CN111209960A - CSI system multipath classification method based on improved random forest algorithm - Google Patents

CSI system multipath classification method based on improved random forest algorithm Download PDF

Info

Publication number
CN111209960A
CN111209960A CN202010008037.4A CN202010008037A CN111209960A CN 111209960 A CN111209960 A CN 111209960A CN 202010008037 A CN202010008037 A CN 202010008037A CN 111209960 A CN111209960 A CN 111209960A
Authority
CN
China
Prior art keywords
samples
class
feature
los
nlos
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.)
Granted
Application number
CN202010008037.4A
Other languages
Chinese (zh)
Other versions
CN111209960B (en
Inventor
史伟光
李耀辉
李婉琪
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.)
Tianjin Polytechnic University
Original Assignee
Tianjin Polytechnic University
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 Tianjin Polytechnic University filed Critical Tianjin Polytechnic University
Priority to CN202010008037.4A priority Critical patent/CN111209960B/en
Publication of CN111209960A publication Critical patent/CN111209960A/en
Application granted granted Critical
Publication of CN111209960B publication Critical patent/CN111209960B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Probability & Statistics with Applications (AREA)
  • Mobile Radio Communication Systems (AREA)
  • Monitoring And Testing Of Transmission In General (AREA)

Abstract

The invention belongs to the field of wireless positioning, and relates to a CSI system multipath classification method based on an improved random forest algorithm. The method aims at distinguishing two propagation modes of LOS and NLOS, and comprises the following steps: clustering the energy of all samples by using a K-means clustering algorithm according to the propagation characteristics of radio signals in LOS and NLOS environments, and constructing a characteristic factor based on the K-means clustering algorithm; calculating the inter-class scattering distance and the intra-class scattering distance of all samples, and obtaining an optimal characteristic combination according to a Fisher criterion; and training and testing different feature combinations by using a random forest algorithm based on a C4.5 algorithm to finish multipath classification. The invention has the characteristics of effectively avoiding the phenomenon that the multipath classification is restricted by an accurate threshold value and meeting the requirements of high accuracy and low computation amount in the multipath classification application scene.

