CN101820639B - WLAN (Wireless Local Area Network) indoor three-layer ANN (Artificial Neural Network) intelligent positioning method based on fault-tolerant technique - Google Patents

Abstract

The invention relates to a WLAN (Wireless Local Area Network) indoor three-layer ANN (Artificial Neural Network) intelligent positioning method based on a fault-tolerant technique, relating to the field of complex system identification and solving the problem that the three-layer ANN has unstable fault-tolerant and unsatisfactory positioning precision since the number of nodes of hidden layers in the traditional three-layer ANN is difficult to confirm. The method comprises steps of: firstly taking the number of an AP (Access Point) as the number of nodes of an ANN input layer and the dimensionality of position information as the number of nodes of an ANN output layer, then utilizing reference point RSS (Received Signal Strength) matrixes and the position coordinates of corresponding reference points to form an ANN training set, training the ANN according to the ANN training set and by selecting the number of nodes of different ANN hidden layers to obtain the number of nodes of the ANN hidden layer with the highest fault-tolerant satisfaction and save the ANN fault-tolerant structure, and finally introducing a test point RSS matrix of a point to be tested into the ANN fault-tolerant structure to obtain the positioning coordinate of the point to be tested. The invention is suitable for positioning in complex systems.

Description

Based on the indoor three layers of ANN intelligent locating method of the WLAN of fault-toleranr technique

Technical field

The present invention relates to complication system identification field, be specifically related to indoor three layers of ANN intelligent locating method based on the WLAN of fault-toleranr technique.

Background technology

Along with the development of wireless network, many technology and the application relevant with the location have appearred, and particularly in the environment sensing application facet, wherein, wireless location is one of important application of wireless lan (wlan).Along with communication service demonstrates diversity, wireless location more and more receives people's attention, and aspect social public service, the important use meaning is arranged.

The navigation system that is applicable to LAN is called the terrestrial wireless location technology; Use at present more widely in the terrestrial wireless navigation system; Utilize the variation of transmitting terminal ambient signals intensity RSS (Received Signal Strength) value to come the method for positioning mobile station to be based on wireless network empirical value database-located method; Since based on the WLAN indoor positioning technology of signal strength signal intensity RSS value in the network of various support 802.11 agreements; Can get access to the signal strength signal intensity RSS of each access point AP (access point) at portable terminal; Thereby be the most economic method, its positioning accuracy under the stable situation of indoor environment can reach Centimeter Level simultaneously, being paid attention to very.

Artificial neural net (ANN) method based on the WLAN indoor positioning is a kind of RSS location technology; Three layers of ANN are that effective method is shone upon in non-linear input and output; Can approach the non-linear relation of any complicacy; And have powerful learning ability, memory capability, a computing capability, in various degree with information processing, storage and the search function of level patrix apery cerebral nervous system.Measure the signal strength signal intensity that arrives access point AP in advance based on the ANN method of WLAN indoor positioning, obtain positional information, through powerful self study and adaptive ability, foundation can be simulated the network configuration of current indoor positioning information.Therefore, three layers of ANN can obtain the positional information of anywhere in the current WLAN indoor environment, and have obtained comparatively ideal positioning accuracy.

In three layers of ANN method commonly used, the node number of input layer depends on the number of access point AP, and the node number of output layer depends on the dimension of position coordinates.But the node number of hiding layer but often obtains through experience and qualitative experiment, does not have complete rationale, owing to lack the rational method that the number of node layer is hidden in selection, makes the training result of three layers of ANN unstable, and positioning accuracy is undesirable.

Summary of the invention

Hide that the node layer number is difficult to confirm and the fault-tolerance instability and the unfavorable problem of positioning accuracy that make three layers of ANN the invention provides indoor three layers of ANN intelligent locating method based on the WLAN of fault-toleranr technique in order to solve among existing three layers of ANN.

