CN105101408A - Indoor Positioning Method Based on Distributed AP Selection Strategy - Google Patents

Abstract

The invention discloses an indoor positioning method based on a distributed AP selection strategy. The method comprises the following steps of using mobile equipment to collect k times of signal intensities of AP nodes on each reference point and processing data; equally dividing a target area into m subareas, classifying fingerprint information of the different subareas, using a fingerprint information base to train a subarea module and selecting the corresponding subarea according to RSS fingerprint information; calculating a correlation of each AP node and each subarea, sorting correlation coefficients, selecting the AP node whose correlation coefficient is greater than Ptau as a positioning node of the subarea; in each subarea, taking the selected positioning node as input of a DBN model and a corresponding position point as output of the DBN model, training the DBN model and constructing a positioning prediction model. By using the method, big noise and the AP node with a weak position distinguishing capability can be effectively removed; positioning precision of an overall WIFI indoor positioning system is increased and algorithm operation time is shortened.

Description

Indoor Positioning Method Based on Distributed AP Selection Strategy

technical field

The invention relates to an indoor positioning method, in particular to a WIFI indoor positioning method based on a distributed AP selection strategy.

Background technique

Indoor positioning is a technology used to obtain the location information of indoor target objects, and has broad application prospects in civilian and military fields. Common positioning algorithms are mainly based on Received Signal Strength Indication (RSSI), Time of Arrival (TOA), Time Difference of Arrival (TDOA), Angle of Arrival (AOA) and other technologies. Among them, the positioning algorithm based on RSSI has the advantages of low power consumption, low cost and easy implementation, and is widely used in wireless indoor positioning.

The positioning algorithm using RSSI is generally divided into: a positioning algorithm based on a signal propagation model and a positioning algorithm based on a fingerprint model. The traditional location algorithm based on the signal propagation model mainly obtains a large number of sample data, uses the traditional path loss model to establish the functional relationship between the received signal strength indicator (RSSI) and the distance between nodes, and then estimates the location information of the target. Since the wireless signal is easily affected by indoor environment changes during the propagation process, the RSSI value has obvious fluctuations. At the same time, it is difficult to determine the path attenuation index and environmental factors in the traditional path loss model. These factors will affect the ranging accuracy of the model. This leads to an increase in positioning error.

The positioning algorithm based on the fingerprint model is based on a large number of actual sample data to establish a positioning model according to the statistical analysis method, in which the learning model is widely used as a statistical analysis method to train the positioning prediction model, and finally use the positioning prediction model to target The location is predicted. The essence of the method is to establish a mapping relationship between a group of wireless signal strengths and target positions. Zhou et al. used Artificial Neural Network (ANN) as a learning and prediction tool, combined with region division technology to predict the target position, and achieved better positioning prediction results. Sang Nan and others established a nonlinear relationship between RSSI and location information, and used Support Vector Machine (SupportVectorMachine, SVM) to learn and predict location information, effectively reducing positioning errors.

Among them, the positioning algorithm based on the fingerprint model has the advantages of high positioning accuracy, full use of existing facilities, and little impact on users for upgrades and maintenance, and has been widely used. The location fingerprint positioning algorithm is mainly divided into two steps: offline measurement stage and online positioning stage .

In a wireless network with a large number of AP nodes deployed, in order to obtain the target position, the data processed by the above methods are all high-dimensional vectors (the number of AP nodes). At the same time, due to factors such as occlusion of objects and node failures, the positioning effect of using all AP nodes as feature input is not necessarily optimal. Redundant AP nodes increase the difficulty of position estimation, easily causing overfitting problems, and Increased time and space complexity. To solve this problem, an appropriate selection mechanism is used to screen AP nodes, and better AP nodes are obtained as feature inputs, which not only achieves the effect of dimensionality reduction, but also improves the positioning accuracy. In traditional methods, the positioning capability of an AP is usually judged according to the average signal strength received from an AP. The greater the average signal strength, the stronger the positioning capability of the AP. In practice, it is found that this AP selection method is incorrect. For example, the signal strength of an AP is relatively large everywhere in the positioning area, but its fluctuation is small. Although the average signal strength of this AP is large, its positioning ability is weak. . At the same time, the traditional method does not take into account that an AP has different positioning contributions to different locations in the positioning environment. For example, an AP may have a good positioning ability for a certain area, but poor positioning ability for other areas.