Description

CSI system multipath classification method based on improved random forest algorithm
Technical Field
The invention belongs to the field of wireless positioning, and relates to a CSI system multipath classification method based on an improved random forest algorithm.
Background
The WiFi positioning technology has the advantages of economy, convenience, rapidness, and easy deployment, and has been successfully applied to a plurality of fields such as industrial automation, commercial automation, transportation control and management in recent years. Especially in an indoor positioning application scene, the WiFi positioning system under an ideal condition can obtain higher positioning precision, and great development potential and practical value are shown.
In an indoor positioning system based on WiFi, positioning accuracy is a basic index for evaluating positioning performance, and the complexity and variability of indoor environment cause the positioning accuracy to be often limited by factors such as multiple access interference, time delay in a circuit, multipath propagation and the like. Under the condition of multipath propagation, the propagation modes of signals are classified into Line of sight (LOS) propagation and Non-Line of sight (NLOS) propagation. Under the condition of LOS propagation, a wireless signal is directly propagated from a transmitting end to a receiving end, the measurement information of the received signal is relatively stable, and under the condition of NLOS propagation, a straight line path between the transmitting end and the receiving end is blocked, the signal can only reach the receiving end through diffraction, reflection or diffraction, and the measurement information of the received signal can be deviated, so that the positioning accuracy is influenced. The classification of the two signal propagation modes is called multipath classification. In order to improve the accuracy of indoor positioning, researchers have made relevant studies on the problem of multipath classification, but the classification accuracy needs to be further improved.
Based on the above background, the present invention provides a CSI system multipath classification method based on an improved random forest algorithm, aiming at achieving multipath classification with higher accuracy and lower computation, and using Channel State Information (CSI) as a multipath classification reference.
Disclosure of Invention
The invention aims to provide a CSI system multipath classification method based on an improved random forest algorithm. The method comprises the steps of firstly analyzing the propagation characteristics of signals in LOS and NLOS environments, constructing a characteristic factor based on a K-means clustering algorithm, then calculating the inter-class scattering distance and the intra-class scattering distance of different classes according to the Fisher criterion for different classification performances of different characteristic combinations, screening the characteristics according to the Fisher criterion, finally establishing a random forest classification framework based on a C4.5 algorithm, training and testing different characteristic combinations, and obtaining classification results by adopting a simple majority voting mechanism.
The method comprises the following specific steps:
step 1: and establishing a multipath classification system model by taking the distribution characteristics of the statistical characteristics in the LOS and NLOS propagation environments as modeling conditions. The system comprises three parts, namely a data acquisition and preprocessing module, a feature extraction module and a multipath classification module. And acquiring CSI data by using a notebook carrying the wireless network card, performing data preprocessing operation, and performing feature extraction on the preprocessed CSI samples.
Step 2: and constructing a characteristic factor based on a K-means clustering algorithm on the basis of the propagation characteristics of radio signals in LOS and NLOS environments. Considering that the signal energy of the CIR samples in the LOS environment is mainly concentrated in the main path, and the sampling points adjacent to the main path are samples belonging to the same class as the main path, the CIR samples collected under the LOS and NLOS scenes are clustered according to the difference of the energy of the CIR samples through a K-means clustering algorithm.
And step 3: in an LOS scene, firstly, K points are randomly selected from the amplitude of an acquired CIR sample to serve as an initial clustering center, then each point is allocated to the nearest clustering center to form K clusters, then the clustering center of each cluster is recalculated and continuously updated until the iteration times reach the maximum or the elements in the clusters do not change, then the distances of the clustering centers are compared, and the class with the larger distance of the clustering centers is selected to serve as the class to be classified by the method, wherein the class represents the number of multi-paths in the environment.
And 4, clustering the amplitude of the CIR samples under the NLOS environment by adopting the method in the step 3, and defining characteristic factors based on the K-means clustering algorithm as ξ -1/lg (M) -M/R, wherein M represents the sum of all energies of the class to which the maximum energy value belongs in the collected CIR samples, R represents the sum of energies of other path samples except the class to which the maximum energy value belongs in the CIR samples, and-1/lg (M) represents a distance factor.
