CN105866732A - Improved MK model and WKNN algorithm combined mixed indoor positioning method - Google Patents

Abstract

The present invention discloses an improved MK model and WKNN algorithm combined mixed indoor positioning method. According to the method, an improved MK model is used to establish a signal spread model which is more suitable for a complex indoor environment, a nearest adjacent point is helped to filtered in a positioning state, thus a nearest adjacent point with a large difference does not participate the positioning of a WKNN algorithm, according to the Euclidean distance formula, the distance of each nearest adjacent point to an AP is calculated, the distances are taken as weights to be substituted into the formula of the WKNN algorithm to obtain a final estimated position value.

Description

The mixing indoor orientation method that a kind of MK of improvement model and WKNN algorithm combine

Technical field

The present invention relates to the improvement MK model under a kind of complex indoor environment and mixed positioning that WKNN algorithm combines is calculated Method, belongs to wireless indoor field of locating technology.

Background technology

Along with the development of wireless communication technology, demand based on location-based service receives much concern.Outdoor positioning technology is at present Through more ripe, it is fully able to meet location requirement；But, indoor positioning technologies but can not meet growing location clothes The business demand to positioning precision.Using relatively broad indoor positioning technologies to have WIFI location technology, its positioning principle is: will WIFI location label is arranged on target object to be followed the tracks of, and location label periodically sends out wireless signal, AP (Access Point, be called for short: access point) receive signal after, pass the signal to engine of positioning, engine of positioning is wireless according to receive The power of signal, calculates and judges this label present position, show particular location by visualization interface, it is achieved the most fixed Position is followed the tracks of and management.

Indoor orientation method based on WIFI mainly has two kinds: propagation model method and location fingerprint method.Typical indoor letter Number propagation model has: linear range path loss model, log-distance path loss model model, decay factor model and MK model. Location algorithm based on propagation model mainly has: three limit positioning modes, hyperbolic fix method, method of least square.Location fingerprint positions Algorithm mainly has: NN (Nearest Neighbor is called for short: nearest neighbor method), KNN (K-Nearest Neighbors, abbreviation: K Nearest neighbor method), WKNN (Weighted K Nearest Neighbors, be called for short: weighting K nearest neighbor method), Bayesian probability calculate Method and neural network algorithm etc..

The barrier that block signal is propagated by MK model processes, and adds on the basis of log-distance path loss model model Enter wall and signal attenuation that floor causes:

L (d)=L+10nlog (d)+N_wL_w+ N_fL_f

Wherein L represents the propagation loss at 1 meter of range transmission end；N is path loss coefficient；D represents that transmitting terminal arrives reception The distance of end；N_wAnd N_fRepresent the number on wall and the floor passed from transmitting terminal to receiving terminal, L respectively_wAnd L_fRespectively represent wall and The loss factor on floor.This model belongs to the statistical model in indoor propagation model, fast relative in deterministic models computing, defeated Enter simple, it is not necessary to pretreatment and simplification, it is not necessary to expensive equipment.

Localization method based on location fingerprint is generally divided into off-line training step and tuning on-line stage: (1) off-line positions The main task in stage is to set up a location fingerprint data base.In region, location, arrange test reference point, gather multiple AP Signal intensity, remember as the signal characteristic parameter of this reference point AP through screening (generally use be worth the method being averaged most) Record in the fingerprint database of position.(2) the tuning on-line stage utilizes MS (Mobile Station is called for short: mobile station) to record The signal characteristic parameter of a certain position, by corresponding matching algorithm, according in measured data and location fingerprint data base Data compare, thus (Received Signal Strength Index, is called for short: receive to obtain one or one group of RSSI Signal intensity indicates) and the data of positional information, thus complete location.

WKNN chooses K nearest neighbor point, is then weighted the coordinate of this K nearest neighbor point processing, obtains final position Put estimated value.

