gathering continuous n the collection RSSI value of grid element center point and check-node, to what receiveThe RSSI value of each beaconing nodes is averaged, that is:

，，,For reference point number,, M is schoolTest interstitial content,，For beaconing nodes number, n is natural number.

Step 2: be to utilize according to the path loss parameter of the each reference point of static RSSI radio frequency map calculation of this sceneIndoor signal propagation characteristic model, obtain at l_iRecord beaconing nodes AP_jPath loss parameter, whereinBe the distance between i reference point and j beaconing nodes,For constant under this environment.

Step 3: the path loss parameter nepal rattlesnake plantain root examination point obtaining according to step 2 is carried out sub-clustering, sub-clustering according to being: as attachedShown in Fig. 3, in zones of different, the value difference of signal propagation path loss parameter η, and in a certain region, η is same or similar, tableBe shown in this region radiofrequency signal propagation characteristic similar. Therefore we can be according to the value degree of approximation of η to each reference point placeSub-clustering is carried out in region, known according to above analysis, reference point locations radiofrequency signal propagation characteristic in same cluster identical orSimilar. The reason of above-mentioned sub-clustering be the RSSI of reference point with from it checkpoint relation away from not tight, therefore can make to penetrateFrequently the identical or similar reference point of signal propagation characteristic is classified as in same cluster, and obtains t₀The RSSI of each reference point in moment bunchAnd with the relation between the RSSI of check-node in cluster, thereby make the fingerprint map that utilizes this relation to upgrade have more reliability.

Cluster-dividing method is as follows:

Reference point l_iLocate the path loss parameter availability vector that each beaconing nodes is corresponding. In order to incite somebody to actionAll reference point sub-clusterings, definition path loss parameter similarity is，It representsWithSimilarity degree,,。Value larger,WithMore similar.

Each reference point is divided into K bunch by recycling K means clustering algorithm, is expressed as, and to a certain fixing K, adopts heuristic calculationMethod, even if following formula maximizes:

WhereinIt is classIn mean value, represent the cluster centre of y class,BeBelong to classSample size.

Step 4: the RSSI fingerprint of checkpoint in the RSSI fingerprint of every sub regions internal reference point and this subregion is carried outRobust linear regression. Owing to being subject to the interference such as indoor multipath, diffraction, diffraction, article displacement, personnel walk about at indoor radio signalThe impact of factor, easily there is exceptional value in the RSSI value of each beaconing nodes that checkpoint gets. For reducing " abnormity point " effect,Can apply different weights to different points, give larger weight to the little point of residual error, and the larger point of residual error is givenGive less weight, determine weight according to residual error size, and set up accordingly the least-squares estimation of weighting, iterate to improveWeight coefficient, until the change of weight coefficient is less than certain allowable error. Employing robustness regression is estimated, is utilized it to error termThe sane characteristic distributing, can effectively get rid of the abnormal RSS value that is subject to indoor interference. Special in the time of error abnormal, it is wanted than LSEMuch better.

In every sub regions according to t₀In moment static RSSI radio frequency mapWith，，，, utilize sane multiple linear regression to determine t₀The RSS of each reference point of moment and check-nodeRelation between RSS:, adopt iteration weighted least-squares to estimate backReturn coefficient, determine the power w of each point according to the size of regression residuals_i, to reach sane object, the object function of its optimization is:

Step 5: according to the checkpoint of timing acquisitionWith iIndividual reference position point robustness regression coefficientUpgrade radio frequency map, wherein robustness regression coefficient is f_ijFor:

，,

Positioning stage, to each reference position point l_i, utilize the f of this position_ijReceive the RSS of beaconing nodes with checkpointValue r_k=(r_k1,…,r_kj,…,r_kM) to location point l_iThe radio-frequency fingerprint of corresponding each beaconing nodes periodically updates.

Suppose that moment t is at reference position point l_iPass through f_ijFunction effect obtains the radio-frequency fingerprint of corresponding each beaconing nodesFor, then this radio frequency map is estimated to the position of unknown node for the K neighbor algorithm of weighting.

In sum, the present invention, by according to path loss parameter, reference point being carried out to sub-clustering, is divided into locating areaMultiple subregions, then in the RSSI fingerprint to every sub regions internal reference point and this subregion, the RSSI fingerprint of checkpoint carries outRobust linear regression; Online positioning stage, utilizes checkpoint RSSI and the robustness regression coefficient update radio frequency map of timing acquisition,The K neighbor algorithm for weighting by this radio frequency map again, thus realize location. So, can effectively reduce RSSI randomized jitter withAnd the indoor occupant factors such as interference of walking about cause the out-of-date impact that location Calculation is caused of radio frequency map, realize comparatively exactly chamberInterior location.

More than preferred embodiment of the present invention, all changes of doing according to technical solution of the present invention, the function producing is doneWhen not exceeding the scope of technical solution of the present invention, all belong to protection scope of the present invention.