And 5: each feature belonging to LOS and NLOS is evaluated according to the fisher criterion, considering that different feature combinations will have different classification performance. Under each feature, define the intra-class scattering distance of all samples as
Figure RE-GSB0000186473650000031
In the formula
Figure RE-GSB0000186473650000032
Where k denotes the feature index number, m denotes the number of classified categories, m is 2 in the present invention, x denotes the sample data set, x(k)A set of samples representing the kth feature,
Figure RE-GSB0000186473650000033
mean, D, of the kth set of feature samples representing the ith class of samplesiSet of samples representing class i, niIndicating the number of samples belonging to the i-th class. The intra-class discrete distance characterizes the variance of samples belonging to the same class over the measured features. On the measured characteristics, the scattering distance between classes of all samples is defined as
Figure RE-GSB0000186473650000041
Wherein
Figure RE-GSB0000186473650000042
n denotes the total number of samples of each class, D denotes the sample set, μ(k)Representing the mean of the kth feature sample set among all samples, the inter-class dispersion distance characterizes the similarity of samples belonging to different classes at feature k.
Step 6: by the intra-class scattering distance in step 5
Figure RE-GSB0000186473650000043
And inter-class scattering distance
Figure RE-GSB0000186473650000044
On the basis of which the discriminant function of the feature k is calculated
Figure RE-GSB0000186473650000045
And 7: and (5) screening different feature combinations according to the feature screening method in the step (5) and the step (6), and then training and testing by using a random forest algorithm based on a C4.5 algorithm. Calculating the information gain rate of each feature, selecting the feature with the outstanding information gain rate as a root node, and reserving other features for further splitting, applying the same splitting criterion at each decision node, and enabling the tree to continuously grow until the node becomes a leaf node containing the prediction category, wherein the path from the root node to the leaf node represents a classification rule.
In step 7, an attribute "x" is defined1”、“x2"and" x3"three different features randomly obtained from the training data set, LOS and NLOS class labels are respectively set as" 1 "and" -1 ", the feature values are classified into left child node and right child node, if feature" x3"has a value of 3.4535 and is the root node, it means" x "in all the features3"the information gain ratio is highest," x3"the right child node becomes a leaf node, the class of which is predicted to be 1, which means that all features" x "greater than 3.45353The values of all "belong to LOS, the entropy of the right child node is zero, and" x "is3"the left child node is further split, and the splitting criterion is applied recursively at other decision nodes until a leaf node is reached, indicating that the decision tree construction is complete. In the testing phase, each value to be tested passes through all trees in the forest simultaneously, starting from the root node until it reaches the corresponding leaf node. Obtaining a predicted value of a random forest algorithm through majority voting of each tree, wherein an output predicted value of a tree P (P is 1, the right, and P) is y according to the ith observed value of the test data setipThe prediction output of the whole random forest algorithm is
Figure RE-GSB0000186473650000051
The classification result can be described as being based on yiBy hypothesis testing of, i.e. when yiWhen the value is more than or equal to 0, the LOS is obtained, and when y is greater than or equal to 0iIf < 0, it belongs to NLOS.
Description of the drawings:
FIG. 1 is a structural frame diagram of the present invention;
FIG. 2 is a flow chart of feature factor construction based on a K-means clustering algorithm;
FIG. 3 is a schematic diagram of a decision tree structure in a random forest algorithm;
FIG. 4 is a schematic view of a measured scene;
FIG. 5 is a complexity of the environment in different scenarios;
the specific implementation mode is as follows:
firstly, a data acquisition system is constructed on the basis of a computer and a router which are provided with a wireless network card, CSI data under LOS and NLOS environments are acquired, the characteristics of a preprocessed sample are extracted, the classification performance is improved, and a characteristic factor based on a K-means clustering algorithm is provided by introducing the K-means clustering algorithm.
Under the condition of NLOS propagation, no direct-view path exists in the signal propagation process, so that the signal energy distribution of each subcarrier received by a receiving end is random, and a plurality of paths with relatively large signal energy exist, on the basis, the CIR samples acquired under LOS and NLOS scenes are clustered according to different energies by using the K-means clustering algorithm, firstly, K points are randomly selected as initial clustering centers for the amplitudes of the acquired CIR samples, then, each point is allocated to the nearest clustering center to form K clusters, then, the centers of each cluster are recalculated and continuously updated until the iteration number reaches the maximum or elements in the clusters are not changed, then, the distances between the clustering centers are compared, the cluster center is selected as the class of the cluster, and the number of the cluster symbols is represented as ξ
