CN116405880B - Radio map construction method and system based on federal learning - Google Patents
Radio map construction method and system based on federal learning Download PDFInfo
- Publication number
- CN116405880B CN116405880B CN202310626603.1A CN202310626603A CN116405880B CN 116405880 B CN116405880 B CN 116405880B CN 202310626603 A CN202310626603 A CN 202310626603A CN 116405880 B CN116405880 B CN 116405880B
- Authority
- CN
- China
- Prior art keywords
- training
- crowdsourcing
- global
- samples
- parameters
- 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.)
- Active
Links
- 238000010276 construction Methods 0.000 title claims abstract description 56
- 238000012549 training Methods 0.000 claims abstract description 175
- 238000000034 method Methods 0.000 claims abstract description 48
- 230000005540 biological transmission Effects 0.000 claims description 33
- 238000005192 partition Methods 0.000 claims description 25
- 238000004422 calculation algorithm Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 13
- 238000012216 screening Methods 0.000 claims description 11
- 230000004927 fusion Effects 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 abstract description 24
- 230000008569 process Effects 0.000 abstract description 8
- 230000008859 change Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 6
- 238000010295 mobile communication Methods 0.000 description 5
- 238000005315 distribution function Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000001186 cumulative effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009469 supplementation Effects 0.000 description 1
Classifications
-
- 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
-
- 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
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0252—Radio frequency fingerprinting
- G01S5/02521—Radio frequency fingerprinting using a radio-map
- G01S5/02524—Creating or updating the radio-map
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Remote Sensing (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Radar, Positioning & Navigation (AREA)
- Mobile Radio Communication Systems (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention provides a radio map construction method and a system based on federal learning, which belong to the technical field of wireless communication and comprise the following steps: dividing a plurality of subareas of the target indoor area, and downloading parameters of the subareas to each training user node by a central node; acquiring crowdsourcing samples of each training user node, and training to obtain curve fitting parameters and curve signal intensity estimation error scores of all the subareas; uploading the updated global parameters to a central node by each training user node, and determining the updated global parameters by the central node to obtain the latest global parameters of all the subareas; and constructing a radio map, determining a target subarea by using the signal intensity characteristics, and determining the position of the to-be-positioned point in the target subarea. The invention effectively relieves the phenomenon that the signal intensity between indoor range areas is jumped and the problem that the change rule of the signal intensity of different areas is inconsistent, improves the construction quality of the whole radio map, and simultaneously saves the uplink and downlink communication expenditure in the training process.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to a radio map construction method and system based on federal learning.
Background
With the rapid development of wireless communication, the fifth generation mobile communication technology (5th Generation Mobile Communication Technology,5G) network has been commercially used on a large scale, and the sixth generation mobile communication technology (6th Generation Mobile Communication Technology,6G) has rapidly entered the field of view of the global mobile communication industry as a base for future communications. The 6G technology has great development potential in the fields of intelligent interaction, space communication, emotion and touch communication, touch Internet, multi-sense mixed reality, full-automatic traffic, machine-to-machine coordination and the like.
However, the biggest technical challenge faced by the 6G technology is that the terahertz wave frequency band where the 6G communication is located has many limitations, one is that the terahertz wave communication signal directionally propagates, the coverage area is small, only small cells can be formed, and the transmission path is weakened; secondly, the terahertz wave has the characteristic of super-strong fading, and the terahertz wave signal becomes very sensitive once encountering shadow and also affects the coverage area of communication, so that the terahertz wave signal is more suitable for indoor short-distance wireless communication; thirdly, intermittent connection and rapid channel fluctuation, the coherence time in the terahertz frequency band is very short, doppler dispersion is very large, and the attenuation fluctuation of the transmission channel of the terahertz wave is more obvious due to strong shadow attenuation. Therefore, it is very necessary to draw a radio map in 6G.
The model adopted in the conventional radio map construction is usually a spatial interpolation method, that is, the attribute of other unknown data points near the discrete points in the same area is estimated through spatial correlation and trend analysis between known discrete data, so that the observed data of the discrete points are converted into continuous data curved surfaces. However, the above spatial interpolation method does not take into consideration the problem that the difference between the whole sample values is often large, but directly performs spatial correlation analysis on all the known discrete data to construct a data curved surface of the indoor space, that is, does not take into consideration the influence of the indoor layout on the signal intensity, and the accuracy of the finally obtained radio map still needs to be improved; meanwhile, in the federal learning method, all model parameters in the training process are often transmitted, that is, the transmitted model parameters are not compressed, and the data transmission quantity of the final model is also required to be reduced.
Therefore, for the above method for constructing a radio map using federal learning, a solution for corresponding improvement needs to be proposed.
Disclosure of Invention
The invention provides a radio map construction method and system based on federal learning, which are used for solving the defects that in the prior art, aiming at the situation that the indoor range radio map constructed in a 6G network has uneven signal intensity distribution, the uplink and downlink communication transmission cost is overlarge, so that the construction process is low in precision and low in efficiency.
In a first aspect, the present invention provides a radio map construction method based on federal learning, including:
determining a plurality of training user nodes of the target indoor area;
dividing a plurality of subareas of the target indoor area, and downloading global parameters and global scores of the subareas to each training user node by a central node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node;
acquiring crowdsourcing samples of each training user node, and training to obtain surface fitting parameters and surface signal intensity estimation error scores of all subareas based on the crowdsourcing samples and the initial training parameters;
uploading the surface fitting parameters and the surface signal strength estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, wherein the central node determines to update the global parameters by utilizing the best surface fitting parameters and the best surface signal strength estimation error scores of all the subareas, and obtains the latest global parameters of all the subareas;
and constructing a radio map based on the latest global parameters, determining a target subarea in the radio map by utilizing signal intensity characteristics, and determining the position of a to-be-positioned point in the target subarea.
According to the radio map construction method based on federal learning provided by the invention, a plurality of training user nodes of a target indoor area are determined, and the method comprises the following steps:
dividing all user nodes into n user groups, and determining m users in the same user group during each round of distributed training;
if the user group with the number i is determined to train in the current training round, the user group with the number i+1 is determined to train in the next training round;
and after the n user groups finish traversing training, training is restarted from the 1 st user group.