Contents of the invention

In order to solve the above technical problems, the object of the present invention is to provide an indoor positioning method based on a distributed AP selection strategy, which can avoid the influence of relatively large noise and AP nodes with weak position resolution capabilities, and at the same time reduce the calculation time required for positioning and Improve positioning accuracy.

Technical scheme of the present invention is:

An indoor positioning method based on a distributed AP selection strategy, comprising the following steps:

S01: Arrange D AP nodes in the indoor environment to ensure that any point in the indoor environment is covered by signals from three or more AP nodes, and at the same time divide the target area into N small areas, with the center of the small area as Signal strength collection points, a total of N reference points, used to collect sample data in the offline phase;

S02: Establish a two-dimensional rectangular coordinate system for the target area, obtain the coordinate positions of N reference points, and use the mobile device to collect the signal strength of the AP node for k times at each reference point, and process the data;

S03: Divide the target area into m sub-areas by means of an equal area, classify the fingerprint information of different sub-areas, use the fingerprint information database to train the sub-area module, and select the corresponding sub-area according to the RSS fingerprint information;

S04: Calculate the correlation between each AP node and each sub-region, sort the correlation coefficients, and select the AP node with a correlation coefficient greater than P _τ as the positioning node of the sub-region, where P _τ represents the correlation coefficient threshold;

S05: In each sub-region, use the selected positioning node as the input of the DBN model, and the corresponding position point as the output of the DBN model, train the DBN model, and build a positioning prediction model;

S06: In the online prediction stage, obtain the RSS fingerprint information received by the mobile device and use it as the input vector of the sub-region module, and the output vector of the sub-region module is the sub-region to which the RSS fingerprint information belongs;

S07: Use the signal strength of the AP node corresponding to the sub-region obtained in step S06 as an input vector of the DBN model, and use the trained DBN model to obtain the position of the target.

Preferably, the step S02 specifically includes the following steps:

S21: Receive the signal strength RSS value of each AP node at each reference point to form a D-dimensional signal vector RSS ^D ; each reference point collects signal strength values k times to form a k*D signal strength matrix, and the matrix Row i and column j represent the signal strength RSS value of the jth AP node received in the i-th collection; i is a positive integer less than or equal to k; j is a positive integer less than or equal to D;

RSS ^D ＝(rss ₁ ,rss ₂ ,rss ₃ ,...,rss _j ,...,rss _D )(1)

In the formula: rss _j ∈ [-100,0] represents the signal strength of the jth AP node received at the reference point.

S22: Build a fingerprint database. The training data set contains N records, each record is expressed as a vector r, and the vector contains the signal strength of the available AP nodes and the location of the sampling point:

r=(RSS ⁱ ,L ⁱ )=(rss ₁ ,rss ₂ ,rss ₃ ,... _, rss _D ,L ⁱ )(2)

In the formula: RSS ⁱ represents the RSS set received in the i-th collection, D represents the number of AP nodes, L ⁱ represents the location label corresponding to the RSS ^D vector.

Preferably, the step S03 specifically includes the following steps:

S31: Divide the positioning area S into m sub-areas, and the sub-areas are defined as follows:

$\begin{array}{cccc}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; \}</span>& <span\; class="notranslate">\; \&ForAll;</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; S</span>}& \mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; d</span>}& \mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&cup;</span>}}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span>\end{array}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

In the formula: c represents the cth sub-region, k represents the number of position points in the sub-region, and R _c represents the set of reference position points in the c-th sub-region; Represents the information of the kth position point in the cth subregion;

S32: Classify the signal strength in the fingerprint database according to different sub-regions, and establish a record information table of sub-regions and signal strength.

S33: Construct a sub-area module, determine the BP neural network model and train the established BP neural network according to the record information of the sub-area identification and signal strength, use the signal strength RSSI as input, and use the corresponding sub-area identification area as an output to train the BP neural network, And modify the parameters of the BP neural network during training to reflect the RSSI-area relationship;

S34: Use the actual received signal strength value RSSI and the corresponding sub-area identification area to repeatedly train and verify the established BP neural network, and encapsulate the trained BP neural network into a fixed function, and the input of this function is the received signal strength RSSI, the output is the corresponding sub-area identifier.