Figure RE-GSB0000186473650000061
Where M represents the sum of all energies in the class to which the maximum energy value belongs in the collected CIR samples, R represents the sum of energies of other path samples in the CIR samples except the class to which the maximum energy value belongs, -1/lg (M) represents the distance factor.
And (4) screening the features according to a Fisher criterion aiming at obtaining the optimal feature combination in view of different classification performances of different feature combinations. First, each feature belonging to LOS and NLOS is processed independently, under each feature, the intra-class scattering distance of all samples is
Figure RE-GSB0000186473650000062
Figure RE-GSB0000186473650000063
Figure RE-GSB0000186473650000064
Where k denotes the feature index number, m denotes the number of classified categories, m is 2 in the present invention, x denotes the sample data set, x(k)A set of samples representing the kth feature,
Figure RE-GSB0000186473650000065
mean, D, of the kth set of feature samples representing the ith class of samplesiSet of samples representing class i, niIndicating the number of samples belonging to the i-th class. On the measured characteristic, the scattering distance between classes of all samples is
Figure RE-GSB0000186473650000071
Figure RE-GSB0000186473650000072
Where n represents the total number of samples of each class, D represents the sample set, μ(k)Representing the mean of the kth feature sample set in all samples, the discriminant function of feature k is
Figure RE-GSB0000186473650000073
The above discriminant function is the criterion for feature screening of the present invention.
And finally, training and testing different feature combinations by using a random forest algorithm based on a C4.5 algorithm. FIG. 3 is a schematic diagram of a decision tree structure in a random forest algorithm, with attribute "x1”、“x2"and" x3"three different features randomly obtained from the training dataset, respectively, the class labels for LOS and NLOS are set to" 1 "and" -1 ", respectively. The feature values are now classified into left and right child nodes if the feature "x3"has a value of 3.4535, and in FIG. 3, the feature" x3"is the root node, then its information gain rate is highest in all features," x3"becomes a leaf node, the class of which is predicted to be 1, and is greater than 3.45353The values of all "belong to LOS, the entropy of the right child node is zero, and" x "is3"the left child node is further split, and the splitting criterion is applied recursively at other decision nodes until a leaf node is reached, indicating that the decision tree construction is complete.
In the testing phase, each value to be tested passes through all trees in the forest simultaneously, starting from the root node until it reaches the corresponding leaf node. Obtaining a predicted value of a random forest algorithm through majority voting of each tree, wherein an output predicted value of a tree P (P is 1, the right, and P) is y according to the ith observed value of the test data setipThe prediction output y of the whole random forest algorithmiCan be expressed as
Figure RE-GSB0000186473650000081
The classification result can be described as being based on yiBy hypothesis testing of, i.e. when yiWhen the value is more than or equal to 0, the LOS is obtained, and when y is greater than or equal to 0iIf < 0, it belongs to NLOS.
400 groups of CSI sampling data packets are collected in a laboratory by utilizing a notebook carrying a wireless network card, wherein 200 groups of data packets are in an NLOS environment, the rest 200 groups of data packets are in an LOS environment, the distance between the notebook and an AP is 2.5m, and CSI is obtained by ping the AP under the terminal of the notebook. From 400 collected data packets, 300 data packets are selected for training a classification model, 100 data packets are left for testing the classification performance of the model, 100 decision trees are used, 100 Monte Carlo experiments are carried out, and 100 classification results are averaged to obtain the final classification accuracy. The classification performance of the method is verified under three different scenes, the complexity of the environment under the different scenes is shown in fig. 5, the actually measured scene graph is shown in fig. 4, fig. 4(a) and (b) are LOS and NLOS scene graphs of a scene I respectively, the classification accuracy of the method in the scene I is 97.80%, and the false alarm probability and the missed detection probability of the NLOS are 1.24% and 3.57% respectively. Fig. 4(c) and (d) are LOS and NLOS scene graphs of scene two, respectively, the classification accuracy of the method in scene two is 94.48%, and the false alarm probability and the false drop probability of NLOS are 6.14% and 5.03%, respectively. Fig. 4(e) and (f) are LOS and NLOS scene graphs of scene three, respectively, the classification accuracy of the above method in scene three is 88.72%, and the false alarm probability and the false drop probability of NLOS are 6.34% and 16.23%, respectively.