According to the radio map construction method based on federal learning provided by the invention, crowd-sourced samples of each training user node are obtained, and the method comprises the following steps:
acquiring acquisition position coordinates and acquisition position received signal strength vectors of each training user node, wherein the acquisition position received signal strength vectors are determined by crowdsourcing sample numbers and total signal sources;
and forming crowdsourcing samples of each training user node by the acquisition position coordinates and the acquisition position received signal strength vector.
According to the radio map construction method based on federal learning, provided by the invention, the curve fitting parameters and the curve signal intensity estimation error scores of all the subareas are obtained based on the crowdsourcing samples and the initial training parameters, and the method comprises the following steps:
Determining a training sample threshold for each sub-region;
if the number of crowdsourcing samples in each sub-area is smaller than the training sample threshold, enhancing the crowdsourcing samples to obtain enhanced crowdsourcing samples;
if the number of the crowdsourcing samples in each sub-area is determined to be larger than the training sample threshold value multiplied by a preset proportion, the reliability weight of each crowdsourcing sample is obtained, and the crowdsourcing samples are screened based on the reliability weight to obtain screened crowdsourcing samples;
and calculating to obtain the curve fitting parameter and the curve signal intensity estimation error score by using the enhanced crowdsourcing sample or the screened crowdsourcing sample and the initial training parameter.
According to the radio map construction method based on federal learning provided by the invention, if the number of crowdsourcing samples of each sub-area is determined to be smaller than the training sample threshold, the crowdsourcing samples are enhanced to obtain enhanced crowdsourcing samples, and the method comprises the following steps:
obtaining a crowded-package sample average value, and determining to construct a random disturbance virtual sample, wherein the value range of the random disturbance virtual sample is between 0 and 1;
and adding the random disturbance virtual sample with the acquisition position coordinates and the acquisition position received signal strength vector in the crowdsourcing sample respectively to obtain the enhanced crowdsourcing sample.
According to the radio map construction method based on federal learning provided by the invention, if the number of crowdsourcing samples in each sub-area is determined to be larger than the training sample threshold value multiplied by a preset proportion, the reliability weight of each crowdsourcing sample is calculated by using a cross-domain clustering and crossing algorithm, the crowdsourcing samples are screened based on the reliability weight, and the screened crowdsourcing samples are obtained, and the method comprises the following steps:
calculating the reliability weight of each crowdsourcing sample by using a cross-domain clustering and crossing algorithm;
determining a partition in each sub-region, and screening a crowdsourcing sample with the highest normalized weight in each partition;
if no crowdsourcing sample exists in any partition, determining a sample corresponding to the maximum weight as the crowdsourcing sample of any partition in the crowdsourcing samples until the number of the screened crowdsourcing samples is equal to the number of the partitions, and obtaining the screened crowdsourcing samples.
According to the radio map construction method based on federal learning provided by the invention, the curve fitting parameter and the curve signal intensity estimation error score are calculated by using the enhanced crowdsourcing sample or the screened crowdsourcing sample and the initial training parameter, and the method comprises the following steps:
Acquiring global parameters and global scores of each sub-region from the central node, determining the global scores as original global scores, and determining a first global score of each sub-region in current user data by using the global parameters;
constructing a radio map of each sub-area by using the current user data and adopting a weighted sample-based surface fitting model, acquiring a surface training parameter of the radio map, and adopting a second global score of the surface training parameter in the current user data;
if the original global score is determined to be larger than the second global score, ending training, otherwise continuing training;
taking the first global score and the second global score as weights, and carrying out weighted fusion and averaging on the global parameter and the curved surface training parameter to obtain a third global score;
determining the highest global score in the first global score, the second global score and the third global score, and taking the curve fitting parameter corresponding to the highest global score as the latest curve fitting parameter;
and calculating the real signal intensity and the estimated signal intensity corresponding to the latest curve fitting parameter by adopting a least square method, and obtaining the curve signal intensity estimated error score by utilizing the constraint of the preset minimum value and the preset maximum value of the received signal intensity.
According to the radio map construction method based on federal learning provided by the invention, the variable length global parameter transmission protocol comprises the following steps:
adding an area index field in an uploaded or downloaded transmission message header;
and determining to upload or download the curve fitting parameters in the untrained subareas or subareas with the curve signal intensity estimation error scores smaller than a preset score threshold according to the area index field.
In a second aspect, the present invention also provides a radio map construction system based on federal learning, including:
a user determining module for determining a plurality of training user nodes of the target indoor area;
the parameter downloading module is used for dividing a plurality of subareas of the target indoor area, and the central node downloads global parameters and global results of the subareas to each training user node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node;
the local training module is used for acquiring crowdsourcing samples of each training user node, and training the crowdsourcing samples and the initial training parameters to obtain curved surface fitting parameters and curved surface signal intensity estimation error scores of all the subareas;
The parameter uploading module is used for uploading the curve fitting parameters and the curve signal intensity estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, and the central node determines to update the global parameters by utilizing the best curve fitting parameters and the best curve signal intensity estimation error scores of all the subareas so as to obtain the latest global parameters of all the subareas;
and the construction positioning module is used for constructing a radio map based on the latest global parameters, determining a target subarea by utilizing signal intensity characteristics in the radio map, and determining the position of a to-be-positioned point in the target subarea.
In a third aspect, the present invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing any one of the federal learning-based radio map construction methods described above when executing the program.
In a fourth aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a federal learning-based radio map construction method as described in any one of the above.
