CN113945215A - RFID indoor positioning method based on stacking model - Google Patents

RFID indoor positioning method based on stacking model Download PDF

Info

Publication number
CN113945215A
CN113945215A CN202111181914.9A CN202111181914A CN113945215A CN 113945215 A CN113945215 A CN 113945215A CN 202111181914 A CN202111181914 A CN 202111181914A CN 113945215 A CN113945215 A CN 113945215A
Authority
CN
China
Prior art keywords
training
rssi
model
value
data set
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111181914.9A
Other languages
Chinese (zh)
Inventor
鲁建厦
张相华
包秦
谭健
徐峰聪
龚辉
赵浩竣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202111181914.9A priority Critical patent/CN113945215A/en
Publication of CN113945215A publication Critical patent/CN113945215A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20—Instruments for performing navigational calculations
    • G01C21/206—Instruments for performing navigational calculations specially adapted for indoor navigation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S11/00—Systems for determining distance or velocity not using reflection or reradiation
    • G01S11/02—Systems for determining distance or velocity not using reflection or reradiation using radio waves
    • G01S11/06—Systems for determining distance or velocity not using reflection or reradiation using radio waves using intensity measurements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K7/00—Methods or arrangements for sensing record carriers, e.g. for reading patterns
    • G06K7/10—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation
    • G06K7/10009—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves
    • G06K7/10019—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves resolving collision on the communication channels between simultaneously or concurrently interrogated record carriers.
    • G06K7/10079—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves resolving collision on the communication channels between simultaneously or concurrently interrogated record carriers. the collision being resolved in the spatial domain, e.g. temporary shields for blindfolding the interrogator in specific directions
    • G06K7/10089—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves resolving collision on the communication channels between simultaneously or concurrently interrogated record carriers. the collision being resolved in the spatial domain, e.g. temporary shields for blindfolding the interrogator in specific directions the interrogation device using at least one directional antenna or directional interrogation field to resolve the collision
    • G06K7/10099—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves resolving collision on the communication channels between simultaneously or concurrently interrogated record carriers. the collision being resolved in the spatial domain, e.g. temporary shields for blindfolding the interrogator in specific directions the interrogation device using at least one directional antenna or directional interrogation field to resolve the collision the directional field being used for pinpointing the location of the record carrier, e.g. for finding or locating an RFID tag amongst a plurality of RFID tags, each RFID tag being associated with an object, e.g. for physically locating the RFID tagged object in a warehouse
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Toxicology (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Electromagnetism (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Automation & Control Theory (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses an RFID (radio frequency identification) positioning method based on a stacking model, which realizes the positioning of a target label by establishing the stacking model. Firstly, arranging equipment and labels in a positioning area; reading the RSSI value of each label through a reader, and smoothing by using a filtering algorithm to obtain a more stable data set, so that the trained model is more accurate and the prediction is more accurate; then establishing a stacking model, and training by using the smoothed data, wherein the stacking model is a layered model, a first layer uses a base learner, and a second layer uses a logistic regression layer, so as to prevent the overfitting of the whole model and obtain a more accurate positioning effect than a single model; and finally, positioning the target label through the trained model.