Claims (2)

1. A CSI system multipath classification method based on an improved random forest algorithm comprises the following specific steps:
step 1: establishing a multipath classification system model by taking the distribution characteristics of statistical characteristics in LOS and NLOS propagation environments as modeling conditions, wherein the system comprises three parts, namely a data acquisition and preprocessing module, a characteristic extraction module and a multipath classification module, and is characterized in that a notebook carrying a wireless network card is used for acquiring CSI data, executing data preprocessing operation and extracting the characteristics of preprocessed CSI samples;
step 2: based on the propagation characteristics of radio signals in LOS and NLOS environments, characteristic factors based on a K-means clustering algorithm are constructed, and in view of the fact that the signal energy of CIR samples in the LOS environment is mainly concentrated in a main path, and sampling points adjacent to the main path are samples belonging to the same class as the main path, the CIR samples acquired in the LOS and NLOS scenes are clustered according to the difference of the energy of the CIR samples through the K-means clustering algorithm;
and step 3: in an LOS scene, firstly, randomly selecting K points as initial clustering centers for the amplitude of an acquired CIR sample, then distributing each point to the nearest clustering center to form K clusters, recalculating the clustering center of each cluster and continuously updating until the iteration times reach the maximum or the elements in the clusters do not change, then comparing the distances of the clustering centers, and selecting the class with the larger distance of the clustering centers as the class to be classified by the invention, wherein the class number represents the number of multi-paths in the environment;
step 4, clustering the amplitude of the CIR samples under the NLOS environment by adopting the method in the step 3, and defining characteristic factors based on a K-means clustering algorithm as ξ -1/lg (M) -M/R on the basis, wherein M represents the sum of all energies of the class to which the maximum energy value belongs in the collected CIR samples, R represents the sum of energies of other path samples except the class to which the maximum energy value belongs in the CIR samples, and-1/lg (M) represents a distance factor;
and 5: considering that different feature combinations have different classification performances, each feature belonging to LOS and NLOS is evaluated according to the Fisher criterion, under each feature, the intra-class scattering distance of all samples is defined as
Figure FSA0000199377030000021
In the formula
Figure FSA0000199377030000022
Figure FSA0000199377030000023
Where k denotes the feature index number, m denotes the number of classified categories, m is 2 in the present invention, x denotes the sample data set, x(k)A set of samples representing the kth feature,
Figure FSA0000199377030000024
mean, D, of the kth set of feature samples representing the ith class of samplesiSet of samples representing class i, niRepresenting the number of samples belonging to the ith class, and the discrete distance in the class is characterized by the variance of the samples belonging to the same class on the measured feature, and the scattering distance between classes of all the samples on the measured feature is defined as
Figure FSA0000199377030000025
Wherein
Figure FSA0000199377030000026
n denotes the total number of samples of each class, D denotes the sample set, μ(k)Representing the mean value of the kth characteristic sample set in all samples, wherein the inter-class dispersion distance represents the similarity of samples belonging to different classes at the characteristic k;
step 6: by the intra-class scattering distance in step 5
Figure FSA0000199377030000027
And inter-class scattering distance
Figure FSA0000199377030000028
On the basis of which the discriminant function of the feature k is calculated
Figure FSA0000199377030000029
And 7: according to the feature screening method in the steps 5 and 6, different feature combinations are screened, then a random forest algorithm based on a C4.5 algorithm is used for training and testing, the information gain rate of each feature is calculated, the feature with the prominent information gain rate is selected as a root node, other features are left to be further split, the same segmentation standard is applied to each decision node, the tree is continuously grown until the node becomes a leaf node containing a prediction category, and the splitting is stopped, wherein a path from the root node to the leaf node represents a classification rule.
2. The method for classifying multipath of CSI system based on improved random forest algorithm as claimed in claim 1, wherein in step 7, attribute "x" is defined1”、“x2"and" x3"three different features randomly obtained from the training dataset, LOS and NLOS class labels are set to" 1 "and" -1 ", respectively, classifying the feature values as leftChild node and right child node if feature "x3"has a value of 3.4535 and is the root node, it means" x "in all the features3"the information gain ratio is highest," x3"the right child node becomes a leaf node, the class of which is predicted to be 1, which means that all features" x "greater than 3.45353The values of all "belong to LOS, the entropy of the right child node is zero, and" x "is3"the left child node is further split, and the splitting criterion is recursively applied at other decision nodes until reaching a leaf node, which indicates that the decision tree is completely constructed; in the testing stage, each value to be tested simultaneously passes through all trees in the forest, starting from the root node until the value reaches the corresponding leaf node, the predicted value of the random forest algorithm is obtained through majority voting of each tree, and the output predicted value of the tree P (P is 1,ipthe prediction output of the whole random forest algorithm is
Figure FSA0000199377030000031
The classification result can be described as being based on yiBy hypothesis testing of, i.e. when yiWhen the value is more than or equal to 0, the LOS is obtained, and when y is greater than or equal to 0iIf < 0, it belongs to NLOS.
CN202010008037.4A 2020-01-06 2020-01-06 CSI system multipath classification method based on improved random forest algorithm Active CN111209960B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010008037.4A CN111209960B (en) 2020-01-06 2020-01-06 CSI system multipath classification method based on improved random forest algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010008037.4A CN111209960B (en) 2020-01-06 2020-01-06 CSI system multipath classification method based on improved random forest algorithm