Distance between information point generally uses Euclidean distance and represents:

$L=\frac{1}{n}{\left(\underset{i=1}{\overset{n}{\Σ}}{\left|s-s_i\right|}^2\right)}^{\frac{1}{2}}$

Wherein s represents the signal message recorded, s_iRepresent the signal characteristic parameter in finger print data.

The formula of WKNN is:

$(\hat{x},\hat{y})=\underset{i=1}{\overset{K}{\Σ}}\frac{\frac{1}{d_i+\mathrm{\ϵ}}}{\underset{j=1}{\overset{K}{\Σ}}\frac{1}{d_j+\mathrm{\ϵ}}}(x_i,y_i)$

Wherein d_jRepresent the weights of jth nearest neighbor point, (x_i, y_i) represent i-th nearest neighbor point coordinate.

Localization method based on propagation model considers this indoor signal propagation law, but have ignored ad-hoc location to letter Number unique impact；Localization method based on location fingerprint, by collecting location fingerprint, by received signal strength and positional information Combine, strengthen the utilization to positional information, but have ignored the propagation law of signal.

Summary of the invention

Present invention aim at for above-mentioned the deficiencies in the prior art, it is provided that a kind of improvement MK model and WKNN algorithm phase In conjunction with hybrid locating method, the method can realize the location of higher precision under complex indoor environment well.

The present invention solves its technical problem and is adopted the technical scheme that: a kind of improvement MK model and WKNN algorithm combine Mixing indoor orientation method, the method utilizes the MK model improved to set up to be more suitable for the signal propagating mode of complex indoor environment Type, helps screening nearest neighbor point at positioning stage so that differ bigger nearest neighbor point and be not involved in the location of WKNN algorithm, root According to Euclidean distance formula, try to achieve each nearest neighbor point distance to AP, using these distances as weights, be updated to WKNN algorithm Formula in try to achieve final estimation positional value.

Method flow:

Step 1: arrange sampled point in localizing environment, measures RSSI at each sampled point, according to the RSSI of sampled point and Positional information sets up fingerprint database；

Step 2: determine the barrier kind in localizing environment, it is thus achieved that the MK model of improvement:

l₀It is constant, k_iAnd k_jRepresent the number through wall and floor, l respectively_wAnd l_fRepresent different types of wall respectively With the attenuation quotient on floor, k_ijRepresent the number through different barriers, l_ijRepresent the decay through i-th kind of barrier of jth Coefficient.

Method of least square or method of maximum likelihood is utilized to obtain the parameter of the improvement MK model under this localizing environment；

Step 3: all fingerprints in location fingerprint data base are updated in this MK model calculate error, and obtain average Error；

Step 4: according to selecting to measure point, acquisition RSSI group.The element of this RSSI group is updated to MK model formation respectively In, it is calculated corresponding point of measuring and arrives the distance vector L specified between AP；

Step 5: compared with respective Neighbor Points by the element in L, if error is calculated more than in step 3 Mean error, just gives up this Neighbor Points, meets the requirements without Neighbor Points, just chooses that Neighbor Points that error is minimum；

Step 6: obtain the position that this Neighbor Points is corresponding；

Step 7: using the inverse of the distance of each Neighbor Points obtained in step 6 as weights, be updated to formula WKNN algorithm In formula, it may be assumed that

$(\hat{x},\hat{y})=\frac{\underset{i=1}{\overset{K}{\Σ}}\frac{1}{w_i}(x_i,y_i)}{\underset{j=1}{\overset{K}{\Σ}}\frac{1}{w_j}}$

w_jIt is the weights of WKNN algorithm, (x_i,y_i) it is the coordinate of i-th reference point, then estimation obtains last position.

Beneficial effect:

1, the present invention is improved on the basis of MK model, more accurately simulation indoor signal propagation condition, utilizes The bigger nearest neighbor point of MK model filter error improved, using the inverse of qualified nearest neighbor point and the distance of AP as weights, Consider the RSSI of sampled point and the relation of positional information in indoor signal propagation law and location fingerprint data base, thus Improve positioning precision.