Description

RFID indoor positioning method based on stacking model
Technical Field
The invention relates to the technical field of indoor positioning, in particular to an RFID indoor positioning method based on a stacking model.
Background
The RFID technology identifies a target tag and acquires tag data through a radio frequency signal, is one of the main forces for the development of the Internet of things, has the characteristics of small volume, low cost, non-line-of-sight identification and the like, and is widely applied to the fields of warehouse storage, industrial production, book management and the like. With the further development of the internet of things, whether the object position can be accurately acquired becomes one of the key factors that the technology of the internet of things can break through.
The application of the RFID technology to indoor object positioning has been studied more, and the conventional RFID indoor positioning deduces the position of the tag by a distance measurement method, but has many problems in accuracy and stability. With the rapid development of machine learning theory and the application in various fields in recent years, the application of machine learning to RFID indoor positioning has obvious feasibility.
Disclosure of Invention
In order to solve the defects of the prior art and realize the purposes of improving the positioning precision and the stability by a stacking method in an integrated learning algorithm, the invention adopts the following technical scheme:
an RFID indoor positioning method based on a stacking model comprises the following steps:
s1, deploying k readers, m reference tags and tags to be positioned in a positioning range;
s2, the RSSI values returned by the labels are received through the reader, the RSSI values read by all the readers of each reference label are combined with the coordinates of the reference label to serve as a training sample, the same reference label is read for multiple times to obtain a plurality of training samples, and the training samples of all the reference labels are integrated into a training data set; for the tags to be positioned, combining the RSSI value read by all readers of each tag to be positioned with the coordinates of the tag to be positioned to serve as a sample to be positioned, and integrating the sample to be positioned into a data set to be positioned in the same way;
s3, constructing a stacking model, training through a training data set, wherein the stacking model is a layered model, a first layer uses a base learner, and a second layer uses a logistic regression layer to prevent overfitting of the whole model, and the method comprises the following steps:
s31, before learning, dividing the training data set into training sets DtrainAnd test set Dtest；
S32, in the first layer of training, adopting K-fold cross validation, and inputting the training set D for each base learnertrainDividing the test set into K parts again, taking 1 part as a test set, taking the rest K-1 parts as a training set, predicting the test set after training by a base learner to obtain a predicted value, sequentially taking 1 part as a test set, and integrating the obtained K predicted values according to the positions of the original test set to obtain a predicted data set corresponding to the base learner; merging prediction data sets generated by all the base learners on the same layer to obtain a secondary training set P;
s33, in the second layer of training, the secondary training set P is used as the characteristic input of the training set, the corresponding RFID reference label coordinate is used as the label input of the training set, and after the training is finished, the test set D is passedtestTesting, detecting the effect of the complete stacking model, and comparing the effect with the effect of a single base learner to judge whether the model parameters are improved;
and S4, inputting the RSSI value in the data set to be positioned into a trained stacking model to obtain the presumed coordinates of the label to be positioned.
Further, performing smoothing processing on the RSSI value of the training data set obtained in the step S2 through kalman filtering to obtain a data set containing RSSI data with smaller fluctuation as input of the training starting model of the step S3; and smoothing the RSSI value of the data set to be positioned obtained in the S2 through Kalman filtering, and then taking the RSSI value as the input of the training model in the S4, wherein the more accurate model can be obtained through the smoothing of the Kalman filtering, so that the estimation of the position of the RFID tag is more accurate.
Further, the kalman filtering algorithm is divided into three phases: an initialization phase, a prediction phase and an updating phase;
the initialization phase: setting an initial value of the state of the filter;
the prediction stage is as follows: estimating the state of the training sample at the current moment according to the state quantity and the control quantity of the training sample at the previous moment, and calculating a corresponding covariance matrix, wherein the process is represented by the following formula:
Figure BDA0003297621700000021
Pt -＝FPt-1FT+Q
wherein the content of the first and second substances,
Figure BDA0003297621700000022
is a prior state estimation value of RSSI at the time t;
Figure BDA0003297621700000023
is a posterior state estimation value of RSSI at the time of t-1; u. oft-1Is a control variable; f is a state transition matrix; b is a control matrix; pt -Estimating covariance a priori for time t; pt-1Estimating covariance for the posteriori at time t-1; q is the process noise covariance;