According to the radio map construction method and system based on federal learning, sub-region user data training is performed by using federal learning, and the central node is used for updating parameters, so that the problem that the signal intensity between indoor range regions is hopped and the change rule of the signal intensity of different regions is inconsistent is effectively solved, the overall radio map construction quality is improved, and meanwhile, the uplink and downlink communication overhead in the training process is saved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a federal learning-based radio map construction method provided by the invention;
fig. 2 is a schematic diagram of indoor space and base station distribution provided by the present invention;
FIG. 3 is one of the thermodynamic diagrams of the distribution of a radio map in indoor space provided by the present invention;
FIG. 4 is a second thermodynamic diagram of the distribution of a radio map in indoor space provided by the present invention;
FIG. 5 is a third thermodynamic diagram of the distribution of a radio map in indoor space provided by the present invention;
FIG. 6 is a fourth thermodynamic diagram of the distribution of a radio map in indoor space provided by the present invention;
FIG. 7 is a graph comparing cumulative distribution functions of positioning errors of various radio map construction methods provided by the invention;
fig. 8 is a schematic structural diagram of a federal learning-based radio map construction system provided by the present invention;
fig. 9 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Aiming at the limitations existing in the prior art when a radio map is built, the invention provides a radio map building method based on federal learning, and fig. 1 is a flow diagram of the radio map building method based on federal learning, provided by the embodiment of the invention, as shown in fig. 1, including:
Step 100: determining a plurality of training user nodes of the target indoor area;
step 200: dividing a plurality of subareas of the target indoor area, and downloading global parameters and global scores of the subareas to each training user node by a central node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node;
step 300: acquiring crowdsourcing samples of each training user node, and training to obtain surface fitting parameters and surface signal intensity estimation error scores of all subareas based on the crowdsourcing samples and the initial training parameters;
step 400: uploading the surface fitting parameters and the surface signal strength estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, wherein the central node determines to update the global parameters by utilizing the best surface fitting parameters and the best surface signal strength estimation error scores of all the subareas, and obtains the latest global parameters of all the subareas;
step 500: and constructing a radio map based on the latest global parameters, determining a target subarea in the radio map by utilizing signal intensity characteristics, and determining the position of a to-be-positioned point in the target subarea.
According to the embodiment of the invention, a radio map construction method is adopted by federal learning, the federal learning is used as a distributed machine learning paradigm, the problem of data island can be effectively solved, participants can be modeled in a combined way on the basis of not sharing data, the data island can be broken technically, the calculated amount is reduced, and the cooperation of artificial intelligence (Artificial Intelligence, AI) is realized. Therefore, in order to protect user privacy, improve training efficiency, and reduce data transmission overhead, the embodiment of the invention proposes to construct a radio map by using a distributed (federal) training method.
Firstly, training user selection is carried out, the indoor area is divided into subareas, parameters of each subarea are issued to user nodes by a central node, secondly, local training of the user nodes is carried out, for data of each user node, crowdsourcing samples are distributed to the corresponding subareas, curve fitting parameters of each subarea are trained, after one round of training is finished, the local user node uploads the best parameters and best results of each subarea obtained by training to the central node, global parameters are updated, and finally, when all rounds of training are finished, a radio map is built by the latest global parameters of each subarea, and the position of a to-be-positioned point is obtained.
It should be noted that, the crowdsourcing sample adopts the idea of the crowdsourcing method, which distributes the tedious sample measurement work to the common user, and the user collects the sample at a non-designated position, so as to avoid the process of site survey of the special person, reduce the labor cost, and ensure that the crowdsourcing sample is updated frequently so as to adapt to the change of the environment. However, the method for constructing the indoor radio map based on the crowdsourcing samples has the following problems: (1) Inaccurate crowdsourcing samples are marked, the crowdsourcing samples are usually collected by users at unspecified positions, and compared with the measurement of field survey professionals, the marking of the positions of the crowdsourcing samples has larger errors in the mode of active marking by the crowdsourcing users or the passive marking mode of track matching. (2) Unequal sample dimensions may enable more Access Point (AP) signals to be received in some location areas and fewer AP signals to be received in other locations due to limited WIFI signal propagation distances, and the number of APs received may be different even in the same location due to differences in user operation. (3) Uneven spatial distribution, due to limitations of indoor layout and uncontrolled user acquisition behavior, places where some easy to reach may have more crowded samples and places where they are not easy to reach contain fewer crowded samples, and even places where they do not. (4) Different measurement devices, due to the different mobile smartdevices used by different users, may acquire different AP dimensions, even at the same location and time, and may also differ in signal reception strength (Received Signal Strength, RSS) of the same AP.
Specifically, in the user selection stage, m users are randomly selected from the same user group each time by a method of grouping the users into user groups, and participate in the j-th round of distributed training;
in the parameter downloading stage, dividing the indoor area into P sub-areas; the central node downloads global parameters and global score of the P sub-areas to each user node as initial training parameters of the user;
in a user node local training stage, for each user's data, distributing each crowdsourcing sample of the user to the affiliated subarea, traversing the P subareas in turn, and utilizing the crowdsourcing sample of the current area of the current user to train the curve fitting parameters of the subareas and the signal intensity estimation error scores of the corresponding curves;
after the local training is completed, parameter uploading is carried out, and the local user uploads the best parameters and best scores of each sub-area obtained through training to the central node. The central node receives m groups of parameters and corresponding scores of m users, selects a group of parameters with highest fitting surface scores of each sub-area, and updates a global parameter, so that one round of distributed training is completed;
finally, entering a positioning stage, after all rounds of training are completed, constructing a radio map by setting reference points through the latest global parameter of the P sub-areas, dividing an indoor space into Q sub-areas, and determining the sub-areas to which the to-be-positioned points belong according to the signal intensity characteristics of the test sample Later, in subregion->And positioning the inner positioning point to obtain the position of the positioning point.
According to the invention, the sub-region user data training is performed by adopting federal learning, and the central node is used for updating parameters, so that the phenomenon that the signal intensity between indoor range regions is jumped and the problem that the signal intensity change rules of different regions are inconsistent are effectively solved, the overall radio map construction quality is improved, and meanwhile, the uplink and downlink communication overhead in the training process is saved.
On the basis of the above embodiment, determining a plurality of training user nodes of the target indoor area includes:
dividing all user nodes into n user groups, and determining m users in the same user group during each round of distributed training;
if the user group with the number i is determined to train in the current training round, the user group with the number i+1 is determined to train in the next training round;
and after the n user groups finish traversing training, training is restarted from the 1 st user group.
Specifically, when selecting a training user node, the embodiment of the invention divides all users into n user groups, each round of distributed training selects m users of the same user group, if the user group with the number i is selected in the current round of training, the user group with the number i+1 is selected in the next round of training, the training is continued, and when the n user groups traverse once, the training is started from the 1 st user group.
On the basis of the above embodiment, dividing the plurality of sub-areas of the target indoor area includes:
when the target area is divided into a plurality of subareas, the subareas which the crowdsourcing samples belong to can be determined directly according to the two-dimensional coordinates of the positions corresponding to the crowdsourcing samples according to the indoor building layout of the target area, and the specific mode of carrying out surface fitting on each signal source by utilizing the crowdsourcing samples contained in the subareas can refer to a patent application file with application publication number CN109059919A, and the details are not repeated; the final fitting of the signal source curved surface, namely the signal intensity fitting function of the signal source, hasForm of (1), wherein->For fitting coefficients +.>、Fitting order for polynomial, ++>For signal receiving point coordinates +.>And->Respectively->And->Fitting index of>Representing crowd-sourced sample numbers.