Preferably, the BP neural network is a five-layer neural network, the number of hidden layers is 2 layers, the number of nodes in the input layer is D, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is 6, respectively. 4. The BP neural network structure of D: 6: 4: 1 is adopted, the training function is the traincgf algorithm, and the number of training times and the target error are set to 100 and 0.00004 respectively.

Preferably, the step S04 specifically includes the following steps:

S41: In this sub-area, scan the signals of the jth AP node at k positions in the area, and the correlation coefficient of the AP node in this sub-area is defined as follows:

$\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}& \mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ...</span>\mathrm{<span\; class="notranslate">\; D.</span>}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

In the formula: P _j represents the correlation between the jth AP node and the sub-region, x _i represents the X-axis coordinate of the i-th position point, x represents the mean value of the X-axis coordinates of the k position points in the region, and y _i Represents the Y-axis coordinate of the i-th location point, y represents the mean value of the Y-axis coordinates of the k location points in the area, rss _ji represents the mean value of the signal strength received by the i-th location point from the j-th AP node, rss _j Indicates the mean value of the signal strength of the jth AP node received by k position points, k represents the number of position points in the sub-area, and D represents the number of AP nodes in the positioning area;

S42: In each sub-region, the correlation coefficient vectors of D AP nodes are expressed as:

P(AP)＝[P ₁ ,P ₂ ,...,P _D ](5)

If P _j >P _τ , it means that the jth AP node is related to the sub-region, otherwise, it is not related, and P _τ represents the threshold of the correlation coefficient.

Preferably, the step S05 specifically includes the following steps:

S51: Obtain all position point information in the sub-area and the signal strength of the corresponding AP node in the fingerprint database, as the training data set of the sub-area;

S52: Determine the DBN model and use the training data set of the sub-region to train and establish the DBN model, use the received signal strength RSSI of the corresponding AP node as input, and the corresponding location point Location as output to train the established DBN model, and in Modify the parameters of the DBN model during training, so that the final positioning model can correctly reflect the relationship between RSSI-Location;

S53: Use the received signal strength value RSSI of the corresponding AP node and the corresponding location point Location to repeatedly train and verify the established DBN model; adjust the relationship between neurons through the energy function between the visible layer and the hidden layer Weights, and finally through the backpropagation algorithm to fine-tune the weights between the networks.

Preferably, the DBN model is a 5-layer network, wherein the number of layers in the hidden layer is 3 layers, the number of nodes in the input layer is the number of selected AP nodes, the node data in the output layer is 2, and the number of nodes in the hidden layer The numbers are 10, 6, 4 respectively.

Preferably, the DBN model includes RBM1 and RBM2, and both the RBM1 module and the RBM2 module include a visible layer and a hidden layer. When the signal strength set of the AP node is input to the visible layer of RBM1, the hidden layer will extract the input data through the weight of the connection features, and adjust the weights between neurons through the energy function between the visible layer and the hidden layer; the output of the hidden layer of RBM1 will be used as the input of the visible layer of RBM2, and the hidden layer of RBM2 will further extract the depth of the classified data Hierarchical features, the trained DBN model is encapsulated into a fixed function, the input of this function is the received signal strength RSSI, and the output is the location point of the target.

The advantages of the present invention are:

1. This method can effectively remove AP nodes with large noise and weak position resolution ability. By dividing the indoor area into several separate sub-areas, and calculating the correlation between the sub-area and all AP nodes, the AP node with excellent correlation is selected as the training node of the sub-area, and finally through the deep belief network (DeepBeliefNetwork, DBN) The model performs positioning model training. Effectively improve the positioning accuracy of the entire WIFI indoor positioning system.

2. The algorithm is simplified and the running time of the algorithm is reduced.

Description of drawings

The present invention will be further described below in conjunction with accompanying drawing and embodiment:

Fig. 1 is the flow chart that positioning realizes of the present invention;

Fig. 2 is the general frame diagram of the indoor positioning algorithm of the distributed AP selection strategy that the present invention proposes;

Fig. 3 is a schematic diagram of an indoor scene of the present invention;

Figure 4 is a block diagram of the DBN model structure.

Among them: 1. Station, 2. AP node, 3. Pillar.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary only, and are not intended to limit the scope of the present invention. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concept of the present invention.