the updating stage comprises the following steps: the estimated values obtained in the prediction phase are corrected by means of the measured values received and the parameters used in the filter are updated, the specific steps being: firstly, calculating Kalman gain, then updating the prior state estimation value to an optimal value, namely a posterior state estimation value according to the Kalman gain, and finally calculating a covariance matrix corresponding to the optimal value, wherein the process is represented by the following formula:
Figure BDA0003297621700000024
Figure BDA0003297621700000025
Pt＝(I-KtH)Pt -
wherein, KtIs the Kalman gain at time t; h is an observation matrix; r is a measurement error; z is a radical oftIs a measured value;
Figure BDA0003297621700000026
as a residual between the actual and expected observations, PtEstimating covariance for the posteriori at time t; and I is an identity matrix.
The RFID indoor positioning method based on the tracking model as claimed in claim 1, wherein the tag to be positioned for verification is used as a verification sample in S2, the verification data set is obtained through S1, the RSSI value in the verification data set is input into the trained tracking model in S4, the coordinates of the tag to be positioned are estimated and compared with the corresponding actual position in the verification data set, the tracking model is verified, and then the tag to be positioned which is actually to be positioned is estimated.
Further, in S1, the reference tags are arranged at equal intervals in a rectangle to collect the position information thereof, and the arrangement of the reader is such that all the tags can be read.
Further, the base learner in S3 is a strong learner randomfort, the decision trees are integrated together by a bagging method, random attribute selection including sample randomization and feature randomization is introduced, for each decision tree, training samples are randomly and replaceably extracted from a training set, and partial features are randomly extracted from all features, thereby obtaining a strong generalization ability, and for the precision analysis of the regression problem, a decision coefficient R is used2To judge the accuracy of the prediction.
Further, the base learner in S3 is a strong learner XGBOOST, an objective function of XGBOOST is divided into a loss function and a regularization term, and optimization of the objective function is divided into a second-order taylor expansion optimization loss function; the regularization term expands an optimized regularization term; and merging the coefficients to obtain a final objective function.
Further, the base learner in S3 is a strong learner GBDT, which is a gradient boosting decision tree, and is an addition model based on boosting, and gradually approaches to an optimization objective function by using a forward distribution algorithm, and in a regression problem, a negative gradient fitting method is usually used to solve the fitting problem of the loss function.
Further, in S2, the coordinates of the label are (x)i,yi) The received signal strength index of each reader is { RSSIi1,RSSIi2,...,RSSIikCombining the signal strength index of the label with the coordinate thereof to form a sample { RSSI }i1,RSSIi2,...,RSSIik,xi,yiWhere i denotes the ith reference tag and k denotes the kth reader.
Further, the base learner in S3 is a strong learner SVR, and a support vector machine is used for the regression problem for a training data set
Figure BDA0003297621700000031
kiFeature vector representing ith reference tag, representing signal strength indicator { RSSIi1,RSSIi2,...,RSSIik}；hiThe coordinate value (x) corresponding to the signal strength indexi,yi) (ii) a n represents the number of pieces of data; d, representing the model corresponding to the hyperplane divided in the feature space as follows:
f(k)＝ωTφ(k)+b
wherein ω represents a normal vector; k represents a feature vector; b represents a displacement term; phi (k) is a feature vector after k is mapped;
when the SVR is used for RFID label position prediction, the optimization target of the SVR can be formalized as follows:
Figure BDA0003297621700000032
wherein C represents a regularization constant; e represents the allowable model outputs f (k) and trueMaximum deviation of real value h; l∈Is ∈ -insensitive loss function:
Figure BDA0003297621700000041
the invention has the advantages and beneficial effects that:
according to the invention, the collected RSSI value of the label is filtered to obtain a more stable data set, so that the trained model is more accurate and the prediction is more accurate. The effect of a plurality of strong learners is fused by using a stacking model, and a positioning effect which is more accurate than that of a single model is obtained.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a diagram of a positioning environment arrangement in the present invention.
Fig. 3 is a diagram of a generation process of a secondary training set of the stacking model in the present invention (K ═ 5).
Fig. 4 is a flow chart of the training model of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the present invention, are given by way of illustration and explanation only, not limitation.
As shown in fig. 1, an RFID indoor positioning method based on a stacking model includes the following steps:
step 1: as shown in fig. 2, the devices are arranged in the required location area, including arranging k readers in the location area, and the readers are arranged as uniformly as possible on the premise that all tags can be read; arranging m reference tags on the ground at equal intervals in a rectangular shape; and randomly placing a plurality of labels to be positioned.