Based on the above embodiment, obtaining a crowd-sourced sample of each training user node includes:
acquiring acquisition position coordinates and acquisition position received signal strength vectors of each training user node, wherein the acquisition position received signal strength vectors are determined by crowdsourcing sample numbers and total signal sources;
and forming crowdsourcing samples of each training user node by the acquisition position coordinates and the acquisition position received signal strength vector.
Specifically, crowdsourcing samples in embodiments of the present inventionFrom the acquisition position coordinates->And position->RSS vector received thereat->) Constitution (S)>Representing crowd-sourced sample number,/->Representing the total number of signal sources.
Based on the above embodiment, training based on the crowdsourcing samples and the initial training parameters to obtain the curve fitting parameters and the curve signal strength estimation error scores of all the subareas, including:
determining a training sample threshold for each sub-region;
if the number of crowdsourcing samples in each sub-area is smaller than the training sample threshold, enhancing the crowdsourcing samples to obtain enhanced crowdsourcing samples;
if the number of the crowdsourcing samples in each sub-area is determined to be larger than the training sample threshold value multiplied by a preset proportion, the reliability weight of each crowdsourcing sample is obtained, and the crowdsourcing samples are screened based on the reliability weight to obtain screened crowdsourcing samples;
and calculating to obtain the curve fitting parameter and the curve signal intensity estimation error score by using the enhanced crowdsourcing sample or the screened crowdsourcing sample and the initial training parameter.
If it is determined that the number of crowdsourcing samples in each sub-region is smaller than the training sample threshold, enhancing the crowdsourcing samples to obtain enhanced crowdsourcing samples, including:
Obtaining a crowded-package sample average value, and determining to construct a random disturbance virtual sample, wherein the value range of the random disturbance virtual sample is between 0 and 1;
and adding the random disturbance virtual sample with the acquisition position coordinates and the acquisition position received signal strength vector in the crowdsourcing sample respectively to obtain the enhanced crowdsourcing sample.
If it is determined that the number of crowdsourcing samples in each sub-area is greater than the training sample threshold multiplied by a preset proportion, calculating a reliability weight of each crowdsourcing sample by using a cross-domain clustering and crossing algorithm, screening the crowdsourcing samples based on the reliability weight, and obtaining screened crowdsourcing samples, wherein the method comprises the following steps:
calculating the reliability weight of each crowdsourcing sample by using a cross-domain clustering and crossing algorithm;
determining a partition in each sub-region, and screening a crowdsourcing sample with the highest normalized weight in each partition;
if no crowdsourcing sample exists in any partition, determining a sample corresponding to the maximum weight as the crowdsourcing sample of any partition in the crowdsourcing samples until the number of the screened crowdsourcing samples is equal to the number of the partitions, and obtaining the screened crowdsourcing samples.
The step of calculating the curve fitting parameter and the curve signal strength estimation error score by using the enhanced crowdsourcing sample or the screened crowdsourcing sample and the initial training parameter, including:
Acquiring global parameters and global scores of each sub-region from the central node, determining the global scores as original global scores, and determining a first global score of each sub-region in current user data by using the global parameters;
constructing a radio map of each sub-area by using the current user data and adopting a weighted sample-based surface fitting model, acquiring a surface training parameter of the radio map, and adopting a second global score of the surface training parameter in the current user data;
if the original global score is determined to be larger than the second global score, ending training, otherwise continuing training;
taking the first global score and the second global score as weights, and carrying out weighted fusion and averaging on the global parameter and the curved surface training parameter to obtain a third global score;
determining the highest global score in the first global score, the second global score and the third global score, and taking the curve fitting parameter corresponding to the highest global score as the latest curve fitting parameter;
and calculating the real signal intensity and the estimated signal intensity corresponding to the latest curve fitting parameter by adopting a least square method, and obtaining the curve signal intensity estimated error score by utilizing the constraint of the preset minimum value and the preset maximum value of the received signal intensity.
Specifically, in the local training stage, the embodiment of the invention performs different processing according to different conditions of the training sample:
when the sub-area is subjected to surface fitting and the samples of a certain area of the user are not enough to fit the surface, namely the number of samples in the sub-areaAt the time, a virtual sample is constructed by adding a random disturbance between (0, 1) to the average value of the existing samples until the number of samples in the subarea is +.>Wherein->And representing the minimum number of samples required by the subarea surface fitting algorithm for subarea surface fitting. The specific generation mode is as follows:
wherein, the method comprises the following steps of) Represents the position of the i-th sample, +.>Representing the RSS value of the j-th AP received by the i-th sample, the random function is used to generate a random number in the range (0, 1).
Surface fitting of sub-regions to the number of samples of a user's regionIn this case, the reliability weight of each sample is calculated by the cross-domain clustering and crossing algorithm, so as to prepare for the subsequent sample screening, and the specific manner of calculating the reliability weight of each sample by using the crowdsourcing samples contained in the sub-region can refer to the patent application document with application publication number CN109059919a, which will not be described herein.
Further, if the number of samples in the sub-region And screening the samples based on the reliability weight of each sample in the subarea, and particularly, screening the samples by using a subarea screening algorithm. Carrying out finer region division on each sub-region, and calculating the partition to which each sample belongs; selecting one sample with the highest normalized weight from each partition, and taking the sample as a representative sample of the partition; if no samples exist in some partitions, selecting samples with large weights from the global samples for supplementation until the total number k of the selected samples is equal to the number M of the partitions. It should be noted that, the sample screening performed by using the partition screening algorithm is merely a preferred implementation manner of the embodiment of the present invention, and the present invention is not limited in any way.
And (3) establishing a polynomial model of signal intensity in space distribution by using reference point fingerprints of each AP for samples of each sub-region after region division, enabling fitting error square sum between model fitting signal intensity and real signal intensity of all reference points in the region to be minimum according to Least square criterion (LS), and solving fitting parameters of the model, thereby constructing an RSS distribution function of the sub-region.
Comparing the front and back sets of surface fitting parameters of the sub-region with the signal strength estimation error scores of the corresponding surfaces as shown in fig. 2, including:
(1) The global parameter and global score for the sub-region are obtained from the central node and the global score at this time is denoted as socre_0. The score of the surface is evaluated on the current user data using the global parameter, denoted score_1.