2, during the present invention preferably make use of indoor signal propagation law and location fingerprint data base the RSSI of sampled point with The relation of positional information, filters out the Neighbor Points that error is bigger, thus improves positioning precision, reduces position error, in complicated room Under interior environment, performance is the most outstanding.

Accompanying drawing explanation

Fig. 1 is the position distribution schematic diagram of sampled point in region, location.

Fig. 2 is the RSSI distribution schematic diagram of the AP of three known location.

The range error schematic diagram of the sampled point that MK model is obtained is improved according to Fig. 3.

Fig. 4 is the hybrid algorithm and NN algorithm and the error comparison diagram of W2NN algorithm improved.

Detailed description of the invention:

Below in conjunction with Figure of description, the invention is described in further detail.

In the concrete application scenarios of the present invention is certain teaching and research room.Participate in location area about 240 square metres, be distributed in three teaching and research In room, totally 56 sampled points, in 10 sampling point distributions corridor between teaching and research room, between teaching and research room by hardwood plate every Open, between one of them teaching and research room and corridor, have thick clamping plate.The horizontal range of neighbouring sample point 2 meters, vertical distance is mostly 1 Rice.Semen setariae 2 mobile phone from exploitation program measuring function in morning by being provided with WIFI signal RSS stands in sampled point towards AP Orientation measurement RSSI, each sampled point is at least sampled 30 times.Three AP are placed on appointment position: AP1:(6, and 2)；AP2 (5, 9)；AP3 (10.5,1), unit is rice.Sampling point position distribution is as it is shown in figure 1, abscissa represents region, location horizontal direction, vertical Region, coordinate representation location vertical direction, unit is all rice.

The simulation process of the present invention includes the following:

(1) the on-line training stage

The RSSI collection of each sampled point obtained by handheld device is stored in the file of correspondence, was worth most through the past and is averaged Value processes the final RSSI value obtaining each sampled point, sets up location fingerprint data base.According to the location fingerprint data base set up Just can be seen that the relation between RSSI and position.Fig. 2 is the RSSI distribution schematic diagram of each AP.In figure, abscissa represents positioning area Territory horizontal direction, vertical coordinate represents region, location vertical direction, and unit is all rice.

In teaching and research room, barrier can be divided into: the solid wood between teaching and research room, and the veneer between corridor and teaching and research room is done Dividing plate on public table.The MK model improved is:

$L\left(d\right)=l_0+10\mathrm{\γ}log\left(d\right)+N_{w}W+\underset{i=1}{\overset{n}{\Σ}}{k}_{i}l+N_{g}G$

Wherein N_wAnd N_gRepresent the number of solid wood and the veneer passed from AP to handheld device respectively, W and G table respectively Show through solid wood and the loss factor of veneer, l represents the loss factor of desk upper spacer, k_iRepresent that AP sets with hand-held Standby space bar number.

It is determined by positioning the number of the these three barrier between AP and the handheld device in region and obtaining before RSSI value be updated to improve MK model formation in.The formula of method of least square is:

$L\left(d\right)=\left[\begin{array}{ccccccccc}1& 10{\log }_{10}\left(d\right)& k_{11}& \mathrm{...}& k_{1n}& \mathrm{...}& k_{m1}& \mathrm{...}& k_{mn}\end{array}\right]\left[\begin{array}{c}{l}_0\\ \mathrm{\γ}\\ l_{11}\\ .\\ .\\ .\\ l_{1n}\\ .\\ .\\ .\\ l_{m1}\\ .\\ .\\ .\\ l_{mn}\end{array}\right]$