The indoor three layers of ANN intelligent locating method of WLAN based on fault-toleranr technique of the present invention, its position fixing process is:

Step 1: in WLAN indoor positioning environment; N reference point, a U test point and M access point AP are set; Define the position coordinates of each reference point and each test point simultaneously; Guarantee that each reference point is covered by the signal that two or more access point AP send, guarantee that simultaneously each test point is also covered by the signal that two or more access point AP send;

Step 2: according to the number M of access point AP and the dimension initialization ANN structure of position coordinates, be specially: the number M of getting access point AP is the node number of ANN input layer, and the dimension that coordinate is put in fetch bit is the node number of ANN output layer;

Step 3: will form reference point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same reference point; And with the position coordinates of each reference point and the reference point RSS matrix composition ANN training set of corresponding reference point; Utilize said ANN training set to train a plurality of ANN; Obtain a plurality of ANN structures, it is different that the ANN of said a plurality of ANN hides the node layer number;

Step 4: will form test point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same test point, according to the slope α value of almost function, the amount of jitter x that the WLAN navigation system is allowed _iTest point RSS matrix with any test point; The ANN that calculates each ANN structure hides the fault-tolerant measures of dispersion of each node of layer; And the fault-tolerant measures of dispersion of calculating the hiding layer of ANN of each ANN structure is 0 node number; Said fault-tolerant measures of dispersion is that the position coordinates of ANN output layer output was constant after 0 expression ANN input layer added the amount of jitter that the WLAN navigation system allows;

Step 5: the fault-tolerant measures of dispersion of hiding layer according to the ANN of each ANN structure is total node number that the ANN of 0 node number and each ANN structure hides layer; The ANN of the said fault-tolerant satisfaction λ of fault-tolerant satisfaction

record that calculates each ANN structure when the highest hides the node layer number, preserves the corresponding ANN structure and is the ANN fault-tolerant architecture with said ANN organization definition;

Step 6: from all test points, choose tested point,, obtain the elements of a fix of tested point to the test point RSS matrix of ANN fault-tolerant architecture importing tested point.

Beneficial effect of the present invention is: the present invention proposes a kind of hiding node layer number that can confirm three layers of ANN, and guarantee the good indoor three layers of ANN intelligent locating method of WLAN of ANN structure fault freedom; The present invention has defined fault-tolerant satisfaction λ as the standard of weighing fault-tolerance; And then three layers of good ANN structure of fault freedom have effectively been selected; Replace original dependence empirical value and confirmed the situation of three layers of ANN structure, thereby guaranteed the validity and the reliability of WLAN indoor locating system; Because the network of zmodem has stronger noise resisting ability, thereby has improved the WLAN indoor position accuracy effectively, position error of the present invention is less than 0.81m.

Description of drawings

Fig. 1 is the flow chart of the indoor three layers of ANN intelligent locating method of the WLAN based on fault-toleranr technique of the present invention; Fig. 2 is the laboratory experiment scene sketch map described in the instance analysis in the embodiment two of the present invention; Fig. 3 is the reference point and the test point distribution schematic diagram in the rooms 1211 described in the instance analysis in the embodiment two of the present invention.

Embodiment

Embodiment one: specify this execution mode according to Figure of description 1, the indoor three layers of ANN intelligent locating method of the described WLAN of this execution mode based on fault-toleranr technique, its position fixing process is:

Step 1: in WLAN indoor positioning environment; N reference point, a U test point and M access point AP are set; Define the position coordinates of each reference point and each test point simultaneously; Guarantee that each reference point is covered by the signal that two or more access point AP send, guarantee that simultaneously each test point is also covered by the signal that two or more access point AP send;

Step 2: according to the number M of access point AP and the dimension initialization ANN structure of position coordinates, be specially: the number M of getting access point AP is the node number of ANN input layer, and the dimension that coordinate is put in fetch bit is the node number of ANN output layer;

Step 3: will form reference point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same reference point; And with the position coordinates of each reference point and the reference point RSS matrix composition ANN training set of corresponding reference point; Utilize said ANN training set to train a plurality of ANN; Obtain a plurality of ANN structures, it is different that the ANN of said a plurality of ANN hides the node layer number;

Step 4: will form test point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same test point, according to the slope α value of almost function, the amount of jitter x that the WLAN navigation system is allowed _iTest point RSS matrix with any test point; The ANN that calculates each ANN structure hides the fault-tolerant measures of dispersion of each node of layer; And the fault-tolerant measures of dispersion of calculating the hiding layer of ANN of each ANN structure is 0 node number; Said fault-tolerant measures of dispersion is that the position coordinates of ANN output layer output was constant after 0 expression ANN input layer added the amount of jitter that the WLAN navigation system allows;

Step 5: the fault-tolerant measures of dispersion of hiding layer according to the ANN of each ANN structure is total node number that the ANN of 0 node number and each ANN structure hides layer; The ANN of the said fault-tolerant satisfaction λ of fault-tolerant satisfaction

record that calculates each ANN structure when the highest hides the node layer number, preserves the corresponding ANN structure and is the ANN fault-tolerant architecture with said ANN organization definition;

Step 6: from all test points, choose tested point,, obtain the elements of a fix of tested point to the test point RSS matrix of ANN fault-tolerant architecture importing tested point.

