CN109068267B

CN109068267B - Indoor positioning method based on LoRa SX1280

Info

Publication number: CN109068267B
Application number: CN201810876943.9A
Authority: CN
Inventors: 姚英彪; 陈宇翔; 姜显扬; 许晓荣; 刘兆霆; 冯维
Original assignee: Hangzhou Dianzi University
Current assignee: HANGZHOU CCRFID MICROELECTRONICS Co.,Ltd.
Priority date: 2018-08-03
Filing date: 2018-08-03
Publication date: 2020-06-23
Anticipated expiration: 2038-08-03
Also published as: CN109068267A

Abstract

The invention discloses an indoor positioning method based on LoRa SX 1280. The method comprises an off-line fingerprint acquisition modeling stage and an on-line real-time positioning stage. An off-line fingerprint acquisition modeling stage: firstly, special fingerprint features based on LoRa SX1280 are needed, wherein the special fingerprint features comprise ranging fingerprints and RSSI fingerprints; then, when the special fingerprint is input into the fingerprint library, the special fingerprint needs to be preprocessed by adopting a Gaussian filtering method, and finally, a positioning calculation model is obtained by utilizing an improved support vector regression method. And (3) an online real-time positioning stage: firstly, acquiring t times in an indoor area by utilizing a mobile node device and an anchor point device to form t groups of special fingerprint characteristics; secondly, a median filtering method is adopted for the t groups of special fingerprint features to obtain a unique fingerprint feature; and finally, substituting the unique fingerprint characteristic into the obtained calculation model, and calculating to obtain the position coordinate of the node to be positioned. The invention eliminates the problem that the fingerprint is easy to be interfered.

Description

Indoor positioning method based on LoRa SX1280

Technical Field

The invention relates to the technical field of Internet of things, in particular to an indoor positioning method based on LoRa SX 1280.

Background

The LoRa technology is an ultra-long distance wireless transmission scheme based on a spread spectrum technology adopted and popularized by Semtech corporation, and belongs to one of Low Power Wide Area Network (LPWAN) communication technologies. The LoRa technology is not limited by the compromise between transmission distance and power consumption, and a system which can realize long transmission distance, low power consumption and multiple nodes is provided for users, so that the system is developed into a network.

As the audience population for the LoRa technology increases, and as such, continues to evolve. Semtech subsequently introduced SX1280 chips, meaning the appearance of LoRa in the 2.4GHz band. The chip has a built-in ranging engine function, i.e. a Time Of Arrival (TOA) ranging engine based on Time-Of-flight (tof). The ranging function of SX1280 is based on the measurement of the round-trip time-of-flight of wireless signals between SX1280 transceivers. This process uses the LoRa modulation scheme and therefore benefits from all the advantages of long range and low power operation granted by LoRa.

With the continuous expansion and deepening of the application of the internet of things, the location-based service has wider and wider application. Indoor positioning has some specificity due to special environmental factors, such as lack of unified infrastructure, difficulty in obtaining consistency and universality of indoor positioning technology, and serious interference effect due to indoor environmental changes. Currently, indoor positioning methods mainly include a positioning method based on ranging and a positioning method based on fingerprints. The ranging-based location method has different methods such as TOA, time difference of arrival (TDOA), Received Signal Strength (RSSI), etc., according to the transmission model. In the methods, the RSSI-based ranging is widely applied due to low hardware cost and low calculation cost, but the positioning error is large due to poor RSSI ranging accuracy. Although SX1280 employs TOA ranging, practical tests show that the ranging error of SX1280 is also relatively large indoors. Therefore, the positioning error obtained by performing positioning calculation only according to the TOA ranging value of SX1280 is also relatively large. The fingerprint-based positioning method mainly utilizes the difference of the space positions of wireless signals, and physical position information corresponding to the position fingerprint of a target node is searched through a position fingerprint relation database and a matching algorithm so as to estimate the position of the node. Common matching algorithms are Nearest Neighbor (NN) algorithm, neural network algorithm, vector machine algorithm, K-nearest neighbor (KNN) algorithm, etc. The existing fingerprint positioning method mainly adopts RSSI fingerprint or magnetic field fingerprint and the like. In an actual indoor environment, due to the influences of multipath, shadow effect, personnel walking and the like, the RSSI values from the AP points at fixed positions often show complex time-varying statistical characteristics, which may reduce the positioning accuracy of the system and increase the complexity of the system.

Disclosure of Invention

The invention discloses an indoor system and a positioning method based on LoRa SX 1280.

The indoor positioning system used for solving the technical problem comprises: a mobile node device, an anchor point device, a gateway and a positioning server; wherein the mobile node device and the anchor point device are integrated with a LoRa SX1280 module.

The LoRa SX1280 module supports a LoRa communication mode and a LoRa ranging engine mode.

The mobile node apparatus can be configured on any mobile equipment, and the position of the mobile node apparatus is not fixed. The mobile node device adopts an SX1280 module, an independent antenna, a vibration sensor and a battery for power supply.

The anchor point device is a device which is placed at a position with known coordinates and is not changed, and is mainly used for assisting the mobile node device to complete the collection of fingerprint information. The anchor point device adopts an SX1280 module, an independent antenna and mains supply for power supply.

The gateway is mainly used for communication between the mobile node device and the positioning server.

The positioning server is used for performing positioning calculation after receiving a positioning request and relevant information of the mobile node device.

The technical scheme adopted by the invention for solving the technical problem comprises the following steps: an off-line fingerprint acquisition modeling stage and an on-line real-time positioning stage.

Firstly, an off-line fingerprint acquisition modeling stage comprises the following specific steps:

1-1, carrying out grid division on an area to be positioned, and acquiring special fingerprint characteristics based on LoRa SX1280 for each grid point area by utilizing a mobile node device and an anchor point device;

the special fingerprint features comprise ranging fingerprints obtained based on a LoRa SX1280 ranging engine mode and Received Signal Strength (RSSI) fingerprints obtained based on a LoRa SX1280 communication mode; the distance measurement fingerprint and the RSSI fingerprint are combined, and the dimensionality of the characteristic fingerprint is increased.

Further, let the center position coordinate of the ith grid point region be P_i＝(x_i,y_i) Noting the ith grid point regionThe collected characteristic fingerprint matrix is

For the

When l is 2j-1 (odd column) represents a received signal strength RSSI fingerprint value obtained by the jth anchor point device based on the communication mode at the kth acquired mesh point region, and when l is 2j (even column) represents a ranging fingerprint value obtained by the jth anchor point device based on the ranging engine mode at the kth acquired mesh point region, where i is 1,2 … N, j is 1,2 … M, l is 1,2 … 2M, and k is 1,2 … T. R_iThe number of columns of (2M) is the dimension of the feature fingerprint vector. In the above, N is the number of grid points in the region to be located, M is the number of anchor points, and T is the number of acquisition times at each grid point.

And 1-2, replacing the extreme data with a mean value by adopting a Gaussian filtering method aiming at the special fingerprint characteristics acquired by each grid point, and then storing the filtered special fingerprint characteristics and the two-dimensional coordinate position corresponding to the grid point into a fingerprint database.

And 1-3, taking the fingerprints in the fingerprint database as a training set, and then constructing a calculation model between the special fingerprint characteristics and the two-dimensional coordinates corresponding to the grid points by using an improved support vector regression algorithm.

Secondly, in an online real-time positioning stage, the specific steps include:

and 2-1, acquiring t times of special fingerprints at the position of the mobile node device (node to be positioned) to form t groups of special fingerprint characteristics.

2-2, obtaining a unique fingerprint feature by adopting a median filtering method for the t groups of special fingerprint features.

And 2-3, substituting the unique fingerprint characteristic into the calculation model obtained in the step 1-3, and calculating to obtain the position coordinate of the node to be positioned.

Further, the method for constructing the improved support vector regression calculation model in the step 1-3 and the method for calculating the position coordinates of the node to be positioned in the step 2-3 both adopt the improved support vector regression ISVR algorithm proposed in the literature [1 ].

The invention has the following beneficial effects:

the invention provides an indoor positioning system and method based on a LoRa SX1280 chip, which are essentially fingerprint-based indoor positioning system and method and can solve the problem of designing a high-precision indoor positioning system by using the LoRa SX1280 chip. The difference from the traditional fingerprint positioning method is that: the fingerprint adopts the fused fingerprint of the ranging value and the RSSI value, so that the problem that the fingerprint is easy to interfere is solved; the fingerprints are preprocessed in an off-line stage and an on-line stage, so that the consistency of the acquired fingerprints and the positions is improved; the ISVR provided by the inventor in the earlier stage is used for positioning calculation, so that the accuracy of a calculation model of fingerprint features and positions is improved. Due to the improvement of the three aspects, the positioning accuracy of the indoor positioning system and the indoor positioning method disclosed by the invention is higher than that of the traditional fingerprint positioning system and the fingerprint positioning method.

[1] The indoor positioning algorithm [ J ] based on improved support vector regression, the academic newspaper of instruments and meters, 2017,38(9): 84-91.

Drawings

FIG. 1(a) is a diagram of an indoor positioning system according to the present invention;

FIG. 1(b) is a diagram of an indoor positioning system according to the present invention;

FIG. 2 illustrates a particular fingerprint feature of the present invention;

fig. 3 is a schematic diagram of obtaining an RSSI fingerprint by using a LoRa communication mode according to the present invention;

FIG. 4 is a schematic diagram of a LoRa ranging engine mode for obtaining a ranging fingerprint according to the present invention;

FIG. 5 is a diagram illustrating the steps of a support vector regression positioning method of the present invention;

fig. 6 is a schematic diagram of an indoor positioning environment in an embodiment of the present invention.

Detailed Description

The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

Fig. 1(a) and (b) are device diagrams of an indoor positioning system based on LoRa SX1280 according to the present invention. Including both mobile node devices fig. 1(a) and fixed anchor point devices fig. 1 (b). And an LoRa module containing an SX1280 chip is required to be installed on the movable node device and the fixed anchor point device. The SX1280 chip of alt corporation of america possesses a LoRa communication mode and a LoRa ranging engine mode.

Fig. 2 shows a specific fingerprint feature based on LoRa SX1280 according to the present invention. Such fingerprint features include ranging fingerprints and RSSI fingerprints; as shown in fig. 2, the RSSI fingerprint value is combined with the distance measurement fingerprint value obtained based on the LoRa distance measurement engine mode, so as to increase the dimension of the eigenvector and improve the positioning accuracy.

Fig. 3 is a schematic diagram of obtaining an RSSI fingerprint by using the LoRa communication mode according to the present invention. The acquisition device of fig. 1 needs to be utilized, and a LoRa module containing an SX1280 chip is mounted on a movable node device and a fixed anchor device. The SX1280 chip itself possesses the LoRa communication mode and the LoRa ranging engine mode. In the communication mode, a mobile node device is used for broadcasting and paging anchor point devices (AP access points) fixed around, and the RSSI value of each anchor point is read after an anchor point reply is received.

Fig. 4 is a schematic diagram of obtaining a ranging fingerprint by using the LoRa ranging engine mode according to the present invention. After the RSSI fingerprint is acquired in the last step, the movable node device enters a ranging engine mode host mode, each anchor point device enters a ranging engine mode slave mode, the movable node device and each anchor point device perform ranging one by one, and ranging values between the movable node device and each anchor point are read out.

FIG. 5 is a step diagram of a location method based on LoRa SX1280 support vector regression according to the present invention;

the method comprises the following steps: s1 off-line acquisition stage; and carrying out grid division on the area to be positioned, and reasonably deploying the fixed anchor point device. And the special fingerprint characteristics containing the ranging value and the RSSI value are collected by utilizing the movable node device and the anchor point device. And then, carrying out data filtering on the initial fingerprint characteristics by using a Gaussian filtering method, namely, replacing the extreme data with the mean value by using the Gaussian filtering method for the fingerprint characteristics acquired at each reference point position. And then storing the characteristic sample and the corresponding coordinate position which are fused by the RSSI fingerprint and the ranging fingerprint after the Gaussian filtering method into a fingerprint database. And an off-line training model between the position coordinates and the reference point fingerprint features is constructed by utilizing an improved support vector regression algorithm. S2, an online real-time positioning stage; the method comprises the steps of utilizing a movable node device to collect wireless signal characteristics for t times in real time at a position to be positioned, and adopting a median filtering data preprocessing method, namely adopting a method of taking median in each dimension for the t fingerprint characteristics at the same position, and only selecting to obtain a unique fingerprint characteristic. And substituting the fingerprint characteristics into the offline training model obtained in the step S1, so as to calculate and obtain the position coordinates of the node to be positioned.

The specific implementation steps of the support vector regression positioning method based on LoRa SX1280 are as follows:

step 1:

and carrying out grid division according to the area in the area to be positioned, establishing a rectangular coordinate system, selecting N reference point areas and M LoRa access points, fixing each LoRa access point, and installing an anchor point device.

In the embodiment shown in fig. 6, an indoor environment of 16M by 12M is formed by meshing the areas with dotted lines, each mesh is a fingerprint reference point area, a black color × indicates the center of the reference point area, the coordinates of the point are the coordinates of the reference point, and a black triangle is a selected LoRa access point, and a fixed anchor device needs to be installed, where N is 48 and M is 4.

Step 2:

and acquiring the fingerprint vector signal characteristics with the movable node device for T times as 1 group of data in an evenly distributed mode at each reference point area position. The fingerprint vector signal characteristics comprise RSSI value fingerprints collected by the movable nodes and between surrounding fixed anchor point devices and ranging fingerprints based on an LoRa ranging engine mode. Specifically, we use the LoRa ranging engine mode, so we use the SX1280 chip from the alt corporation, which has built-in ranging engine function, i.e., Time of arrival (TOA) fusion ranging engine, and uses the Time-of-flight (tof) ranging method. The ranging function of SX1280 is based on a measurement of the round-trip time-of-flight between a pair of SX1280 transceivers (i.e., here the node to be located and the fixed anchor point). It is easy to understand that, training the feature vector combining the RSSI value and the ranging value in the positioning algorithm based on support vector regression can increase the dimensionality of the feature vector and improve the positioning accuracy.

Let the coordinate of the center position of the ith grid point region be P_i＝(x_i,y_i) Recording the characteristic fingerprint matrix acquired from the ith grid point area as

For the

When l is 2j-1 (odd column) represents a received signal strength RSSI fingerprint value obtained by the jth anchor point device based on the communication mode at the kth acquired mesh point region, and when l is 2j (even column) represents a ranging fingerprint value obtained by the jth anchor point device based on the ranging engine mode at the kth acquired mesh point region, where i is 1,2 … N, j is 1,2 … M, l is 1,2 … 2M, and k is 1,2 … T. R_iThe number of columns of (2M) is the dimension of the feature fingerprint vector. In the above, N is the number of grid points in the region to be located, M is the number of anchor points, and T is the number of acquisition times at each grid point. In practical applications, T is generally greater than 30 to achieve better positioning effect.

And step 3:

and performing initial signal preprocessing on the fingerprint vector signal characteristics uniformly distributed and collected in each reference point area. The initial signal preprocessing adopts a Gaussian filtering method, and specifically adopts the following principle: fingerprint matrix R acquired for each reference point of formula (1)_iCalculating the average value μ of each dimension data_l(mean of each column), where l ═ 1,2 …,2M, represents the dimension. Then setting a threshold probability lambdaIt means that the signal values are distributed in the interval [ mu ]_l-θ_l,μ_l+θ_l]Wherein θ is_lAnd the data can be obtained through the acquired training set data and lambda. Specifically, assuming that each line of data in the acquired training set data obeys normal distribution, the probability density function is recorded as f_l(x) Then theta_lSatisfy the requirement of

Determining theta from equation (2)_lThen R is obtained_iIn each column of elements is in the reasonable range of [ mu ]_l-θ_l,μ_l+θ_l]. Signal values within this range are considered reliable; conversely, if the signal value is not within this range, the signal value is deemed unreliable. Finally, for R_iEach element of

If it is not

Retention

If not, then,

λ may be selected according to the actual environment, and may be 0.8, for example.

And 4, step 4:

using the fingerprint matrix R of each reference point after the preprocessing of step 3_iEach line and position coordinate P of_i＝(X_i，Y_i) Establishing a fingerprint database D, R_i∈R^2MAnd as a training sample set of the SVR algorithm, the total number of samples should be N × T.

And 5:

and constructing a nonlinear relation between the fingerprint vector and the coordinates of the reference point by utilizing an SVR algorithm. First by a non-linear mapping Φ:R^2Minput space R → F^2MMapping into a high-dimensional feature space F, and then constructing an optimal regression function of the position coordinates P and the fingerprint R in the F:

P＝W^T·Φ(R)+b。 (3)

wherein W is a weight coefficient, W ∈ F, b is an offset coefficient

To finally determine the parameters W and b in equation (5), the following convex quadratic programming problem can be solved according to the principle of minimization of structural risk.

Satisfies the following conditions:

wherein C is a penalty constant of ξ_i,

Is the relaxation variable.

The lagrange polynomial of the convex quadratic programming problem of equation (5) is:

in the formula:

lagrange multiplier, and need to satisfy

And according to the condition of the optimal solution, making L to W, b,

calculating the partial derivative as 0 can obtain:

by solving the dual optimization problem, the following can be obtained:

satisfies the following conditions:

0≤α_i ^*,α_i≤C

wherein κ (R)_i,R_j) For kernel functions, the kernel functions are typically Gaussian kernel functions

Wherein x_i,x_jRepresenting the input vector, σ is the gaussian kernel width.

Thereby, can obtain

The optimal penalty constant C and the kernel function width value σ can be obtained by performing a grid search using a support vector machine tool box LIBSVM.

Positioning based on the SVR algorithm, where only indoor plane positioning is considered, requires outputting two-dimensional coordinates (x, y) of the user, i.e., a prediction that multiple outputs are used. The outputs of the traditional SVR algorithm are all one-dimensional, and at the moment, the multiple outputs can be replaced by a plurality of single outputs to realize the multiple-output SVR algorithm. However, the indoor coordinate system is a two-bit related information, so that the two-dimensional coordinate vector obtained by training each coordinate individually can reduce the accuracy of the training model to some extent.

In order to reduce errors of independently constructing an x coordinate model and a y coordinate model and improve the correlation between two-dimensional position information and signal strength, a correction coordinate z which contains information of x and y can be additionally trained in a training stage.

Step 6:

and acquiring the wireless signal characteristics of the node to be positioned in real time by using the movable node device, and preprocessing data. The data preprocessing adopts a median filtering method, and specifically, t fingerprint characteristic vectors are continuously acquired by a movable node at a certain position to be positioned to obtain t records. The t records are recorded as

Wherein for

When l is 2j-1 (odd column), the RSSI fingerprint value from the jth anchor point acquired t time is represented, and when l is 2j (even column), the distance measurement fingerprint value obtained by the jth anchor point acquired t time based on the distance measurement engine mode is represented. Wherein l 1,2.. 2M, j 1,2.. M. Let the median of each column be R_lThen the median vector for Z should be [ R "")₁，R″₂，...R″_l…R″_2M-1，R″_2M]。

Specifically, let t records be as follows, t being 6:

the median vector for Z is [ -55, 8.35, -61, 5.25, -65, 4.58, -71, 10.5 ].

And 7:

and substituting the Z median vector after data preprocessing into the off-line training model constructed in the step 5.

Since in the training phase, 3 training models are constructed for x, y, and z, respectively. Therefore, the prediction result obtained has 3 sets of coordinate values, which are (x ', y'), (x ', z'/x '), (z'/y ', y'), respectively. After 3 groups of coordinate values are obtained by correcting the coordinates, if the obtained coordinate values are the same, the result can be directly output; on the contrary, if the obtained coordinate values are different, one set of coordinate output needs to be selected from the 3 sets of coordinates.

For this problem, a Weighted Inverse K Nearest Neighbor (WIKNN) method is adopted for selection: according to the obtained 3 groups of predicted position coordinate combinations, selecting k position coordinates with the nearest Euclidean distance by calculating the Euclidean distance between the coordinates of each position in the fingerprint database and the predicted position coordinates, and then multiplying fingerprint features corresponding to the selected k coordinates by a weighting coefficient and then summing to obtain the fingerprint feature vectors corresponding to the predicted positions.

In particular, the amount of the solvent to be used,

step 7.1, assuming the selected coordinates are (X ', y'), first calculate (X ', y') and coordinates in the fingerprint database (X) using equation (14)_i，Y_i) A distance d between_iThen according to d_iSelecting the first k reference points from small to large, and calculating the mean vector R of the fingerprint feature vectors of the reference points in the fingerprint database_iFinally, the k feature vectors are weighted by using the formula (15), so as to obtain the fingerprint feature vector R corresponding to the coordinates (x ', y').

In the formula:

and 7.2, calculating the Euclidean distance D (1) between the fingerprint feature vectors R and R 'by using the formula (16) according to the fingerprint feature vector R' acquired at the position (x ', y').

And 7.3, repeating the steps 7.1 and 7.2 for (x ', z'/x '), (z'/y ', y') respectively to obtain Euclidean distances D (2) and D (3) respectively.

And 7.4, sequencing the obtained Euclidean distances D (1), D (2) and D (3), selecting a coordinate corresponding to the D (i) with the minimum Euclidean distance, and taking the coordinate as a final position coordinate. Thereby estimating the actual location of the user.

Finally, the above-mentioned embodiments are not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An indoor positioning method based on LoRa SX1280 is characterized by comprising an offline fingerprint acquisition modeling stage and an online real-time positioning stage, and is specifically realized as follows:

the special fingerprint features comprise ranging fingerprints obtained based on a LoRa SX1280 ranging engine mode and Received Signal Strength (RSSI) fingerprints obtained based on a LoRaSX1280 communication mode; combining the ranging fingerprint with the RSSI fingerprint to increase the dimensionality of the characteristic fingerprint;

further, let the center position coordinate of the ith grid point region be P_i＝(x_i,y_i) Recording the characteristic fingerprint matrix acquired from the ith grid point area as

For the

When l is 2j-1 (odd column) represents the k-th acquired data at the ith grid point regionA received signal strength RSSI fingerprint value obtained by the jth anchor point device based on the communication mode when l is 2j (even column) represents a ranging fingerprint value obtained by the jth anchor point device based on the ranging engine mode and acquired at the kth time at the ith grid point region, wherein i is 1,2.. N, j is 1,2.. M, l is 1,2.. 2M, k is 1,2.. T; r_iThe number of columns of (2) is the dimension 2M of the feature fingerprint vector; in the above, N is the number of grid points of the area to be positioned, M is the number of anchor points, and T is the number of acquisition times at each grid point;

1-2, aiming at the special fingerprint characteristics acquired at each grid point, replacing the data of the extreme points by a mean value by adopting a Gaussian filtering method, and then storing the filtered special fingerprint characteristics and the two-dimensional coordinate position corresponding to the grid point into a fingerprint database;

1-3, taking the fingerprints in the fingerprint database as a training set, and then constructing a calculation model between the special fingerprint characteristics and the two-dimensional coordinates corresponding to the grid points by using an improved support vector regression algorithm;

secondly, in an online real-time positioning stage, the specific steps include:

2-1, acquiring t times of special fingerprints at the position of the mobile node device to form t groups of special fingerprint characteristics;

2-2, obtaining a unique fingerprint characteristic by adopting a median filtering method for the t groups of special fingerprint characteristics;

2-3, substituting the unique fingerprint characteristic into the calculation model obtained in the step 1-3, and calculating to obtain the position coordinate of the node to be positioned;

the Gaussian filtering method is adopted to replace the extreme data with the mean value, and the method is specifically realized as follows:

fingerprint matrix R acquired for each reference point of formula (1)_iCalculating the average value μ of each dimension data_lWherein l 1, 2M, denotes dimension; setting a threshold probability lambda, wherein the threshold probability lambda represents the distribution of the signal value in the interval [ mu ]_l-θ_l，μ_l+θ_l]Wherein θ is_lObtaining the data through the acquired training set data and lambda;

let each column of the collected training set data be compliant withState distribution, with probability density function of f_l(x) Then theta_lSatisfy the requirement of

Determining theta from equation (2)_lThen R is obtained_iIn each column of elements is in the reasonable range of [ mu ]_l-θ_l，μ_l+θ_l](ii) a Signal values within this range are considered reliable; conversely, if the signal value is not within this range, the signal value is deemed unreliable;

for R_iEach element of

If it is not

Retention

If not, then,

lambda is selected according to the actual environment.

2. The indoor positioning method based on LoRa SX1280 as claimed in claim 1, wherein the method for constructing the calculation model in step 1-3 and the method for calculating the position coordinates of the node to be positioned in step 2-3 both adopt an improved support vector regression ISVR algorithm.

3. The indoor positioning method based on LoRa SX1280 as claimed in claim 2, wherein the equipment required for implementing the method comprises a mobile node device, an anchor point device, a gateway, a positioning server; wherein mobile node device and anchor point device all integrate loRa SX1280 module, just loRa SX1280 module support loRa communication mode and loRa ranging engine mode.

4. The LoRa SX 1280-based indoor positioning method according to claim 3, wherein the mobile node device is configured on all mobile devices; the mobile node device adopts an SX1280 module, an independent antenna, a vibration sensor and a battery for power supply.

5. The LoRa SX 1280-based indoor positioning method according to claim 4, wherein the anchor point device is a device fixedly installed at a designated location, and is used for communication with the mobile node device and assisting in completing the collection of information related to positioning of the mobile node device; the anchor point device adopts an SX1280 module, an independent antenna and mains supply for power supply.

6. The LoRa SX 1280-based indoor positioning method according to claim 5, wherein the gateway is used for communication between the mobile node device and the positioning server; the positioning server is used for performing positioning calculation after receiving a positioning request of the mobile node device.

7. The indoor positioning method of claim 6, wherein the Improved Support Vector Regression (ISVR) algorithm is implemented as follows:

constructing a non-linear relationship between the fingerprint vector and the coordinates of the reference point:

first by a non-linear mapping Φ: r^2MInput space R → F^2MMapping into a high-dimensional feature space F, and then constructing an optimal regression function of the position coordinates P and the fingerprint R in the F:

P＝W^T·Φ(R)+b； (3)

wherein W is a weight coefficient, W ∈ F, b is an offset coefficient

In order to finally determine the parameters W and b in the formula (5), solving the following convex quadratic programming problem according to the minimum structural risk principle;

satisfies the following conditions:

wherein C is a penalty constant of ξ_i，ξ_i ^*Is a relaxation variable;

in the formula η_i，

α_i，

Lagrange multiplier, and need to satisfy

Let L pair W, b, ξ according to the conditions under which the optimal solution exists_i ^*Calculating the partial derivative as 0 can obtain:

by solving the dual optimization problem, the following can be obtained:

satisfies the following conditions:

wherein, κ (R)_i，R_j) Selecting a Gaussian kernel function for the kernel function

κ(x_i，x_j)＝exp(-||x_i-x_j||²/2σ²) (10)

Wherein x_i，x_jRepresenting the input vector, sigma is the width of a Gaussian kernel;

thereby obtaining

The optimal penalty constant C and the kernel function width value sigma are obtained by carrying out grid search by using a support vector machine tool box LIBSVM;

in order to reduce errors of independently constructing an x coordinate model and a y coordinate model and improve the correlation between two-dimensional position information and signal strength, a correction coordinate z which contains information of x and y is additionally trained in a training stage.

8. The indoor positioning method of claim 7, wherein the obtaining of the unique fingerprint feature by the median filtering method is specifically realized as follows:

continuously acquiring t fingerprint characteristic vectors by using a movable node at a certain position to be positioned to obtain t records Z:

wherein for

When l is 2j-1 (odd column), the signal strength indicator (RSSI) fingerprint value from the jth anchor point device acquired t time is represented, and when l is 2j (even column), the distance measurement fingerprint value obtained by the jth anchor point device from the jth anchor point device acquired t time based on the distance measurement engine mode is represented; wherein l 1,2.. 2M, j 1,2.. M; in each columnThe number of bits is R ″)_lThen the median vector for Z should be [ R "")₁，R″₂，...R″_l...R″_2M-1，R″_2M]。

9. The indoor positioning method based on LoRa SX1280 as claimed in claim 8, wherein the steps 2-3 are implemented as follows:

in the training stage, 3 training models are respectively constructed for x, y and z; therefore, the obtained prediction result has 3 groups of coordinate values which are (x ', y'), (x ', z'/x '), (z'/y ', y') respectively; obtaining 3 groups of coordinate values by correcting the coordinates, and if the obtained coordinate values are the same, directly outputting the result; on the contrary, if the obtained coordinate values are different, a group of coordinates needs to be selected and output by adopting a Weighted Inverse K Nearest Neighbor (WIKNN) method in the 3 groups of coordinates;

the method comprises the following specific steps:

① assuming the selected coordinates are (X ', y'), first (X ', y') and coordinates in the fingerprint database (X) are calculated using equation (14)_i，Y_i) A distance d between_iThen according to d_iSelecting the first k reference points from small to large, and calculating the mean vector R of the fingerprint feature vectors of the reference points in the fingerprint database_iFinally, weighting the k eigenvectors by using a formula (15) to obtain fingerprint eigenvectors R corresponding to coordinates (x ', y');

in the formula:

② calculating the Euclidean distance D (1) between the fingerprint feature vectors R and R 'by using the formula (16) according to the fingerprint feature vector R' acquired at (x ', y');

③ repeating steps ① and ② above for (x ', z'/x '), (z'/y ', y'), respectively, resulting in Euclidean distances D (2) and D (3), respectively;

④ the Euclidean distances D (1), D (2) and D (3) obtained above are sorted, and the coordinate corresponding to D (i) with the smallest Euclidean distance is selected and used as the final position coordinate.