Example:

As shown in Figure 1, an indoor positioning method based on a distributed AP selection strategy includes the following steps:

S01: Arrange D AP nodes in the indoor environment to ensure that any point in the indoor environment is covered by signals from three or more AP nodes, and at the same time divide the target area into N small areas, with the center of the small area as Signal strength collection points, a total of N reference points, used to collect sample data in the offline phase;

S02: Establish a two-dimensional rectangular coordinate system for the target area, obtain the coordinate positions of N reference points, and use the mobile device to collect the signal strength of the AP node for k times at each reference point, and process the data;

S03: Divide the target area into m sub-areas by means of an equal area, classify the fingerprint information of different sub-areas, use the fingerprint information database to train the sub-area module, and select the corresponding sub-area according to the RSS fingerprint information;

S04: Calculate the correlation between each AP node and each sub-region, and sort the correlation coefficients, select the AP node with a correlation coefficient greater than P _τ , as the positioning node of the sub-region, and P _τ represents the threshold of the correlation coefficient;

S05: In each sub-region, use the selected positioning node as the input of the DBN model, and the corresponding position point as the output of the DBN model, train the DBN model, and build a positioning prediction model;

S06: In the online prediction stage, obtain the RSS fingerprint information received by the mobile device and use it as the input vector of the sub-region module, and the output vector of the sub-region module is the sub-region to which the RSS fingerprint information belongs;

S07: Use the signal strength of the AP node corresponding to the sub-region obtained in step S06 as an input vector of the DBN model, and use the trained DBN model to obtain the position of the target.

The experimental scene of positioning data collection is set in the comprehensive laboratory. The experiment is carried out in the indoor scene shown in Figure 3. The terminal device for data collection is a Samsung smart phone (I9228), and the operating system is Android4.1.2. The laboratory is about 12.8m long, 12.5m wide, and 3m high. There are station 1, computers and other office supplies in the room. The height of AP node 2 is kept at 1.6m. The layout of AP node 2 is shown in Figure 3. In this area, the signals of 25 AP nodes 2 can be collected, and the propagation paths of the signals of some AP nodes 2 are non-line-of-sight, and there are pillars 3 blocking them. During the experiment, the target area was divided into 144 small areas, each with a size of 1m×1m, and the center of the small area was used as the signal strength collection point, a total of 144 reference points were used to collect sample data in the offline stage. To ensure the accuracy of sample data, each signal collection point acquires 600 signal sets of AP node 2, once per second.

Receive the signal strength RSS value of each AP node at each reference point to form a 25-dimensional signal vector RSS ^D ; each reference point collects 600 signal strength values to form a 600*25 signal strength matrix, and the i-th matrix of the matrix The jth column of the row indicates the signal strength RSS value of the jth AP node received in the ith acquisition; i is a positive integer less than or equal to k; j is a positive integer less than or equal to D;

RSS ²⁵ = (rss ₁ , rss ₂ , rss ₃ ,..., rss ₂₅ )(1)

In the formula: rss _j ∈ [-100,0] represents the signal strength of the jth AP node received at the reference point.

Build a fingerprint database. The training data set contains N records. Each record can be expressed as a vector r, which contains the signal strength of the available AP nodes and the location of the sampling point:

$\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; RSS</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

In the formula: D represents the number of AP ^nodes , Li represents the location label corresponding to the RSS ^D vector. If the mobile device does not receive the signal of an AP node, -100 is used to fill the signal strength value of the AP node by default. There are two reasons why the mobile device does not receive the signal of a certain AP node: one is that the AP node fails; the other is that the AP node is blocked by obstacles.

The established fingerprint database is shown in the table:

L _{1 x1} Y ₁ rss ₁ ... rss ₂₅ ... ... ... ... ... ... L _N X _N Y _N rss ₁ ... rss ₂₅

Among them, L ₁ to L _N are selected N reference points, and the information of each reference point includes location information and RSS values of D APs.

Assuming that the area is divided into 4 sub-areas, the definition of the sub-areas is shown in Equation 3:

$\begin{array}{cccc}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; \}</span>& <span\; class="notranslate">\; \&ForAll;</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; S</span>}& \mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; d</span>}& \mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{<span\; class="notranslate">\; \&cup;</span>}}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

In the formula: c represents the cth sub-region, k represents the number of position points in the sub-region, S represents the positioning area (target region), and R _c represents the set of reference position points in the c-th sub-region; Indicates the information of the kth location point in the cth subregion.