$\stackrel{\‾}{Y}=\left[\begin{array}{c}{L}_1\\ L_2\\ L_3\\ .\\ .\\ .\end{array}\right],\mathrm{\β}=\left[\begin{array}{c}{l}_0\\ \mathrm{\γ}\\ l_{11}\\ .\\ .\\ .\\ l_{mn}\end{array}\right],\stackrel{\‾}{X}=\left[\begin{array}{ccccc}1& 10{\log }_{10}\left(d_1\right)& k_{11}& \mathrm{...}& k_{1n}\\ 1& 10{\log }_{10}\left(d_2\right)& k_{21}& \mathrm{...}& k_{2n}\\ 1& 10{\log }_{10}\left(d_3\right)& k_{31}& \mathrm{...}& k_{3n}\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\end{array}\right]$

β=(X^TX)^-1X^TY

Above-mentioned formula is utilized to obtain the MK Model Parameter value of improvement: W value is-1.7688, and G-value is-1.1435, l is- The l of 0.7069, AP1₀Value is for-37.3264, and the γ-value of AP1 is-1.8966.For AP1 signal propagation model formula i.e.:

$L\left(d\right)=-37.3264-18.966log\left(d\right)-1.7688N_w-1.1435N_g-0.7069\underset{i=1}{\overset{n}{\Σ}}{k}_i$

The RSSI value of each sampled point in location fingerprint data base is updated in WKNN algorithmic formula, obtains each sampled point With the distance value of AP1, then poor absolute value is asked to draw by mistake required distance value and each self-corresponding coordinate and the distance of AP1 Difference, obtains mean error.Simulation result shows the distance and actual range that the signal propagation model formula according to AP1 obtains Mean error is: 0.7318 meter.It is each sampled point shown in Fig. 3 and the signal propagation model formula of the actual range AP1 of AP1 draws Range error schematic diagram.

(2) off-line positioning stage

The RSSI measuring altogether five test points carrys out verification algorithm, and the coordinate of these five points is followed successively by: (9.3,4.8), (11.15,3.96), (12.97,0), (4.3,1.4), (3.6,6.16).The signal that these five RSSI value are updated to AP1 is propagated In model formation, distance array L being calculated between these test point and AP1.According to RSSI location fingerprint data before Storehouse carries out coupling and obtains the Neighbor Points of test point.Calculate the distance between Neighbor Points and AP1 corresponding with in distance array L Distance versus, if error is more than 1.414 times of mean error tried to achieve before, then give up this Neighbor Points.Without appointing What Neighbor Points meets this requirement, then take that Neighbor Points that error is minimum.The set of these Neighbor Points will be used for WKNN As its coordinate in algorithmic formula, each Neighbor Points utilizes distance that Euclid's formula tries to achieve as weights, is updated to WKNN and calculates Method formula can be obtained by positioning result.Through calculating, the first Neighbor Points is followed successively by: (10.0,7.45), (12.15,1.0), (12.15,0), (4.0,1.0), (6.0,5.38)；Second Neighbor Points is followed successively by: (9.0,7.45), (16.15,0.0), (14.15,3.0), (4.0,0), (10.0,7.45)；3rd Neighbor Points is followed successively by (7.0,7.45), (14.15,3.0), (15.15,0), (4.0,2.0), (9.0,7.45).

Fig. 4 is that the mixing that the positioning result of NN algorithm and W2NN algorithm combines with the MK model of improvement and WKNN algorithm is fixed The error contrast of position algorithm positioning result, it can be seen that the hybrid algorithm after improvement can filter out the Neighbor Points that error is bigger, It is significantly improved on the whole compared with the positioning precision of NN algorithm and W2NN algorithm.

In sum, the present invention improves MK model and the hybrid locating method that combines of WKNN algorithm preferably make use of In indoor signal propagation law and location fingerprint data base, the RSSI of sampled point and the relation of positional information, filter out error bigger Neighbor Points, thus improve positioning precision, reduce position error, under complex indoor environment, performance is the most outstanding.