(2) And constructing a radio map of the area by using the data of the current user and adopting a curve fitting model based on weighted samples to obtain curve parameters trained parameter, and calculating a score score_2 of the fitting curve on the data of the current user.
(3) Judging the signal intensity estimation error score score_0 of the initial curved surface of the sub-region and the signal intensity estimation error score score_2 of the latest curved surface of the sub-region: if score_0> score_2, the curve effect of previous round fitting is considered to be better, the round does not need to be trained any more, and training of the area of the user is finished; if score_0 is less than or equal to score_2, the curve effect of the current round fitting is considered to be better, and the subsequent steps are continued.
(4) Averaging global parameters and trained parameter according to the weights of score_1 and score_2 to obtain a fusion parameter merged parameter; and tests the score of the fusion parameter on the user data, noted score_3.
(5) And comparing the signal intensity estimation error scores score_1, score_2 and score_3 of the curved surfaces corresponding to the three groups of parameters, and selecting a group of parameters with the highest score as a result of the training and uploading the result to the central node.
Further, the embodiment of the present invention calculates a global score for the signal strength estimation error of the fitted surface by:
wherein,,representing the i-th dimension true signal strength of the sample, < >>Representing the signal strength estimated by the ith dimension algorithm of the sample, if +.>No signal is received in a certain dimension->And->Reducing this dimension and then participating in the calculation. The smaller the combined error between the true signal strengths of all samples and the algorithmic estimated signal strengths, the higher the global score.
In addition, when the user locally trains and calculates a global score for fitting the signal intensity estimation error of the curved surface, if a certain subarea is not fitted, namely, when the global parameter of the subarea is an initial value, the RSS predicted value of a sample of the subarea is set to be a minimum value MIN= -120dBm, and MIN is smaller than the minimum value of the receivable signal intensity; and if the predicted value of the sample for the RSS from a certain signal source is larger than the maximum value MAX, setting the RSS value of the sample from the signal source to be the maximum value MAX= -60dBm; if the predicted value of the sample for the RSS from a certain signal source is smaller than the minimum value MIN, setting the RSS value of the sample from the signal source to be the minimum value MIN= -120dBm; MAX is greater than the maximum value of the receivable signal strength.
According to the invention, a continuous signal curved surface, namely a signal intensity fitting function, is constructed in each sub-area by utilizing a partition curved surface fitting technology, and compared with a traditional off-line fingerprint map construction scheme based on site survey, the method can obtain richer fingerprint information by only storing a small amount of relevant fitting coefficients; and the weighted samples are used for fitting the signal curved surface to the sub-region, so that the influence of crowdsourcing sample labeling errors and indoor layout on the fitted curved surface can be effectively reduced, and the accuracy of radio map construction is improved.
On the basis of the above embodiment, the variable length global parameter transmission protocol includes:
adding an area index field in an uploaded or downloaded transmission message header;
and determining to upload or download the curve fitting parameters in the untrained subareas or subareas with the curve signal intensity estimation error scores smaller than a preset score threshold according to the area index field.
It can be understood that the embodiment of the invention adopts a variable-length global parameter transmission protocol in the uploading and downloading processes of the parameters.
And a field index of 'region index' is newly added in the message header of parameter downloading and uploading transmission and is used for identifying which sub-region of the P sub-regions corresponds to the surface fitting parameter. And in the subsequent uploading and downloading of parameters, only the curved surface fitting parameters of the subareas with global scores smaller than the threshold value, which are not trained or fitted in the earlier stage, are uploaded and downloaded so as to save the uplink and downlink communication overhead.
In the embodiment of the invention, a continuous signal curved surface is constructed in each sub-area by using a partition curved surface fitting technology, namely a signal strength fitting function, and uplink and downlink parameter communication in a distributed training process is completed by using a variable-length global parameter transmission protocol and adding a new 'area index' field index in a message header of parameter downloading and uploading transmission, and compared with a fixed-length global parameter transmission protocol, only meaningful parameters are uploaded or downloaded, so that communication expenditure is effectively saved.
On the basis of the above embodiments, the embodiments of the present invention further describe the solution of the present invention in conjunction with an application example in a specific scenario.
Taking the indoor target area shown in fig. 2 as an example, the signal intensity measurement is performed by using an intelligent device, n=4 signal sources can be measured in total in an indoor space scene of about 400 m×400 m, and the signal intensity of at least one signal source can be received at any position in the scene. The map in fig. 2 is uniformly divided into 25 sub-areas, i.e. each sub-area has a density of 80 m by 80 m. And distributing the data acquisition task to n=9 groups of users in the system, wherein each group of users comprises 10 users, each group of users moves in different areas to obtain training samples in the areas, the moving range of each user is 240 m by 240 m, and the number of data samples of each user is about 4000. The data samples of each user are randomly divided into training sets and verification sets according to the proportion of 99:1, and the verification sets of all users form a test set. The training set is used for training the partitioned surface fitting model, and the testing set is used for testing the prediction performance of the radio map obtained through training.
The embodiment of the invention uses the signal strength estimation error score and the model uplink and downlink overhead to perform performance evaluation. The method for constructing the radio map based on the federal learning partition surface fitting is compared with the method for constructing the radio map based on the multi-layer perceptron (Multilayer Perceptron, MLP), the method for constructing the radio map without adopting a sample weighting algorithm and the method for constructing the radio map without adopting a variable length message protocol, and the method comprises the following steps:
a radio map construction method based on a multi-layer perceptron (MLP) comprises the following steps: a radio map is constructed by adopting a neural network consisting of 4 fully connected layers, wherein the input and output dimensions of the 4 fully connected layers are (2,256), (256,1024), (1024,256) and (256,4) respectively.
The radio map construction method without adopting the sample weighting algorithm is different from the invention in that: and (3) directly carrying out polynomial surface fitting on all unweighted samples in the subareas without adopting a sample weighting algorithm to construct the radio map.
The radio map construction method without adopting the variable length message protocol is different from the invention in that: and when the parameters are uploaded and downloaded, the parameters are transmitted by using a fixed message length parameter transmission protocol.