Step 2: the RSSI value of the tag is received by the reader, and the RSSI values of the reference tag and the tag to be positioned can be received in batches. For reference tag TiLet its coordinate be (x)i,yi) The received signal strength index of each reader is { RSSIi1,RSSIi2,...,RSSIikCombining the signal strength index of the reference label with the coordinate thereof to form a training sample { RSSI }i1,RSSIi2,...,RSSIik,xi,yiAnd reading the same reference label for multiple times to obtain multiple training samples. And finally integrating the training samples acquired by all the reference labels to serve as an original data set for subsequent use. And collecting the RSSI value and the coordinates of the label to be positioned by the same method, and integrating the RSSI value and the coordinates into a data set for verifying the positioning effect of the model after training is finished.
And step 3: and (3) processing the original data set obtained in the step (2) by using a Kalman filtering algorithm, smoothing the RSSI value to obtain a data set D containing less dynamic RSSI data, and performing subsequent model training by using the data set D to obtain a more accurate model so as to accurately infer the position of the RFID tag.
Further, the kalman filtering algorithm is mainly divided into three stages: an initialization phase, a prediction phase and an update phase.
An initialization stage: the initial value of the filter state is set,
a prediction stage: and estimating the state of the sample at the current moment according to the state quantity and the control quantity of the sample at the previous moment, and calculating a corresponding covariance matrix. The process is expressed by the following formula:
Figure BDA0003297621700000051
Pt -＝FPt-1FT+Q
wherein the content of the first and second substances,
Figure BDA0003297621700000052
is a prior state estimation value of RSSI at the time t;
Figure BDA0003297621700000053
is a posterior state estimation value of RSSI at the time of t-1; u. oft-1Is a control variable;f is a state transition matrix; b is a control matrix. Pt -Estimating covariance a priori for time t; pt-1Estimating covariance for the posteriori at time t-1; q is the process noise covariance.
And (3) an updating stage: the kalman filter uses the received measurements to correct the estimates obtained during the prediction phase and updates the parameters used in the filter. The method comprises the following specific steps: firstly, calculating Kalman gain, then updating the prior state estimation value to an optimal value, namely a posterior state estimation value, according to the obtained Kalman gain, and finally calculating a covariance matrix corresponding to the optimal value. The process is expressed by the following formula:
Figure BDA0003297621700000054
Figure BDA0003297621700000055
Pt＝(I-KtH)Pt -
wherein, KtIs the Kalman gain at time t; h is an observation matrix; r is a measurement error; z is a radical oftIs a measured value;
Figure BDA0003297621700000056
as a residual between the actual and expected observations, PtEstimating covariance for the posteriori at time t; and I is an identity matrix.
And 4, step 4: as shown in fig. 3 and 4, the data set obtained in step 3 is trained using a two-layer stacking model. The first layer model regresses the dataset using RandomForest, XGBOOST, GBDT, SVR. And the second layer selects a simpler Logistic regression model to integrate the effect of the first layer model.
Further, RandomForest introduces random attribute selection including sample randomness and feature randomness on the basis of integrating decision trees together by a bagging method. For each decision tree, training is performed randomly and with a drop backTraining samples are extracted from training sets, and partial features are extracted from all the features randomly, so that strong generalization capability is obtained. For precision analysis of regression problems, a decision coefficient R is used2To judge the accuracy of the prediction.
The target function of the XGBOOST is divided into a loss function and a regularization item, and the optimization of the target function can be divided into a second-order Taylor expansion optimization loss function; the regularization term expands an optimized regularization term; and merging the coefficients to obtain a final objective function.
The GBDT is also called a gradient lifting decision tree, which is an addition model based on boosting thought, a forward distribution algorithm is utilized to gradually approach an optimization objective function, and in a regression problem, a negative gradient fitting method is generally used to solve the fitting problem of a loss function.
SVR uses support vector machines for regression problem, for a training sample set
Figure BDA0003297621700000061
The model corresponding to the hyperplane divided in the feature space can be expressed as follows:
f(k)＝ωTφ(k)+b
where ω is a normal vector; k is a feature vector; b is a displacement term; phi (k) is the feature vector after k is mapped.
When the SVR is used for RFID label position prediction, assuming that the maximum deviation between the allowable model output f (k) and the true value h is E, the optimization target of the SVR can be formalized as follows:
Figure BDA0003297621700000062
wherein ω is a normal vector; b is a displacement term; c is a regularization constant; k is a radical ofiIs a feature vector, i.e., the signal strength indicator { RSSI } collected in step 2i1,RSSIi2,...,RSSIik}；hiIs a coordinate value, x, of the corresponding labeli、yiRespectively transmitting the training signals into the training machine for training; n represents the number of pieces of data; l∈Is ∈ -insensitive loss function:
Figure BDA0003297621700000063
further, before the first layer model training, the data set obtained in the step 3 is divided into a training set DtrainAnd test set Dtest. The training of the first layer model adopts the thought of K-fold cross validation, and the divided training set DtrainAnd dividing the test data into K parts again, taking 1 part as a test set, taking the rest K-1 parts as a training set, predicting the test set after training by a base learner, and integrating K predicted values according to the positions of the original test set to obtain a predicted data set corresponding to the learner. For example, when performing regression prediction using RandomForest, the training set D will betrainAre equally divided into D1,D2,…,DKTaking D1For this test set, D2,D3,…,DKFor this training set, regression training is performed on D using the RandomForest model1Performing prediction to obtain a prediction set P1(ii) a Then get D2For this test set, D1,D3,…,DKFor the training set, obtaining a prediction set P after regression training2(ii) a K prediction sets P are obtained after K times of training1,P2,…,PKIntegrating K prediction sets into a final prediction set P of RandomForestRF＝{P1,P2,…,PK}; in the same way, a prediction set P corresponding to XGBOST, GBDT and SVR is obtainedXGB，PGBDT，PSVR(ii) a The secondary training set P ═ { P } resulting from the first layer model trainingRF,PXGB,PGBDT,PSVRAnd (4) taking the secondary training set P as the characteristic input of the Logistic regression of the second-layer model, taking the corresponding RFID label coordinates as the label input, obtaining a complete stacking model after the training of the second-layer model is finished, and using the test set DtestAnd testing the effect of the stacking model, and comparing the effect with the effect of a single base learner to judge whether the model parameters need to be improved.
And 5: and (4) predicting the RSSI value of the label to be positioned processed by the Kalman filter in the step (3) by using a complete stacking model to obtain the coordinate of the label to be positioned.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. An RFID indoor positioning method based on a stacking model is characterized by comprising the following steps:
s1, deploying a reader, a reference tag and a tag to be positioned in the positioning range;
s2, the RSSI values returned by the labels are received through the reader, the RSSI values read by all the readers of each reference label are combined with the coordinates of the reference label to serve as a training sample, the same reference label is read for multiple times to obtain a plurality of training samples, and the training samples of all the reference labels are integrated into a training data set; for the tags to be positioned, combining the RSSI value read by all readers of each tag to be positioned with the coordinates of the tag to be positioned to serve as a sample to be positioned, and integrating the sample to be positioned into a data set to be positioned in the same way;
s3, constructing a stacking model, training through a training data set, wherein the stacking model is a layered model and comprises a base learner and a logistic regression layer, and the method comprises the following steps:
s31, before learning, dividing the training data set into training sets DtrainAnd test set Dtest；
S32, adopting K-fold cross validation in the training of the base learners, and inputting the training set D for each base learnertrainDividing into K parts again, taking 1 part as a test set, taking the rest K-1 parts as a training set, predicting the test set after training by a base learner to obtain a predicted value, taking 1 part as the test set in sequence, and pre-predicting the obtained K partsMeasuring values, and integrating according to the positions of the original test set to obtain a prediction data set corresponding to the base learner; merging prediction data sets generated by all the base learners on the same layer to obtain a secondary training set P;
s33, in the training of the logistic regression layer, the secondary training set P is used as the characteristic input of the training set, the corresponding reference label coordinate is used as the label input of the training set, and after the training is finished, the test set D is passedtestTesting and comparing with the effect of a single base learner to judge whether the model parameters are improved;
and S4, inputting the RSSI value in the data set to be positioned into a trained stacking model to obtain the presumed coordinates of the label to be positioned.
2. The RFID indoor positioning method based on the stacking model as claimed in claim 1, wherein the RSSI value of the training data set obtained in S2 is smoothed through Kalman filtering to obtain a data set containing smaller fluctuation RSSI data as the input of S3 training stacking model; and smoothing the RSSI value of the data set to be positioned obtained in the step S2 through Kalman filtering, and taking the RSSI value as the input of the training model in the step S4.
3. The RFID indoor positioning method based on the stacking model as claimed in claim 1, wherein the Kalman filtering algorithm is divided into three stages: an initialization phase, a prediction phase and an updating phase;
the initialization phase: setting an initial value of the state of the filter;