Embodiment two: this execution mode is to the further specifying of embodiment one, in the embodiment one described in the step 4 according to the slope α value of almost function, the amount of jitter x that the WLAN navigation system is allowed _iWith the test point RSS matrix of any test point, the detailed process of fault-tolerant measures of dispersion that the ANN that calculates each ANN structure hides each node of layer is:

With the test point RSS matrix of any test point as input variable X _iThe ANN input layer of each ANN structure of substitution, and add the amount of jitter x that the WLAN navigation system is allowed to the ANN input layer _i, the input variable of analyzing the ANN input layer is by X _iBecome X _i+ x _iAfter the output variable Y of ANN output layer _k ³Variation tendency,

1. any node q when the hiding layer of ANN satisfies $\underset{i=1}{\overset{n_1}{\mathrm{\Σ}}}{W}_{\mathrm{Ij}}{X}_i\≥\mathrm{\α}+{\mathrm{\θ}}_j,$ $Y_j^2=1$ The time, if $\underset{i=1}{\overset{n_1}{\mathrm{\Σ}}}{W}_{\mathrm{Ij}}{x}_i\≥\mathrm{\α}+{\mathrm{\θ}}_j-\underset{i=1}{\overset{n_1}{\mathrm{\Σ}}}{W}_{\mathrm{Ij}}{X}_i,$ Y then _k ³No change, the fault-tolerant measures of dispersion of node q is 0, wherein, n ₁Expression ANN input layer number, W _IjExpression ANN input layer and ANN hide the weights between the layer, and j is a natural number, θ _jExpression ANN hides layer threshold value, Y _j ²Expression ANN hides the output variable of layer, Y _k ³The output variable of expression ANN output layer;

2. the node q when the hiding layer of ANN satisfies ${\mathrm{\&theta;}}_j\&le;\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i<\mathrm{\&alpha;}+{\mathrm{\&theta;}}_j,$ $Y_j^2=\frac{1}{\mathrm{\&alpha;}}(\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i-{\mathrm{\&theta;}}_j+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i)$ The time,

If $\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i\&GreaterEqual;\mathrm{\&alpha;}(\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k)$ Or $\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_q}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i<\mathrm{\&alpha;}{\mathrm{\&theta;}}_k,$ Y then _k ³No change, the fault-tolerant measures of dispersion of node q is 0, wherein, n ₂Expression ANN hides node layer number, W _JkExpression ANN hides the weights between layer and the ANN output layer, θ _kThe threshold value of expression ANN output layer;

If $\mathrm{\&alpha;}{\mathrm{\&theta;}}_k\&le;\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i<\mathrm{\&alpha;}(\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k),$ Output variable then $Y_k^3=\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j-\frac{1}{\mathrm{\&alpha;}}{\mathrm{\&theta;}}_k+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i,$ Y _k ³Variable quantity do $\mathrm{\&Delta;}=\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i,$ The fault-tolerant measures of dispersion of node q is not 0;

3. the node q when the hiding layer of ANN satisfies $\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i<{\mathrm{\&theta;}}_j,$ $Y_j^2=0$ The time,

If $\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i<{\mathrm{\&theta;}}_j-\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i,$ Y then _k ³No change, the fault-tolerant measures of dispersion of node q is 0;

4. preceding p node when the hiding layer of ANN satisfies $\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i\&GreaterEqual;\mathrm{\&alpha;}+{\mathrm{\&theta;}}_j,$ $Y_j^2=1$ And ANN hides the back n of layer ₂-p node satisfies ${\mathrm{\&theta;}}_j\&le;\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i<\mathrm{\&alpha;}+{\mathrm{\&theta;}}_j,$ $Y_j^2=\frac{1}{\mathrm{\&alpha;}}(\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i-{\mathrm{\&theta;}}_j+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i)$ The time, wherein, p is a natural number, and smaller or equal to n ₂,

If hiding preceding p node of layer, satisfies ANN $\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i\&GreaterEqual;0;$ ANN hides the back n of layer ₂-p node satisfies

More near α+θ _j,

More little perhaps

More near θ _j, Big more; Satisfy simultaneously

$\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j\&GreaterEqual;\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k$ Or

$\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{\mathrm{\&theta;}}_j<{\mathrm{\&theta;}}_k,$

Y then _k ³Complete no change, the fault-tolerant measures of dispersion of each node is 0;

If hiding preceding p node of layer, satisfies ANN $\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i\&GreaterEqual;0;$ ANN hides the back n of layer ₂-p node satisfies

More near α+θ _j,

More little perhaps

More near θ _j,

Big more; Satisfy simultaneously

${\mathrm{\&theta;}}_k\&le;\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j<\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k$ Then,

$Y_k^3=\frac{1}{\mathrm{\&alpha;}}\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{X}_i+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{x}_i-\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{\mathrm{\&theta;}}_j-\frac{1}{\mathrm{\&alpha;}}{\mathrm{\&theta;}}_k,$

Y _k ³Variable quantity do $\mathrm{\&Delta;}=\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{x}_i,$ The fault-tolerant measures of dispersion of each node of ANN output layer is not 0.

In this execution mode; The transfer function of the node of three layers of ANN adopts the almost function; Like formula (1), to ANN input layer input signal, the ANN input layer is left intact and passes to ANN with complete interconnected mode and hide layer; The hiding layer of ANN after weighted sum, is handled the output variable Y that obtains the hiding layer of ANN to input signal according to transfer function _j ², like formula (2), ANN hides the output variable Y of layer _j ²Pass to the ANN output layer, the output vector Y of ANN output layer with complete interconnected mode again _k ³, like formula (3),

$Y_j^2=f(\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}{Y}_i^1-{\mathrm{\&theta;}}_j)=f(\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}{X}_i-{\mathrm{\&theta;}}_j)$

$Y_k^3=f(\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{Y}_j^2-{\mathrm{\&theta;}}_k)---\left(3\right)$

Add the amount of jitter x that the WLAN navigation system is allowed to the ANN input layer _i, the variation of research ANN output layer, the generalization ability of three layers of ANN of analysis is to the output variable Y of the hiding layer of ANN _j ², the branch situation is discussed the output variable Y of ANN output layer _k ³, and then whether accomplished fault-tolerant measures of dispersion be 0 judgement.

To this execution mode, carry out instance analysis:

In indoor scene shown in Figure 2, experimentize, this experiment scene dimensioned area is 66.43 * 24.9m ², height 3m.And have 19 laboratories, 1 meeting room and 1 table tennis room, wherein; 1201-1227 representes the room respectively, 1203,1204,1209,1210,1218,1219,1225 and 1224 expression teacher offices, and all the other 19 numbers are represented 19 laboratories;

representes elevator, and the material of wall is a fragment of brick, aluminium alloy window and metallic door; Each access point AP is Linksys access point AP, and arrow mark is the position of 1 to No. 9 AP placement among the figure, and uses AP1; AP2 ..., AP9 indicates; Be fixed on the 2m height, support IEEE 802.11g standard, transmission rate 54Mbps.Receiver is 1.2m overhead.

Fig. 3 is that the reference point and the test point in rooms 1211 distributes.At each reference point and test point place, 3 minutes signal strength signal intensity (Received Signal Strength, RSS) measurement, twice of per second sampling have all been carried out.72 reference points and 56 test points are wherein arranged, the reference point in square marks " " expression rooms 1211, the test point in circular mark " zero " expression rooms 1211.100 sampled values are got at each reference point or test point place, and totally 100 * 128 sampled values are used for the ANN training and testing.

In indoor scene shown in Figure 2, test, use the wireless network card of Intel PRO/Wireless 3945ABG networkconnection to connect networking, under Windows XP operating system, collect from the signal strength signal intensity RSS of 9 Linksys AP value.

The present invention selects the rooms 1211 of Fig. 3 as the experiment place; Reference point and test point that the position is set is as shown in Figure 3; The locating area in experiment place shown in Figure 3 is irregular, and covering performance is better, and the diverse location in the zone detects the WLAN sample of signal value from different AP.Use NetStumbler signals collecting software, the RSS of each point measures and all adopts the poll acquisition mode, and AP of every connection gathers 3 minute datas in each position, twice of per second sampling.At each test point place, carry out 1 minute WLAN signals collecting, twice of per second sampling.

Get α=1, the maximum jitter amount that the WLAN navigation system is allowed is 5dB, is respectively 10,20,25 and at 30 o'clock at hiding node layer number, calculates their corresponding fault-tolerant satisfactions respectively, and by formula test point error E, wherein E are calculated in (4) _x, E _yBe respectively the experiment elements of a fix and the actual location coordinate difference of X and Y direction.

$E=\sqrt{{E_x}^2+{E_x}^2}---\left(4\right)$

Hiding the node layer number is 10 o'clock, and the node that fault-tolerant measures of dispersion does not satisfy corresponding requirements has 2, so fault-tolerant satisfaction rate is 80%, the test point error of this moment is 2.86m.

Hiding the node layer number is 20 o'clock, and the node that fault-tolerant measures of dispersion does not satisfy corresponding requirements has 1, so fault-tolerant satisfaction rate is 95%, the test point error of this moment is 2.74m.

Hiding the node layer number is 25 o'clock, and the node that fault-tolerant measures of dispersion does not satisfy corresponding requirements has 0, so fault-tolerant satisfaction rate is 100%, the test point error of this moment is 0.81m.

Hiding the node layer number is 30 o'clock, and the node that fault-tolerant measures of dispersion does not satisfy corresponding requirements has 1, so fault-tolerant satisfaction rate is 97%, the test point error of this moment is 2.11m.

Contrast is different hides the fault-tolerant satisfaction that the node layer number calculates, and corresponding test point is as shown in table 1 in WLAN indoor positioning error.

The different fault-tolerant calculation result's contrasts of hiding the node layer number of table 1

Hide the node layer number	10	20	25	30
					Fault-tolerant satisfaction	80％	95％	100％	97％
The test point position error	2.86m	2.74m	0.81m	2.11m

Obviously, fault-tolerant satisfaction is high more, and the fault-tolerance of ANN is good more, and the error of test point is more little.With respect to relying on empirical value or qualitative experiment to confirm to hide the ANN of node layer number, using fault-tolerant satisfaction to hide the node layer number as Standard Selection has very big effect for the indoor position accuracy that improves the ANN method.

Claims (2)

1. based on the indoor three-layer artificial neural network ANN of the WLAN of fault-toleranr technique intelligent locating method, it is characterized in that its position fixing process is:

Step 1: in WLAN indoor positioning environment; N reference point, a U test point and M access point AP are set; Define the position coordinates of each reference point and each test point simultaneously; Guarantee that each reference point is covered by the signal that two or more access point AP send, guarantee that simultaneously each test point is also covered by the signal that two or more access point AP send;

Step 2: based on the number M of access point AP and the dimension initialization artificial neural network ANN structure of position coordinates; Be specially: the number M of getting access point AP is the node number of artificial neural network ANN input layer, and the dimension that coordinate is put in fetch bit is the node number of artificial neural network ANN output layer;

Step 3: will form reference point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same reference point; And the position coordinates of each reference point and the reference point RSS matrix group of corresponding reference point become artificial neural net ANN training set; Utilize said artificial neural net ANN training set to train a plurality of artificial neural net ANN; Obtain a plurality of artificial neural net ANN structures, it is different that the artificial neural net ANN of said a plurality of artificial neural net ANN hides the node layer number;

Step 4: will form test point RSS matrix about a plurality of signal strength signal intensity RSS values from diverse access point AP of same test point; According to the slope α value of almost function, amount of jitter xi that the WLAN navigation system is allowed and the test point RSS matrix of any test point; The artificial neural net ANN that calculates each artificial neural net ANN structure hides the fault-tolerant measures of dispersion of each node of layer; And the fault-tolerant measures of dispersion of calculating the hiding layer of artificial neural net ANN of each artificial neural net ANN structure is 0 node number; Said fault-tolerant measures of dispersion is that the position coordinates of artificial neural net ANN output layer output was constant after 0 expression artificial neural net ANN input layer added the amount of jitter that the WLAN navigation system allows;

Step 5: the fault-tolerant measures of dispersion of hiding layer according to the artificial neural net ANN of each artificial neural net ANN structure is total node number that the artificial neural net ANN of 0 node number and each artificial neural net ANN structure hides layer; The artificial neural net ANN of the said fault-tolerant satisfaction λ of fault-tolerant satisfaction record that calculates each artificial neural net ANN structure when the highest hides the node layer number, preserves corresponding artificial neural net ANN structure and is artificial neural net ANN fault-tolerant architecture with said artificial neural net ANN organization definition;

Step 6: from all test points, choose tested point,, obtain the elements of a fix of tested point to the test point RSS matrix of artificial neural net ANN fault-tolerant architecture importing tested point.