As shown in table 1, the radio map constructed for each scheme has a signal strength estimation error loss, a signal strength estimation error score, an uplink cost and a downlink cost. In all the comparison schemes, the loss, the uplink cost and the downlink cost of the radio map construction method are lower than those of the radio map construction method of the MLP, and the radio map construction method with score higher than that of the MLP is proved to effectively improve the accuracy of the radio map construction and reduce the uplink and downlink communication cost. In addition, the uplink cost and the downlink cost of the algorithm are obviously reduced after the variable length message protocol is added, namely the communication efficiency is further improved.
TABLE 1
As shown in figures 3 to 6, according to the thermodynamic diagram of indoor space distribution of the radio map constructed by the method, the phenomenon that RSS possibly jumps in adjacent areas due to walls, indoor layout and the like, the radio map constructed by the method basically accords with the propagation rule of the RSS in a complex indoor environment, wherein the radio map constructed by the method corresponds to an RSS [0] signal source in figure 3, an RSS [1] signal source in figure 4, an RSS [2] signal source in figure 5 and an RSS [3] signal source in figure 6.
As also shown in Table 2, the positioning errors of the MLP and the two radio map construction methods of the present invention in the current positioning environment are compared, including a multi-layer perceptron (Neural Network) MLP (NN), a multi-layer perceptron (k Nearest Neighbor) MLP (KNN), a partitioned surface fit (NN), and a partitioned surface fit (KNN). The whole indoor space is uniformly divided into q=4 sub-areas, and then the sample points of the test set are subjected to area fingerprint positioning. The MLP method is poor in positioning effect by constructing a radio map only through one group of neural network parameters under such a large positioning scene, and the proposed method for regional surface fitting has a group of own surface fitting parameters for each sub-region, so that the influence of inconsistent RSS distribution rules among different regions on the accuracy of the constructed radio map can be reduced, and therefore, the MLP method has lower positioning error, and the method for constructing the radio map by regional surface fitting has good robustness.
TABLE 2
As shown in fig. 7, the cumulative distribution function (Cumulative Distribution Function, CDF) of the positioning error in the current positioning environment is compared with the MLP and the two radio map construction methods of the present invention. As shown in fig. 7, the CDF curve of the positioning error of the radio map constructed by the present invention is located above the CDF curve of the positioning error of the radio map constructed by the MLP method, and the rising trend of the curve is far faster than that of the CDF curve of the MLP method, which indicates that the radio map construction method of the partition surface fitting has good accuracy.
The federal learning-based radio map construction system provided by the invention is described below, and the federal learning-based radio map construction system described below and the federal learning-based radio map construction method described above can be referred to correspondingly with each other.
Fig. 8 is a schematic structural diagram of a radio map construction system based on federal learning according to an embodiment of the present invention, as shown in fig. 8, including: a user determination module 81, a parameter download module 82, a local training module 83, a parameter upload module 84, and a build location module 85, wherein:
the user determining module 81 is configured to determine a plurality of training user nodes of the target indoor area; the parameter downloading module 82 is configured to divide a plurality of sub-areas of the target indoor area, and the central node downloads global parameters and global scores of the plurality of sub-areas to each training user node based on a variable-length global parameter transmission protocol, as initial training parameters of each training user node; the local training module 83 is configured to obtain crowd-sourced samples of each training user node, and train to obtain surface fitting parameters and surface signal strength estimation error scores of all sub-areas based on the crowd-sourced samples and the initial training parameters; the parameter uploading module 84 is configured to upload, by each training user node, the surface fitting parameter and the surface signal strength estimation error score to the central node based on the variable length global parameter transmission protocol, where the central node determines to update the global parameter with the best surface fitting parameter and the best surface signal strength estimation error score for all sub-regions, and obtains the latest global parameter for all sub-regions; the build localization module 85 is configured to build a radio map based on the latest global parameters, in which a target sub-area is determined using the signal strength characteristics, in which the location of the site to be localized is determined.
Fig. 9 illustrates a physical schematic diagram of an electronic device, as shown in fig. 9, which may include: processor 910, communication interface (Communications Interface), memory 930, and communication bus 940, wherein processor 910, communication interface 920, and memory 930 communicate with each other via communication bus 940. The processor 910 may invoke logic instructions in the memory 930 to perform a federal learning-based radio map construction method comprising: determining a plurality of training user nodes of the target indoor area; dividing a plurality of subareas of the target indoor area, and downloading global parameters and global scores of the subareas to each training user node by a central node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node; acquiring crowdsourcing samples of each training user node, and training to obtain surface fitting parameters and surface signal intensity estimation error scores of all subareas based on the crowdsourcing samples and the initial training parameters; uploading the surface fitting parameters and the surface signal strength estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, wherein the central node determines to update the global parameters by utilizing the best surface fitting parameters and the best surface signal strength estimation error scores of all the subareas, and obtains the latest global parameters of all the subareas; and constructing a radio map based on the latest global parameters, determining a target subarea in the radio map by utilizing signal intensity characteristics, and determining the position of a to-be-positioned point in the target subarea.
Further, the logic instructions in the memory 930 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the federal learning-based radio map construction method provided by the above methods, the method comprising: determining a plurality of training user nodes of the target indoor area; dividing a plurality of subareas of the target indoor area, and downloading global parameters and global scores of the subareas to each training user node by a central node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node; acquiring crowdsourcing samples of each training user node, and training to obtain surface fitting parameters and surface signal intensity estimation error scores of all subareas based on the crowdsourcing samples and the initial training parameters; uploading the surface fitting parameters and the surface signal strength estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, wherein the central node determines to update the global parameters by utilizing the best surface fitting parameters and the best surface signal strength estimation error scores of all the subareas, and obtains the latest global parameters of all the subareas; and constructing a radio map based on the latest global parameters, determining a target subarea in the radio map by utilizing signal intensity characteristics, and determining the position of a to-be-positioned point in the target subarea.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the 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 scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A federal learning-based radio map construction method, comprising:
determining a plurality of training user nodes of the target indoor area;
dividing a plurality of subareas of the target indoor area, and downloading global parameters and global scores of the subareas to each training user node by a central node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node;
acquiring crowdsourcing samples of each training user node, and training to obtain surface fitting parameters and surface signal intensity estimation error scores of all subareas based on the crowdsourcing samples and the initial training parameters;
Uploading the surface fitting parameters and the surface signal strength estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, wherein the central node determines to update the global parameters by utilizing the best surface fitting parameters and the best surface signal strength estimation error scores of all the subareas, and obtains the latest global parameters of all the subareas;
and constructing a radio map based on the latest global parameters, determining a target subarea in the radio map by utilizing signal intensity characteristics, and determining the position of a to-be-positioned point in the target subarea.
2. The federal learning-based radio map construction method according to claim 1, wherein determining a plurality of training user nodes for a target indoor area comprises:
dividing all user nodes into n user groups, and determining m users in the same user group during each round of distributed training;
if the user group with the number i is determined to train in the current training round, the user group with the number i+1 is determined to train in the next training round;
and after the n user groups finish traversing training, training is restarted from the 1 st user group.
3. The federal learning-based radio map construction method according to claim 1, wherein obtaining crowd-sourced samples for each training user node comprises:
acquiring acquisition position coordinates and acquisition position received signal strength vectors of each training user node, wherein the acquisition position received signal strength vectors are determined by crowdsourcing sample numbers and total signal sources;
and forming crowdsourcing samples of each training user node by the acquisition position coordinates and the acquisition position received signal strength vector.
4. A federal learning-based radio map construction method according to claim 3, wherein training to obtain the surface fitting parameters and the surface signal strength estimation error scores for all sub-regions based on the crowd-sourced samples and the initial training parameters comprises:
determining a training sample threshold for each sub-region;
if the number of crowdsourcing samples in each sub-area is smaller than the training sample threshold, enhancing the crowdsourcing samples to obtain enhanced crowdsourcing samples;
if the number of the crowdsourcing samples in each sub-area is determined to be larger than the training sample threshold value multiplied by a preset proportion, the reliability weight of each crowdsourcing sample is obtained, and the crowdsourcing samples are screened based on the reliability weight to obtain screened crowdsourcing samples;
And calculating to obtain the curve fitting parameter and the curve signal intensity estimation error score by using the enhanced crowdsourcing sample or the screened crowdsourcing sample and the initial training parameter.
5. The federal learning-based radio map construction method of claim 4, wherein if it is determined that the number of crowdsourcing samples for each sub-region is less than the training sample threshold, performing enhancement processing on the crowdsourcing samples to obtain enhanced crowdsourcing samples, comprising:
obtaining a crowded-package sample average value, and determining to construct a random disturbance virtual sample, wherein the value range of the random disturbance virtual sample is between 0 and 1;
and adding the random disturbance virtual sample with the acquisition position coordinates and the acquisition position received signal strength vector in the crowdsourcing sample respectively to obtain the enhanced crowdsourcing sample.
6. The federal learning-based radio map construction method of claim 4, wherein if it is determined that the number of crowdsourcing samples for each sub-region is greater than the training sample threshold multiplied by a preset ratio, then calculating a reliability weight for each crowdsourcing sample using a cross-domain clustering intersection algorithm, screening the crowdsourcing samples based on the reliability weights, and obtaining screened crowdsourcing samples, comprising:
Calculating the reliability weight of each crowdsourcing sample by using a cross-domain clustering and crossing algorithm;
determining a partition in each sub-region, and screening a crowdsourcing sample with the highest normalized weight in each partition;
if no crowdsourcing sample exists in any partition, determining a sample corresponding to the maximum weight as the crowdsourcing sample of any partition in the crowdsourcing samples until the number of the screened crowdsourcing samples is equal to the number of the partitions, and obtaining the screened crowdsourcing samples.
7. The federal learning-based radio map construction method according to claim 4, wherein calculating the surface fitting parameter and the surface signal strength estimation error score using the enhanced crowd-sourced samples or the screened crowd-sourced samples, and the initial training parameter, comprises:
acquiring global parameters and global scores of each sub-region from the central node, determining the global scores as original global scores, and determining a first global score of each sub-region in current user data by using the global parameters;
constructing a radio map of each sub-area by using the current user data and adopting a weighted sample-based surface fitting model, acquiring a surface training parameter of the radio map, and adopting a second global score of the surface training parameter in the current user data;
If the original global score is determined to be larger than the second global score, ending training, otherwise continuing training;
taking the first global score and the second global score as weights, and carrying out weighted fusion and averaging on the global parameter and the curved surface training parameter to obtain a third global score;
determining the highest global score in the first global score, the second global score and the third global score, and taking the curve fitting parameter corresponding to the highest global score as the latest curve fitting parameter;
and calculating the real signal intensity and the estimated signal intensity corresponding to the latest curve fitting parameter by adopting a least square method, and obtaining the curve signal intensity estimated error score by utilizing the constraint of the preset minimum value and the preset maximum value of the received signal intensity.
8. The federally learned based radio map construction method according to claim 1, wherein the variable length global parameter transmission protocol comprises:
adding an area index field in an uploaded or downloaded transmission message header;
and determining to upload or download the curve fitting parameters in the untrained subareas or subareas with the curve signal intensity estimation error scores smaller than a preset score threshold according to the area index field.
9. A federal learning-based radio map construction system, comprising:
a user determining module for determining a plurality of training user nodes of the target indoor area;
the parameter downloading module is used for dividing a plurality of subareas of the target indoor area, and the central node downloads global parameters and global results of the subareas to each training user node based on a variable-length global parameter transmission protocol to serve as initial training parameters of each training user node;
the local training module is used for acquiring crowdsourcing samples of each training user node, and training the crowdsourcing samples and the initial training parameters to obtain curved surface fitting parameters and curved surface signal intensity estimation error scores of all the subareas;
the parameter uploading module is used for uploading the curve fitting parameters and the curve signal intensity estimation error scores to the central node by each training user node based on the variable-length global parameter transmission protocol, and the central node determines to update the global parameters by utilizing the best curve fitting parameters and the best curve signal intensity estimation error scores of all the subareas so as to obtain the latest global parameters of all the subareas;
And the construction positioning module is used for constructing a radio map based on the latest global parameters, determining a target subarea by utilizing signal intensity characteristics in the radio map, and determining the position of a to-be-positioned point in the target subarea.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the federal learning-based radio map construction method according to any one of claims 1 to 8 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310626603.1A CN116405880B (en) | 2023-05-31 | 2023-05-31 | Radio map construction method and system based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310626603.1A CN116405880B (en) | 2023-05-31 | 2023-05-31 | Radio map construction method and system based on federal learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116405880A CN116405880A (en) | 2023-07-07 |
| CN116405880B true CN116405880B (en) | 2023-09-12 |
Family
ID=87014503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310626603.1A Active CN116405880B (en) | 2023-05-31 | 2023-05-31 | Radio map construction method and system based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116405880B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009082717A2 (en) * | 2007-11-19 | 2009-07-02 | Ezmcom, Inc. | A method for authenticating a communication channel between a client and a server |
| CA2894863A1 (en) * | 2015-06-22 | 2016-12-22 | The Governing Council Of The University Of Toronto | Indoor localization using crowdsourced data |
| CN108449709A (en) * | 2018-03-16 | 2018-08-24 | 华中科技大学 | A kind of indoor radio map constructing method based on crowdsourcing sample |
| CN108919177A (en) * | 2018-07-16 | 2018-11-30 | 华中科技大学 | A kind of positioning map construction method based on virtual information source estimation and trajectory corrector |
| WO2019211134A1 (en) * | 2018-05-02 | 2019-11-07 | Telefonaktiebolaget Lm Ericsson (Publ) | First network node, third network node, and methods performed thereby, for handling a performance of a radio access network |
| CN111474516A (en) * | 2020-04-17 | 2020-07-31 | 华中科技大学 | Multi-level indoor positioning method and system based on crowdsourcing sample surface fitting |
| CN113207089A (en) * | 2021-05-08 | 2021-08-03 | 苏州云享阁智能科技有限公司 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
| KR102333893B1 (en) * | 2020-07-02 | 2021-12-01 | 한경대학교 산학협력단 | Method and Apparatus for Received Signal Strength Indicator Database Update based on Crowdsourced Data for Maintenance of Indoor Localization System |
| CN113919483A (en) * | 2021-09-23 | 2022-01-11 | 南昌大学 | Method and system for constructing and positioning radio map in wireless communication network |
| WO2022120997A1 (en) * | 2020-12-10 | 2022-06-16 | 中国科学院深圳先进技术研究院 | Distributed slam system and learning method therefor |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8265590B2 (en) * | 2009-10-02 | 2012-09-11 | At&T Mobility Ii Llc | Providing information pertaining to usage of a mobile wireless communications device |
-
2023
- 2023-05-31 CN CN202310626603.1A patent/CN116405880B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009082717A2 (en) * | 2007-11-19 | 2009-07-02 | Ezmcom, Inc. | A method for authenticating a communication channel between a client and a server |
| CA2894863A1 (en) * | 2015-06-22 | 2016-12-22 | The Governing Council Of The University Of Toronto | Indoor localization using crowdsourced data |
| CN108449709A (en) * | 2018-03-16 | 2018-08-24 | 华中科技大学 | A kind of indoor radio map constructing method based on crowdsourcing sample |
| WO2019211134A1 (en) * | 2018-05-02 | 2019-11-07 | Telefonaktiebolaget Lm Ericsson (Publ) | First network node, third network node, and methods performed thereby, for handling a performance of a radio access network |
| CN108919177A (en) * | 2018-07-16 | 2018-11-30 | 华中科技大学 | A kind of positioning map construction method based on virtual information source estimation and trajectory corrector |
| CN111474516A (en) * | 2020-04-17 | 2020-07-31 | 华中科技大学 | Multi-level indoor positioning method and system based on crowdsourcing sample surface fitting |
| KR102333893B1 (en) * | 2020-07-02 | 2021-12-01 | 한경대학교 산학협력단 | Method and Apparatus for Received Signal Strength Indicator Database Update based on Crowdsourced Data for Maintenance of Indoor Localization System |
| WO2022120997A1 (en) * | 2020-12-10 | 2022-06-16 | 中国科学院深圳先进技术研究院 | Distributed slam system and learning method therefor |
| CN113207089A (en) * | 2021-05-08 | 2021-08-03 | 苏州云享阁智能科技有限公司 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
| CN113919483A (en) * | 2021-09-23 | 2022-01-11 | 南昌大学 | Method and system for constructing and positioning radio map in wireless communication network |
Non-Patent Citations (1)
| Title |
|---|
| 一种基于支持向量机的认知无线电频谱感知方案;陈思吉;王欣;申滨;;重庆邮电大学学报(自然科学版)(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116405880A (en) | 2023-07-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hu et al. | Reinforcement learning for decentralized trajectory design in cellular UAV networks with sense-and-send protocol | |
| CN112886996B (en) | Signal receiving method, user equipment, electronic equipment and computer storage medium | |
| KR102030128B1 (en) | Resource allocating method for wireless backhaul network and apparatus based on machine learning | |
| JP7279856B2 (en) | Method and apparatus | |
| Zhang et al. | Multi-agent deep reinforcement learning-based cooperative spectrum sensing with upper confidence bound exploration | |
| CN107113764B (en) | Method and device for improving positioning performance of artificial neural network | |
| Krijestorac et al. | Spatial signal strength prediction using 3D maps and deep learning | |
| CN105372628A (en) | Wi-Fi-based indoor positioning navigation method | |
| Redondi | Radio map interpolation using graph signal processing | |
| CN111148030A (en) | Fingerprint database updating method and device, server and storage medium | |
| Hashimoto et al. | SICNN: Spatial interpolation with convolutional neural networks for radio environment mapping | |
| Thrane et al. | Drive test minimization using deep learning with Bayesian approximation | |
| Podevijn et al. | Performance comparison of RSS algorithms for indoor localization in large open environments | |
| Terán et al. | Wlan-ble based indoor positioning system using machine learning cloud services | |
| CN111698695A (en) | LTE fingerprint type positioning method based on neural network | |
| CN103369670A (en) | Improved DV-hop (distance vector-hop) location method based on hop count optimization | |
| CN114710747A (en) | Indoor node-based layout method and device for communication fusion system | |
| CN110769499A (en) | Fingerprint positioning method based on dual-radio-frequency mixed RSSI | |
| CN109996171A (en) | Heredity-TABU search optimization Amorphous localization method for wireless sensor network | |
| CN116405880B (en) | Radio map construction method and system based on federal learning | |
| Cheerla et al. | RSS based Wi-Fi positioning method using multi layer neural networks | |
| Chen et al. | DeepMetricFi: Improving Wi-Fi fingerprinting localization by deep metric learning | |
| Espindola et al. | A deep autoencoder and RNN model for indoor localization with variable propagation loss | |
| Zheng et al. | RSS-based indoor passive localization using clustering and filtering in a LTE network | |
| Evmorfos et al. | Deep actor-critic for continuous 3D motion control in mobile relay beamforming networks |
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 |