Claims (1)

1. the radio frequency map adaptive location method based on sub-clustering mechanism and robustness regression, is characterized in that, comprises followingStep:

(1) target area is divided into uniform grid, and in target area, places appropriate check-node and beaconing nodes, pointThe RSSI of the each beaconing nodes not receiving in each grid element center point and check-node measurement, builds static RSSI radio frequency map;

(2) according to described static RSSI radio frequency map, calculate each reference point, i.e. the path loss of each grid element center point ginsengNumber;

(3) according to path loss parameter, reference point is carried out to sub-clustering, target area is divided into multiple subregions;

(4) the RSSI fingerprint of checkpoint in the RSSI fingerprint of every sub regions internal reference point and this subregion is carried out to sane linearityReturn;

(5) utilize checkpoint RSSI and the robustness regression coefficient update radio frequency map of timing acquisition, and this radio frequency map is used forThe K neighbor algorithm of weighting estimates the position of unknown node;

In described step (1), target area is divided into uniform grid, the check-node of placing in target area is uniformly distributedIn target area, respectively in Si Ge corner, target area place a described beaconing nodes, beaconing nodes by mobile node withBase station communication; Described static RSSI radio frequency map is at t₀Moment obtains in the following ways: in each grid element center point and verificationNode gathers RSSI value continuous n time, the RSSI value of the each beaconing nodes receiving is averaged, that is:

$u_{ij}\left(t_0\right)=\frac{1}{n}{\mathrm{rssi}}_{ij}$

$r_{kj}\left(t_0\right)=\frac{1}{n}{\mathrm{rssi}}_{kj}$

Wherein, u_ij(t₀) be t₀The mean value of the RSSI value of the n that moment i reference point receives a j beaconing nodes, r_kj(t₀) be t₀The mean value of the RSSI value of the n that moment k check-node receives a j beaconing nodes, i=(1 ..., N),N is reference point number, k=(1 ..., M), M is check-node number, j=(1 ..., P), P is beaconing nodes number, n is natureNumber;

In described step (2), the method for calculating the path loss parameter of each reference point is: utilize indoor signal propagation characteristic mouldType P_r=α+10 η log (d), obtain at reference point l_iPlace records beaconing nodes AP_jPath loss parameterWherein d_ijBe the distance between i reference point and j beaconing nodes, α is normal under this environmentNumber;

In described step (3), the method for reference point being carried out to sub-clustering according to path loss parameter is as follows:

Reference point l_iLocate the vector for path loss parameter that each beaconing nodes is correspondingTableShow, definition path loss parameter similarity is:

Similarity(η(l_u),η(l_v))＝-(η_u-η_v)^TΣ^-1(η_u-η_v)，

Above formula representsWithSimilarity degree, 1≤u≤M, 1≤v≤M;

Each reference point is divided into K bunch by recycling K means clustering algorithm, is expressed as C={c₁,c₂,…,c_k, to a certain fixing K, makeFollowing formula maximizes:

$J_C(\mathrm{\Σ},X)=\underset{y=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{N}{\Σ}}\[-{({\mathrm{\η}}_i-c_y)}^T{\mathrm{\Σ}}^{-1}({\mathrm{\η}}_i-c_y)\]-ln\left|{\mathrm{\Σ}}^{-1}\right|$

WhereinClass C_yIn mean value, represent the cluster centre of y class, n_yTo belong toClass C_ySample size;

In described step (4), by the RSSI fingerprint of checkpoint in the RSSI fingerprint of every sub regions internal reference point and this subregionThe method of carrying out robust linear regression is: in every sub regions according to t₀U in moment static RSSI radio frequency map_ij(t₀) andr_kj(t₀), utilize sane multiple linear regression to determine t₀Pass between the RSS of each reference point of moment and the RSS of check-nodeSystem: u_ij(t₀)＝f_ij(r_1j(t₀),…,r_kj(t₀),…,r_Mj(t₀)), adopt iteration weighted least-squares to estimate regression coefficient,Determine the power w of each point according to the size of regression residuals_i, to reach sane object, the object function of its optimization is:

$\min f({\mathrm{\β}}_0,{\mathrm{\β}}_1,...,{\mathrm{\β}}_p)=\underset{i=1}{\overset{N}{\Σ}}{w}_i{(Y_i-\underset{j=1}{\overset{p}{\Σ}}{x}_{ij}{\mathrm{\β}}_j)}^2$

In described step (5), utilize the checkpoint RSSI vector r of timing acquisition_kj(t₀)＝(r_1j(t_i),…,r_kj(t_i),…,r_Mj(t_i)) and i reference position point robustness regression coefficientUpgrade radio frequency map, wherein robustness regression systemNumber is f_ijFor:

$u_i={\widehat{\mathrm{\β}}}_0+{\widehat{\mathrm{\β}}}_1{r}_{1j}+{\widehat{\mathrm{\β}}}_2{r}_{2j}+...+{\widehat{\mathrm{\β}}}_M{r}_{Mj},1\≤i\≤N,1\≤j\≤M$

At positioning stage, to each reference position point l_i, utilize the f of this position_ijReceive the RSS value of beaconing nodes with checkpointr_k＝(r_k1,…,r_kj,…,r_kM) to location point l_iThe radio-frequency fingerprint of corresponding each beaconing nodes periodically updates; While suppositionCarve t at reference position point l_iPass through f_ijThe radio-frequency fingerprint that function effect obtains corresponding each beaconing nodes isAgain this radio frequency map is estimated to the position of unknown node for the K neighbor algorithm of weighting.