According to different sub-regions, the signal strength in the fingerprint database is classified according to different sub-regions, and the record information of sub-regions and signal strength is expressed as:

area ₁ rss ₁ ... rss ₂₅

area ₁ rss ₁ ... rss ₂₅ ... ... ... ... area ₄ rss ₁ ... rss ₂₅

area ₁ indicates the first sub-area, and rss ₁ indicates the signal strength of the first AP node.

Construct the sub-area module, determine the BP neural network model and train the established BP neural network according to the record information of the sub-area identification and signal strength, use the signal strength RSSI as input, and use the corresponding sub-area identification area as the output to train the BP neural network, and Modify the parameters of the BP neural network during training to reflect the RSSI-area relationship; the BP neural network used is a five-layer neural network; in the five-layer neural network used, the number of hidden layers is 2 layer, the number of nodes in the input layer is 25, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is 6 and 4 respectively, that is, the BP neural network structure of 25:6:4:1 is adopted, and the training function is the traincgf algorithm , the number of training times and the target error are set to 100 and 0.00004 respectively, and the sub-area module of the BP network is shown in Figure 2.

Use the actual received signal strength value RSSI and the corresponding sub-area identification area to repeatedly train and verify the established BP neural network, and encapsulate the trained BP neural network into a fixed function. The input of this function is the received signal strength RSSI, The output is the corresponding sub-area ID.

Assuming that in this sub-area, the signal of the j-th AP node at k positions in the scanning area is scanned, then the correlation coefficient of the AP node in this area is defined as follows:

$\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; rss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}& \mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1...25</span>}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

In the formula: P _j represents the correlation between the jth AP node and the area, x _i represents the X-axis coordinate of the i-th position point, x represents the mean value of the X-axis coordinates of the k position points in the area, and y _i represents The Y-axis coordinate of the i-th location point, y represents the mean value of the Y-axis coordinates of the k location points in the area, rss _ji represents the mean value of the signal strength received by the i-th location point from the j-th AP node, and rss _j represents The mean value of the signal strength of the jth AP node received by the k position points in the area, and k represents the number of position points in the sub-area.

In each sub-region, the correlation coefficient vector of 25 AP nodes is expressed as:

P(AP)=[P ₁ , P ₂ , . . . ,] (5)

If P _j >P _τ , it means that the jth AP node is related to the sub-region, otherwise it is not related. The AP nodes related to this area are selected as feature input for training the localization model.

P _τ represents the threshold value of the correlation coefficient, which is determined by trial-and-error method.

Obtain all location point information in the sub-area and the signal strength of the corresponding AP node in the fingerprint database, and use it as the training data set for the sub-area.

Determine the DBN model and use the training data set of the sub-area to train and build the DBN model, take the received signal strength RSSI of the corresponding AP node as input, and the corresponding location point Location as the output to train the established DBN model, and in the training Modify the parameters of the DBN model so that the final location model can correctly reflect the relationship between RSSI-Location; the DBN model used is a 5-layer network, in which the number of layers in the hidden layer is 3 layers, and the number of nodes in the input layer is 25. The node data of the output layer is 2, the node numbers of the hidden layer are 10, 6, and 4 respectively, and the network structure is 25:10:6:4:2.

Use the actual received signal strength value RSSI of the corresponding AP node and the corresponding location point Location to repeatedly train and verify the established DBN model.

The weights between neurons are adjusted through the energy function between the visible layer and the hidden layer, and finally the weights between the networks are fine-tuned through the backpropagation algorithm. The DBN model is shown in Figure 4, including the RBM1 module and the RBM2 module, and both the RBM1 module and the RBM2 module include a visible layer and a hidden layer. The training process is as follows: when the signal strength set of the AP node is input to the visible layer of RBM1, the hidden layer will extract the characteristics of the input data through the weight of the connection, and adjust the neural network through the energy function between the visible layer and the hidden layer. The weight between elements; the output of the hidden layer of RBM1 will be used as the input of the visible layer of RBM2, and the new hidden layer will further extract the deep features of the classification data. Encapsulate the trained DBN model into a fixed function, the input of this function is the received signal strength RSSI, and the output is the position point of the target.

The RSS fingerprint information received at the test point is used as the input vector of the sub-region module, and the output vector of the sub-region module is the sub-region to which the fingerprint information belongs.