the prediction stage is as follows: estimating the state of the training sample at the current moment according to the state quantity and the control quantity of the training sample at the previous moment, and calculating a corresponding covariance matrix, wherein the process is represented by the following formula:
Figure FDA0003297621690000011
Pt -＝FPt-1FT+Q
wherein the content of the first and second substances,
Figure FDA0003297621690000021
is a prior state estimation value of RSSI at the time t;
Figure FDA0003297621690000022
is a posterior state estimation value of RSSI at the time of t-1; u. oft-1Is a control variable; f is a state transition matrix; b is a control matrix; pt -Estimating covariance a priori for time t; pt-1Estimating covariance for the posteriori at time t-1; q is the process noise covariance;
the updating stage comprises the following steps: the estimated values obtained in the prediction phase are corrected by means of the measured values received and the parameters used in the filter are updated, the specific steps being: firstly, calculating Kalman gain, then updating the prior state estimation value to an optimal value, namely a posterior state estimation value according to the Kalman gain, and finally calculating a covariance matrix corresponding to the optimal value, wherein the process is represented by the following formula:
Figure FDA0003297621690000023
Figure FDA0003297621690000024
Pt＝(I-KtH)Pt -
wherein, KtIs the Kalman gain at time t; h is an observation matrix; r is a measurement error; z is a radical oftIs a measured value;
Figure FDA0003297621690000025
as a residual between the actual and expected observations, PtEstimating covariance for the posteriori at time t; and I is an identity matrix.
4. The RFID indoor positioning method based on the tracking model as claimed in claim 1, wherein the tag to be positioned for verification is used as a verification sample in S2, the verification data set is obtained through S1, the RSSI value in the verification data set is input into the trained tracking model in S4, the coordinates of the tag to be positioned are estimated and compared with the corresponding actual position in the verification data set, the tracking model is verified, and then the tag to be positioned which is actually to be positioned is estimated.
5. The method of claim 1, wherein in the step S1, the reference tags are arranged at equal intervals in a rectangle, and the arrangement of the readers is such that all tags can be read.
6. The method of claim 1, wherein the base learner in S3 is a strong learner randomfort, the decision trees are integrated by a bagging method, random attribute selection is introduced, including sample randomization and feature randomization, for each decision tree, training samples are randomly and replaceably extracted from a training set, part of features are randomly extracted from all features, and a decision coefficient R is used for precision analysis of regression problem2To judge the accuracy of the prediction.
7. The method of claim 1, wherein the base learner in the S3 is a strong learner XGBOOST, an objective function of XGBOOST is divided into a loss function and a regularization term, and an optimization of the objective function is divided into a second-order taylor expansion optimization loss function; the regularization term expands an optimized regularization term; and merging the coefficients to obtain a final objective function.
8. The method of claim 1, wherein the base learner in S3 is a strong learner GBDT, GBDT is a gradient boosting decision tree, and is an additive model based on boosting, the forward distribution algorithm is used to approach the optimization objective function step by step, and in the regression problem, the fitting problem of the loss function is usually solved by using a negative gradient fitting method.
9. The method of claim 1, wherein in the step S2, the coordinates of the tag are (x)i,yi) The received signal strength index of each reader is { RSSIi1,RSSIi2,...,RSSIikCombining the signal strength index of the label with the coordinate thereof to form a sample { RSSI }i1,RSSIi2,...,RSSIik,xi,yiWhere i denotes the ith reference tag and k denotes the kth reader.
10. The method of claim 9, wherein the base learner in S3 is a strong learner SVR, and the support vector machine is used for regression, and D { (k) is used for a training data set1,h1),(k2,h3),…,(kn,hn)},
Figure FDA0003297621690000031
kiRepresenting the signal strength indicator { RSSI ] for the eigenvector of the ith reference tagi1,RSSIi2,...,RSSIik}；hiThe coordinate value (x) corresponding to the signal strength indexi,yi) (ii) a n represents the number of pieces of data; d, representing the model corresponding to the hyperplane divided in the feature space as follows:
f(k)＝ωTφ(k)+b
wherein ω represents a normal vector; k represents a feature vector; b represents a displacement term; phi (k) is a feature vector after k is mapped;
when the SVR is used for RFID label position prediction, the optimization target of the SVR can be formalized as follows:
Figure FDA0003297621690000032
wherein C represents a regularization constant; e represents the maximum deviation of the allowable model output f (k) from the true value h; l∈Is ∈ -insensitive loss function:
Figure FDA0003297621690000033
CN202111181914.9A 2021-10-11 2021-10-11 RFID indoor positioning method based on stacking model Pending CN113945215A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111181914.9A CN113945215A (en) 2021-10-11 2021-10-11 RFID indoor positioning method based on stacking model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111181914.9A CN113945215A (en) 2021-10-11 2021-10-11 RFID indoor positioning method based on stacking model