2. the indoor three-layer artificial neural network ANN of the WLAN based on fault-toleranr technique according to claim 1 intelligent locating method, it is characterized in that described in the step 4 according to the slope α value of almost function, the amount of jitter x that the WLAN navigation system is allowed _iWith the test point RSS matrix of any test point, the detailed process of fault-tolerant measures of dispersion that the artificial neural net ANN that calculates each artificial neural net ANN structure hides each node of layer is:

With the test point RSS matrix of any test point as input variable X _iThe artificial neural net ANN input layer of each artificial neural net ANN structure of substitution, and add the amount of jitter x that the WLAN navigation system is allowed to artificial neural net ANN input layer _i, the input variable of analyst's artificial neural networks ANN input layer is by X _iBecome X _i+ x _iAfter the output variable of artificial neural net ANN output layer

Variation tendency,

1. when any one node q of the hiding layer of artificial neural network ANN satisfies

If

Then

No change, the fault-tolerant measures of dispersion of node q is 0, wherein, n ₁Expression artificial neural net ANN input layer number, W _IjExpression artificial neural net ANN input layer and artificial neural net ANN hide the weights between the layer, and j is a natural number, θ _jExpression artificial neural net ANN hides layer threshold value,

Expression artificial neural net ANN hides the output variable of layer,

The output variable of expression artificial neural net ANN output layer;

2. the node q when the hiding layer of artificial neural net ANN satisfies

$Y_j^2=\frac{1}{\mathrm{\&alpha;}}(\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{X}_i-{\mathrm{\&theta;}}_j+\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{x}_i)$ The time,

If $\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i\&GreaterEqual;\mathrm{\&alpha;}(\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k)$ Or $\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i<\&GreaterEqual;\mathrm{\&alpha;}{\mathrm{\&theta;}}_k,$ Then

No change, the fault-tolerant measures of dispersion of node q is 0, wherein, n ₂Expression artificial neural net ANN hides node layer number, W _JkExpression artificial neural net ANN hides the weights between layer and the artificial neural net ANN output layer, θ _kThe threshold value of expression artificial neural net ANN output layer;

If $\mathrm{\&alpha;}{\mathrm{\&theta;}}_k\&le;\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j+\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i<\mathrm{\&alpha;}(\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k),$

Output variable then $Y_k^3=\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{X}_i-\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j-\frac{1}{\mathrm{\&alpha;}}{\mathrm{\&theta;}}_k+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Ij}}{W}_{\mathrm{Jk}}{x}_i,$

Variable quantity do

The fault-tolerant measures of dispersion of node q is not 0;

3. when the node q of the hiding layer of artificial neural network ANN satisfies

If

be

no change then, the fault-tolerant measures of dispersion of node q is 0;

4. preceding p node when the hiding layer of artificial neural net ANN satisfies

And artificial neural net ANN hides the back n of layer ₂-p node satisfies

The time, wherein, p is a natural number, and smaller or equal to n ₂,

If hiding preceding p node of layer, satisfies artificial neural net ANN Artificial neural net ANN hides the back n of layer ₂-p node satisfies

More near α+θ _j,

More little perhaps

More near θ _j, Big more; Satisfy simultaneously

$\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j\&GreaterEqual;\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k$ Or

$\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{\mathrm{\&theta;}}_j\&GreaterEqual;\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k,$

complete no change then, the fault-tolerant measures of dispersion of each node is 0;

Satisfied

artificial neural net ANN hides if artificial neural net ANN hides preceding p node of layer

The back n of layer ₂-p node satisfies

More near α+θ _j,

More little perhaps

More near θ _j,

is big more; Satisfy simultaneously

${\mathrm{\&theta;}}_k\&le;\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{X}_i+\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{W}_{\mathrm{Ij}}{x}_i-\frac{1}{\mathrm{\&alpha;}}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{Jk}}{\mathrm{\&theta;}}_j<\mathrm{\&alpha;}+{\mathrm{\&theta;}}_k$ Then,

$Y_k^3=\frac{1}{\mathrm{\&alpha;}}\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{X}_i+\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n_1}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{W}_{\mathrm{ij}}{x}_i-\frac{1}{{\mathrm{\&alpha;}}^2}\underset{j=p+1}{\overset{n_2}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}{\mathrm{\&theta;}}_j-\frac{1}{\mathrm{\&alpha;}}{\mathrm{\&theta;}}_k,$

The variation is ANN artificial neural network output layer of fault tolerance for each node the amount of difference is not zero.

Images