Use the sub-region obtained in step 6, use the signal strength of the AP node corresponding to the sub-region as the input vector of the DBN model, and use the trained DBN model to obtain the position of the final target.

It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and not to limit the present invention. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention shall fall within the protection scope of the present invention. Furthermore, it is intended that the appended claims of the present invention embrace all changes and modifications that come within the scope and metesques of the appended claims, or equivalents of such scope and metes and bounds.

Claims (8)

1. based on an indoor orientation method for distributed AP selection strategy, it is characterized in that, comprise the following steps:

S01: arrange D AP node in indoor environment, ensure the quorum sensing inhibitor that optional position in indoor environment o'clock is sent by the AP node of more than three or three, this target area is divided into N number of zonule simultaneously, with center, zonule for signal strength signal intensity collection point, N number of reference point altogether, for off-line phase collecting sample data;

S02: set up two-dimensional direct angle coordinate system to target area, obtains the coordinate position of N number of reference point, and in each reference point, utilize mobile device to gather the signal strength signal intensity of k AP node, and processes data;

S03: utilized target area the method for homalographic to be divided into m sub regions, classified by the finger print information of different subregion, utilizes training subregion, finger print information storehouse module, selects corresponding subregion according to RSS finger print information;

S04: the correlation calculating each AP node and each sub regions, and relative coefficient is sorted, select relative coefficient to be greater than P _τaP node, as the location node of this subregion, wherein, P _τrepresent the threshold value of relative coefficient;

S05: in every sub regions, utilizes the location node chosen as the input of DBN model, and corresponding location point is as the output of DBN model, and training DBN model, builds location prediction model;

S06: on-line prediction stage, obtains RSS finger print information that mobile device receives and as the input vector of subregion module, the subregion of output vector belonging to this RSS finger print information of subregion module;

S07: the signal strength signal intensity of the AP node utilizing the subregion that obtains in step S06 corresponding, as the input vector of DBN model, utilizes the DBN model of having trained to obtain the position of target.

2. the indoor orientation method based on distributed AP selection strategy according to claim 1, is characterized in that, described step S02 specifically comprises the steps:

S21: the signal strength signal intensity RSS value receiving each AP node in each reference point, forms D dimensional signal vector RSS ^d; Each reference point gathers k signal strength values, constitutes the signal strength signal intensity matrix of k*D, and the signal strength signal intensity RSS value receiving a jth AP node in i-th collection is shown in the i-th row jth list of its matrix; I is the positive integer being less than or equal to k; J is the positive integer being less than or equal to D;

RSS ^D＝(rss ₁,rss ₂,rss ₃,…,rss _j,…，rss _D)(1)

In formula: rss _j∈ [-100,0] represents that reference point place receives the signal strength signal intensity of a jth AP node.

S22: build fingerprint database, training data is concentrated and is comprised N bar record, and every bar record is expressed as vectorial r, comprises the available signal strength signal intensity of AP node and the position of sampled point in vector:

r＝(RSS ⁱ,L ⁱ)＝(rss ₁,rss ₂,rss ₃,…,rss _D,L ⁱ)(2)

In formula: RSS ⁱrepresent the RSS set received in gathering for i-th time, D represents the number of AP node, L ⁱrepresent and correspond to RSS ^dthe location tags of vector.

3. the indoor orientation method based on distributed AP selection strategy according to claim 1, is characterized in that, described step S03 specifically comprises the following steps:

S31: locating area S is divided into m sub regions, shown in subregion is defined as follows:

$\begin{array}{cccc}{R}_c=\left\{L_1^c,L_2^c,...,L_k^c\right\}& \&ForAll;L_k^c\&Element;S& and& S=\underset{c=1}{\overset{m}{\&cup;}}{R}_c\end{array}---\left(3\right)$

In formula: c represents c sub regions, k represents the number of location point in this subregion, R _crepresent reference position point set in c sub regions; represent the information of a kth location point in c sub regions;

S32: signal strength signal intensity in fingerprint database classified according to different subregions, sets up the record information list of subregion and signal strength signal intensity.

S33: build subregion module, determine BP neural network model and train with the recorded information of signal strength signal intensity the BP neural net set up according to subregion mark, using signal strength signal intensity RSSI as input, corresponding subregion mark area is as output training BP neural net, and the parameter revising BP neural net in training makes it reflect the relation of RSSI-area;

S34: identify area with the actual signal strength values RSSI received with corresponding subregion and go repetition training and verify the BP neural net set up, the BP neural net of having trained is packaged into a fixing function, this function is input as received signal strength RSSI, exports and is corresponding subregion mark.

4. the indoor orientation method based on distributed AP selection strategy according to claim 3, it is characterized in that, described BP neural net is five layers of neural net, and the number of plies of hidden layer is 2 layers, and the interstitial content of input layer is D, the interstitial content of output layer is 1, the interstitial content of hidden layer is respectively 6,4, namely adopts the BP neural network structure of D:6:4:1, training function is traincgf algorithm, and frequency of training and target error are set to 100 and 0.00004 respectively.

5. the indoor orientation method based on distributed AP selection strategy according to claim 1, is characterized in that, described step S04 specifically comprises the following steps:

S41: in this subregion, the signal of a jth AP node on k location point in scanning area, this AP node is defined as follows at the relative coefficient of this subregion:

$\begin{array}{cc}{P}_j=\frac{\left|\underset{i=1}{\overset{k}{\&Sigma;}}\left(x_i-x\right)\left(y_i-y\right)\left({\mathrm{rss}}_{ji}-{\mathrm{rss}}_j\right)\right|}{\sqrt{\underset{i=1}{\overset{k}{\&Sigma;}}{\left(x_i-x\right)}^2\underset{i=1}{\overset{k}{\&Sigma;}}{(y_i-y)}^2\underset{i=1}{\overset{k}{\&Sigma;}}{\left({\mathrm{rss}}_{ji}-{\mathrm{rss}}_j\right)}^2}}& j=1...D\end{array}---\left(4\right)$

In formula: P _jrepresent the correlation of a jth AP node and this subregion, x _irepresent the X-axis coordinate of i-th location point, x represents the average of the X-axis coordinate of k location point in this region, y _irepresent the Y-axis coordinate of i-th location point, y represents the average of the Y-axis coordinate of k location point in this region, rss _jirepresent that i-th location point receives the average of the signal strength signal intensity of a jth AP node, rss _jrepresent the signal strength signal intensity average of the jth AP node that k location point receives, k represents the number of location point in this subregion, and D represents the number of AP node in locating area;

S42: in every sub regions, the relative coefficient vector representation of D AP node is:

P(AP)＝[P ₁,P ₂,...,P _D](5)

If P _j>P _τrepresent that a jth AP node is relevant to this subregion, on the contrary then uncorrelated, P _τrepresent the threshold value of relative coefficient.

6. the indoor orientation method based on distributed AP selection strategy according to claim 1, is characterized in that, described step S05 specifically comprises the following steps:

S51: the signal strength signal intensity obtaining the AP node of all location point information and correspondence in this subregion in fingerprint database, as the training dataset of this subregion;

S52: determine DBN model and go to train with the training dataset of this subregion and set up DBN model, using the signal strength signal intensity RSSI of the corresponding A P node received as input, corresponding location point Location goes to train the DBN model set up as exporting, and in training, revise the parameter of DBN model, make final location model correctly can reflect the relation of RSSI-Location;

S53: go repetition training also to verify the DBN model set up with signal strength values RSSI and the corresponding location point Location of the actual corresponding A P node received; Adjust the weights between neuron by the energy function between visible layer and hidden layer, carry out the weights between trim network finally by back-propagation algorithm.

7. the indoor orientation method based on distributed AP selection strategy according to claim 6, it is characterized in that, described DBN model is 5 layer networks, wherein the number of plies of hidden layer is 3 layers, the interstitial content of input layer is the number of the AP node chosen, and the node data of output layer is 2, and the interstitial content of hidden layer is respectively 10,6,4.

8. the indoor orientation method based on distributed AP selection strategy according to claim 7, it is characterized in that, DBN model comprises RBM1 and RBM2, RBM1 module and RBM2 module all comprise visible layer and hidden layer, when the signal strength signal intensity set of AP node is input to the visible layer of RBM1, weights by connecting are extracted the feature of input data by hidden layer, and adjust the weights between neuron by the energy function between visible layer and hidden layer; The output of RBM1 hidden layer is using the input as RBM2 visible layer, RBM2 hidden layer extracts the profound feature of grouped data further, the DBN model of having trained is packaged into a fixing function, the input of this function is received signal strength RSSI, exports the location point being target.