CN111523667B

CN111523667B - RFID positioning method based on neural network

Info

Publication number: CN111523667B
Application number: CN202010365734.5A
Authority: CN
Inventors: 吴浩; 石海涛
Original assignee: Tianjin University
Current assignee: Tianjin University
Priority date: 2020-04-30
Filing date: 2020-04-30
Publication date: 2023-06-27
Anticipated expiration: 2040-04-30
Also published as: CN111523667A

Abstract

The invention discloses an RFID positioning method based on a neural network, which comprises a card reader and a processor, wherein the processor realizes the following steps through program codes: the positioning algorithm module reads card reader coordinate data, reference tag coordinate data and real coordinate data of a point to be measured, and obtains measured coordinate and positioning error data of the point to be measured by adopting a LANDMARC algorithm, a BVER algorithm and a VIRE algorithm; the error analysis module extracts time stamps and error main characteristics in the measured coordinates and positioning error data of the point to be measured to construct a training set; the positioning correction module corrects the minimum error position point coordinate and the minimum error position point time to output a point coordinate correction value to be detected and a position coordinate corresponding to the maximum occurrence possibility.

Description

RFID positioning method based on neural network

Technical Field

The invention mainly relates to the technical field of indoor positioning based on an RFID network, in particular to an RFID positioning method based on a neural network.

Background

At present, under indoor environment, radio frequency identification (Radio Frequency Identification, RFID) technology is widely paid attention to and applied due to the characteristics of information carrying function, reliable identification of transmission and the like. The RFID positioning technology utilizes the unique identification characteristic of the tag to the object, and obtains the position information of the electronic tag according to the signal sent by the electronic tag and received by the card reader. The basis of RFID indoor positioning is to combine the received signal intensity, phase and other parameters of RFID signals, and utilize a positioning algorithm to complete the calculation of distance and azimuth, namely, a plurality of card readers are placed in the indoor in advance, when a mobile object with an electronic tag enters the recognition range of the card reader, the received signal can be uploaded to an upper computer, and the upper computer can realize the positioning algorithm by calculating the signal attenuation degree of the tag and the position information of the adjacent known tag.

Indoor positioning algorithms based on RFID can be divided into two main categories: non-ranging positioning algorithms and ranging positioning algorithms. The RSSI positioning method in the ranging positioning algorithm is easier to operate than other positioning methods, and the thought of the RSSI positioning method is to estimate the position of an object by measuring the RSSI of a signal strength value, and more advanced positioning algorithms comprise a LANDMARC algorithm, a BVERE algorithm and a VIRE algorithm.

The LANDMARC (Location Identification Based on Dynamic Active RFIDCalibration) algorithm is well-established for its simplicity and high positioning accuracy, and the boundary virtual tag algorithm (Boundary Virtual Label Algorithm, BVIRE) is obtained by inserting the grid virtual reference tag and the boundary reference tag with similar parties on the basis of the LANDMARC algorithm. Two weights are adopted in the boundary virtual tag algorithm, which is 1 more than that of the Landmarc algorithm, and a threshold value TH is adopted in the selection of adjacent tags to exclude the tags with large errors with small probability, so that the BVER algorithm is greatly improved in positioning accuracy. The VIRE algorithm introduces the concepts of virtual reference labels and adjacent maps while utilizing the principle of LANDMARC, provides the concepts of the virtual reference labels, utilizes an interpolation method to estimate the signal strength values of the virtual reference labels, and takes the virtual reference labels as actual reference labels to perform later calculation and positioning, thereby improving the accuracy and the positioning calculation feasibility.

Conventional approaches require that all test samples be stored in a database. This will seriously affect the positioning efficiency and accuracy, since in a complex indoor environment most of the test samples are noisy. And a large amount of data needs to be collected as an alignment for high accuracy.

In recent years, machine learning has been increasingly applied to rfid processing. However, since the acquired RSSI signal is usually noisy, the existing features are all selected based on the raw data, and a large amount of computation and tuning work are required to improve the model accuracy. Therefore, the organic combination of error correction and positioning calculation realizes the aims of reducing positioning error, improving positioning accuracy and increasing positioning calculation efficiency.

Disclosure of Invention

The invention aims to design an RFID positioning method based on a neural network, which is used for analyzing errors, reducing original noise and data volume and reducing algorithm errors, calculated amount and calculation complexity.

In view of the above problems, the present invention provides an RFID positioning method based on a neural network, which calculates a minimum error value of a card reader based on a random position to read position data of a moving object at a certain position. And carrying out positioning calculation according to the optimization schemes of the three positioning algorithms, carrying out data analysis on the result data of the positioning calculation to obtain error data of the three algorithms, taking the error data as a modeling object, and modeling by using a neural network algorithm to obtain the position coordinates and the corresponding time of the maximum occurrence possibility. The invention integrates the LANDMARC algorithm, the BVERE algorithm and the VIRE algorithm, and combines the neural network algorithm to obtain a more efficient and accurate result, thereby realizing the improvement of the RFID positioning precision.