Claims (3)

1. one kind is improved MK model and mixing indoor orientation method that WKNN algorithm combines, it is characterised in that described method bag Include following steps:

Step 1: arrange sampled point in localizing environment, measures RSSI at each sampled point, according to RSSI and the position of sampled point Information sets up fingerprint database；

Step 2: determine the barrier kind in localizing environment, it is thus achieved that the MK model of improvement；

$L\left(d\right)=10{\mathrm{\γlog}}_{10}\left(d\right)+l_0+\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{k}_{ij}{l}_{ij}$

l₀It is constant, k_iAnd k_jRepresent the number through wall and floor, l respectively_wAnd l_fRepresent different types of wall and ground respectively The attenuation quotient of plate, k_ijRepresent the number through different barriers, l_ijRepresent the decay system through i-th kind of barrier of jth Number；

Method of least square or method of maximum likelihood is utilized to obtain the parameter of the improvement MK model under this localizing environment；

Step 3: be updated to all fingerprints in location fingerprint data base in this MK model calculate error, and averagely missed Difference；

Step 4: according to selecting to measure point, acquisition RSSI group, the element of this RSSI group is updated in MK model formation respectively, meter Calculation obtains corresponding point of measuring and arrives the distance vector L between appointment AP；

Step 5: the element in L is compared with respective Neighbor Points, if error more than in step 3 calculated averagely Error, just gives up this Neighbor Points, meets the requirements without Neighbor Points, just chooses that Neighbor Points that error is minimum；

Step 6: obtain the position that this Neighbor Points is corresponding；

Step 7: using the inverse of the distance of each Neighbor Points obtained in step 6 as weights, be updated to the formula of formula WKNN algorithm In, it may be assumed that

$(\hat{x},\hat{y})=\frac{\underset{i=1}{\overset{K}{\Σ}}\frac{1}{w_i}(x_i,y_i)}{\underset{j=1}{\overset{K}{\Σ}}\frac{1}{w_j}}$

w_jIt is the weights of WKNN algorithm, (x_i,y_i) it is the coordinate of i-th reference point, then estimation obtains last position.

The mixing indoor orientation method that a kind of MK of improvement model the most according to claim 1 and WKNN algorithm combine, its It is characterised by: described method utilizes the MK model improved to set up the signal propagation model being more suitable for complex indoor environment, in location Stage help screening nearest neighbor point so that differ bigger nearest neighbor point and be not involved in the location of WKNN algorithm, according to Europe several in Obtain range formula, try to achieve each nearest neighbor point distance to AP, using these distances as weights, be updated in the formula of WKNN algorithm Try to achieve final estimation positional value.

The mixing indoor orientation method that a kind of MK of improvement model the most according to claim 1 and WKNN algorithm combine, its It is characterised by: the formula of described method of least square is:

$L\left(d\right)=\left[\begin{array}{ccccccccc}1& 10{\log }_{10}\left(d\right)& k_{11}& \mathrm{...}& k_{1n}& \mathrm{...}& k_{m1}& \mathrm{...}& k_{mn}\end{array}\right]\left[\begin{array}{c}{l}_0\\ \mathrm{\γ}\\ l_{11}\\ .\\ .\\ .\\ l_{1n}\\ .\\ .\\ .\\ l_{m1}\\ .\\ .\\ .\\ l_{mn}\end{array}\right]$

$\stackrel{\‾}{Y}=\left[\begin{array}{c}{L}_1\\ L_2\\ L_3\\ .\\ .\\ .\end{array}\right],\mathrm{\β}=\left[\begin{array}{c}{l}_0\\ \mathrm{\γ}\\ l_{11}\\ .\\ .\\ .\\ l_{mn}\end{array}\right],\stackrel{\‾}{X}=\left[\begin{array}{ccccc}1& 10{\log }_{10}\left(d_1\right)& k_{11}& \mathrm{...}& k_{1n}\\ 1& 10{\log }_{10}\left(d_2\right)& k_{21}& \mathrm{...}& k_{2n}\\ 1& 10{\log }_{10}\left(d_3\right)& k_{31}& \mathrm{...}& k_{3n}\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\end{array}\right]$

β=(X^TX)^-1X^TY。