Publications (2)

Publication Number Publication Date
CN111209960A true CN111209960A (en) 2020-05-29
CN111209960B CN111209960B (en) 2024-01-05

Family

ID=70787825

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010008037.4A Active CN111209960B (en) 2020-01-06 2020-01-06 CSI system multipath classification method based on improved random forest algorithm

Country Status (1)

Country Link
CN (1) CN111209960B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112235816A (en) * 2020-10-16 2021-01-15 哈尔滨工程大学 WIFI signal CSI feature extraction method based on random forest
CN112637926A (en) * 2020-11-28 2021-04-09 重庆工商大学融智学院 Safe relay selection method based on random forest uniform quantitative feature extraction
CN112686221A (en) * 2021-03-11 2021-04-20 成都点泽智能科技有限公司 Pedestrian entrance and exit monitoring method, device and equipment and computer storage medium
CN114339649A (en) * 2022-01-10 2022-04-12 河北阿卡斯物联网科技有限公司 NLOS (non line of sight) identification system with WKNN classification function

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059852A (en) * 2019-03-11 2019-07-26 杭州电子科技大学 A kind of stock yield prediction technique based on improvement random forests algorithm
CN110381440A (en) * 2019-06-16 2019-10-25 西安电子科技大学 The fingerprint indoor orientation method of joint RSS and CSI based on deep learning
CN110636466A (en) * 2019-09-06 2019-12-31 联泰集群（北京）科技有限责任公司 WiFi indoor positioning system based on channel state information under machine learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059852A (en) * 2019-03-11 2019-07-26 杭州电子科技大学 A kind of stock yield prediction technique based on improvement random forests algorithm
CN110381440A (en) * 2019-06-16 2019-10-25 西安电子科技大学 The fingerprint indoor orientation method of joint RSS and CSI based on deep learning
CN110636466A (en) * 2019-09-06 2019-12-31 联泰集群（北京）科技有限责任公司 WiFi indoor positioning system based on channel state information under machine learning

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
DI HE等: "3-D spatial spectrum fusion indoor localization algorithm based on CSI-UCA smoothing technique", 《IEEE ACCESS》 *
吴自鹏: "WLAN环境下的AoA/PDR融合室内定位算法研究", 《中国优秀硕士学位论文全文数据库》 *
张萌等: "基于密度峰值聚类的随机森林室内定位", 《计算机工程与设计》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112235816A (en) * 2020-10-16 2021-01-15 哈尔滨工程大学 WIFI signal CSI feature extraction method based on random forest
CN112235816B (en) * 2020-10-16 2023-03-17 哈尔滨工程大学 WIFI signal CSI feature extraction method based on random forest
CN112637926A (en) * 2020-11-28 2021-04-09 重庆工商大学融智学院 Safe relay selection method based on random forest uniform quantitative feature extraction
CN112637926B (en) * 2020-11-28 2022-06-03 重庆财经学院 Safe relay selection method based on random forest uniform quantitative feature extraction
CN112686221A (en) * 2021-03-11 2021-04-20 成都点泽智能科技有限公司 Pedestrian entrance and exit monitoring method, device and equipment and computer storage medium
CN114339649A (en) * 2022-01-10 2022-04-12 河北阿卡斯物联网科技有限公司 NLOS (non line of sight) identification system with WKNN classification function
CN114339649B (en) * 2022-01-10 2024-03-01 河北阿卡斯物联网科技有限公司 WKNN classified NLOS recognition system

Also Published As

Publication number Publication date
CN111209960B (en) 2024-01-05

Similar Documents

Publication Publication Date Title
CN111209960B (en) CSI system multipath classification method based on improved random forest algorithm
McFee et al. Analyzing Song Structure with Spectral Clustering.
Liu et al. Isolation forest
CN110519128B (en) Random forest based operating system identification method
KR100930799B1 (en) Automated Clustering Method and Multipath Clustering Method and Apparatus in Mobile Communication Environment
CN111833172A (en) Consumption credit fraud detection method and system based on isolated forest
CN111506599B (en) Industrial control equipment identification method and system based on rule matching and deep learning
CN109450834A (en) Signal of communication classifying identification method based on Multiple feature association and Bayesian network
CN107729721B (en) Metabolite identification and disorder pathway analysis method
CN110019074B (en) Access path analysis method, device, equipment and medium
CN105335752A (en) Principal component analysis multivariable decision-making tree-based connection manner identification method
CN107729916B (en) ISODATA-based interference source classification and identification algorithm
CN108924756A (en) Indoor orientation method based on WiFi double frequency-band
CN110544047A (en) Bad data identification method
CN113037410A (en) Channel identification method, device, transmission method, transmission equipment, base station and medium
CN110765329B (en) Data clustering method and electronic equipment
CN106570104A (en) Multi-partition clustering preprocessing method of stream data
CN114355298B (en) Radar composite modulation pulse signal identification method
CN113489514B (en) Power line communication noise identification method and device based on self-organizing mapping neural network
CN110287237A (en) One kind analyzing efficient corporations&#39; data digging method based on social network structure
Wang et al. FCM algorithm and index CS for the signal sorting of radiant points
CN109408498A (en) The identification of time series feature and decomposition method based on eigenmatrix decision tree
CN113794653B (en) High-speed network traffic classification method based on sampling data flow
CN112788063B (en) LDoS attack detection method in SDN based on RF-GMM
CN113239880A (en) Radar radiation source identification method based on improved random forest

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
GR01 Patent grant
GR01 Patent grant