The achievement of the invention is realized by the following steps:

the RFID positioning method based on the neural network comprises a card reader and a processor, wherein the processor realizes the following steps through program codes:

the positioning algorithm module reads card reader coordinate data, reference tag coordinate data and real coordinate data of a point to be measured, and obtains measured coordinate and positioning error data of the point to be measured by adopting a LANDMARC algorithm, a BVER algorithm and a VIRE algorithm;

the error analysis module extracts time stamps and error main characteristics in the measured coordinates and positioning error data of the point to be measured to construct a training set;

the training set processes the timestamp and the error main characteristic through grid encryption and a neural network algorithm to output the minimum error position point coordinate and the minimum error position point time;

and the positioning correction module corrects the minimum error position point coordinate and the minimum error position point time and outputs a coordinate correction value of the point to be detected and a position coordinate corresponding to the maximum occurrence possibility.

Further, the neural network algorithm is to selectively input the minimum error position point coordinates and the minimum error position point time obtained in the RNN, CNN and LSTM neural network algorithm after efficiency calculation and data stability evaluation are carried out on the positioning errors of the LANDMARC algorithm, the BVERE algorithm and the VIRE algorithm.

Further, the positioning correction module corrects the minimum error position point coordinate and the minimum error position point time to find the corresponding measured coordinate of the point to be measured through data screening of the minimum error value, and performs data analysis on the measured coordinate of the point to be measured and the real coordinate of the point to be measured through path fitting of the moving object to realize positioning correction of the measured coordinate of the point to be measured

Advantageous effects

1. The invention provides a neural network analysis algorithm based on an RFID indoor positioning algorithm, which improves the calculation accuracy and the positioning calculation complexity.

2. The invention provides a neural network analysis method based on a time stamp and an error value as a feature set, which utilizes a feedback neural network to convert a positioning problem into an error problem for analysis. Compared with the prior art, the method combines a plurality of algorithms, effectively saves data resources, does not directly analyze the original data, and improves the data operation analysis speed.

Drawings

FIG. 1 is a diagram of the result of a positioning algorithm.

FIG. 2 is a logic block diagram of an analysis method based on three RFID positioning algorithms and a neural network;

fig. 3 is a logic structure diagram based on a feedback neural network.

Detailed Description

The data analysis, network training and data computation of the inventive design are described in detail below with reference to the accompanying drawings. The whole positioning principle block diagram is shown in fig. 1, 2 and 3.

Step 1: and (3) carrying out data analysis by using three positioning algorithm modules, wherein input data are card reader coordinate data, reference tag coordinate data and real coordinate data of the point to be detected, output data are measured coordinates and positioning errors of the point to be detected, and a positioning result is shown in a schematic diagram shown in fig. 1.

The method comprises the steps of performing parallel calculation through three positioning algorithm modules of a LANDMARC algorithm, a BVERE algorithm and a VIRE algorithm, performing simulation by utilizing Matlab, inputting card reader coordinate data, reference tag coordinate data and real coordinate data of a point to be measured, performing positioning calculation by calculating the received signal strength of the reference tag and the tag to be measured, and simultaneously starting the calculation of the three algorithms according to a distributed principle, wherein output data is measured coordinates and positioning errors of the point to be measured.

Step 2: the error analysis module performs network training by utilizing a neural network algorithm, inputs positioning errors of three algorithms, inputs RNN, CNN and LSTM neural network algorithms, and outputs data which are error minimum values and corresponding time stamps.

That is, the positioning error data is preprocessed, the positioning errors of the three algorithms are input by using grid calculation and a neural network algorithm, the RNN, CNN and LSTM neural network algorithm are input at the same time, the neural network algorithm is selected according to the calculation efficiency and the data stability, the time stamp and the error value are used as main characteristic sets in the neural network algorithm to train, and the minimum value of the positioning error and the time stamp corresponding to the occurrence of the positioning error are output by training the selected neural network algorithm.

Step 3: and (3) carrying out data calculation by using a positioning correction algorithm, wherein input data is the minimum value of errors, and output data is the coordinate correction value of the point to be detected, the corresponding maximum possible position coordinate and the corresponding time.

Namely, a positioning correction algorithm is realized by selecting a minimum value of a positioning error, a corresponding measured coordinate of a point to be measured is found through data screening of the minimum value of the error, the measured coordinate of the point to be measured and a true coordinate of the point to be measured are subjected to data analysis through path fitting of a moving object, positioning correction of the measured coordinate of the point to be measured is realized, and a coordinate correction value of the point to be measured and a corresponding position coordinate with the maximum occurrence possibility are output.

Claims

1. The utility model provides a RFID positioning method based on neural network, includes card reader, treater, its characterized in that: the processor implements the following steps by program code:

the training set processes the timestamp and the error main characteristic through grid encryption and a neural network algorithm to output the minimum error position point coordinate and the minimum error position point time; the neural network algorithm is that after efficiency calculation and data stability evaluation are carried out on positioning errors of a LANDMARC algorithm, a BVER algorithm and a VIRE algorithm, the minimum error position point coordinates and the minimum error position point time are selectively input into the RNN, CNN and LSTM neural network algorithm;

the positioning correction module corrects the minimum error position point coordinate and the minimum error position point time and outputs a coordinate correction value of the point to be detected and a position coordinate corresponding to the maximum occurrence possibility; wherein: the positioning correction module corrects the minimum error position point coordinate and the minimum error position point time to find the corresponding measured coordinate of the point to be measured through data screening of the minimum error value, and performs data analysis on the measured coordinate of the point to be measured and the real coordinate of the point to be measured through path fitting of a moving object to realize positioning correction of the measured coordinate of the point to be measured.