Publications (1)

Publication Number Publication Date
CN113945215A true CN113945215A (en) 2022-01-18

Family

ID=79329619

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111181914.9A Pending CN113945215A (en) 2021-10-11 2021-10-11 RFID indoor positioning method based on stacking model

Country Status (1)

Country Link
CN (1) CN113945215A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102692618A (en) * 2012-05-23 2012-09-26 浙江工业大学 RFID (radio frequency identification) positioning method based on RSSI (received signal strength indicator) weight fusion
CN103997717A (en) * 2014-06-12 2014-08-20 福建师范大学 Real-time indoor positioning system and method
CN110996280A (en) * 2019-11-06 2020-04-10 山东大学 RFID indoor positioning fingerprint database updating system and method
CN111461251A (en) * 2020-04-10 2020-07-28 桂林电子科技大学 Indoor positioning method of WiFi fingerprint based on random forest and self-encoder

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102692618A (en) * 2012-05-23 2012-09-26 浙江工业大学 RFID (radio frequency identification) positioning method based on RSSI (received signal strength indicator) weight fusion
CN103997717A (en) * 2014-06-12 2014-08-20 福建师范大学 Real-time indoor positioning system and method
CN110996280A (en) * 2019-11-06 2020-04-10 山东大学 RFID indoor positioning fingerprint database updating system and method
CN111461251A (en) * 2020-04-10 2020-07-28 桂林电子科技大学 Indoor positioning method of WiFi fingerprint based on random forest and self-encoder

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
谷民阳: "仓储危化品定位管理系统研究与开发", 《中国优秀硕士学位论文全文数据库工程科技I辑》, no. 2, pages 2 - 7 *

Similar Documents

Publication Publication Date Title
CN105809672B (en) A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring
CN114564982B (en) Automatic identification method for radar signal modulation type
CN104952753A (en) Measurement Sampling Method
CN110765882B (en) Video tag determination method, device, server and storage medium
CN106663086A (en) Apparatus and method for ensembles of kernel regression models
CN113033665A (en) Sample expansion method, training method and system, and sample learning system
CN113539909A (en) Fault detection method and device, terminal equipment and storage medium
CN113591839B (en) Feature extraction model construction method, target detection method and device
CN113128518B (en) Sift mismatch detection method based on twin convolution network and feature mixing
CN108919182B (en) Target positioning method based on support set and expectation maximization in WIFI environment
CN109859243A (en) A kind of motion target tracking method based on dimension self-adaption block particle
CN113945215A (en) RFID indoor positioning method based on stacking model
CN116720079A (en) Wind driven generator fault mode identification method and system based on multi-feature fusion
CN116228776A (en) Electromechanical equipment welding defect identification method and system
CN116563841A (en) Detection method and detection device for power distribution network equipment identification plate and electronic equipment
CN114708307B (en) Target tracking method, system, storage medium and device based on correlation filter
CN110781968A (en) Extensible class image identification method based on plastic convolution neural network
US11645827B2 (en) Detection method and device for assembly body multi-view change based on feature matching
CN111160464B (en) Industrial high-order dynamic process soft measurement method based on multi-hidden-layer weighted dynamic model
CN113971737A (en) Object recognition method for robot, electronic device, medium, and program product
CN114374931A (en) Fingerprint positioning method based on measurement learning of neighbor component analysis
Nagy Data-driven analysis of fractality and other characteristics of complex networks
Bhattacharjee et al. Active learning for imbalanced domains: the ALOD and ALOD-RE algorithms
CN117708691B (en) Intermittent process monitoring method, storage medium and computer equipment
CN112802547B (en) Gene detection method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination