CN110264154A - A kind of crowdsourcing signal map constructing method based on self-encoding encoder - Google Patents
A kind of crowdsourcing signal map constructing method based on self-encoding encoder Download PDFInfo
- Publication number
- CN110264154A CN110264154A CN201910454208.3A CN201910454208A CN110264154A CN 110264154 A CN110264154 A CN 110264154A CN 201910454208 A CN201910454208 A CN 201910454208A CN 110264154 A CN110264154 A CN 110264154A
- Authority
- CN
- China
- Prior art keywords
- signal
- signal map
- crowdsourcing
- self
- encoding encoder
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/101—Collaborative creation, e.g. joint development of products or services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Human Resources & Organizations (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Remote Sensing (AREA)
- General Engineering & Computer Science (AREA)
- Instructional Devices (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of crowdsourcing signal map constructing method based on self-encoding encoder, includes the following steps: (1) off-line training step: with the imperfect historical signal map being collected under crowdsourcing mode training pattern off-line manner;(2) online phase of regeneration: the model completed using off-line training is carried out missing values to the imperfect historical signal being collected into and infers to rebuild complete signal map.The present invention does not need professional's in-site measurement compared with traditional signal map collection mode;And compared with common crowdsourcing mode, which does not need numerous crowdsourcing participants;And compared with missing values estimating method before, which has better reconstruction precision, and adaptable, precision is high.
Description
Technical field
The present invention relates to crowdsourcing signal map constructing method, in particular to a kind of crowdsourcing signal map based on self-encoding encoder
Construction method.
Background technique
The signal map being made of the signal strength of different location is in website spectrum monitoring, location based service (LBS),
It is played an important role in network struction and the network optimization.For example, signal map can be used for the indoor location based on fingerprint with
Better position service is provided, and signal map can be used to understand current Network status to select suitable position in user
Experience better mobile network service.Signal map, which can be used, in ISP carrys out awareness network situation and optimizes the network architecture to provide
Better service coverage.Especially under the prospect of 5G network, signal map can also be used for analyzing current 4G network with
Instruct the deployment of 5G network.Although its importance, it is time-consuming and expense that signal map is constructed by professional's handheld device
Power.In order to solve this problem, Publication No. CN106157342A Chinese patent " automatic calibration method of signal map and
System " proposes that, using mobile signal receiver collecting signal building signal map, this method effectively reduces manpower but cost
Higher and moving range is smaller.With the development of electronic technology, mobile device is increasingly becoming the indispensable tool of people, mobile simultaneously
Equipment is also equipped with many sensors, this, which makes crowdsourcing mode construct signal map, becomes possibility.However crowdsourcing constructs signal
Map generally requires a large amount of participants, but user is more passive to this performance, this causes participant less.It is asked to solve this
Topic, the Chinese patent " Wi-Fi signal map construction device and method " of Publication No. CN104380294A propose that building user is normal
Database and implicit collecting signal in place, this method can participant a small amount of with effective solution the problem of, but construct place
Database is related to privacy of user while needing frequent updating database.And (the Steering Crowdsourced Signal of document 1
Map Construction via Bayesian Compressive Sensing) it proposes by Bayes's compressed sensing to not
Complete signal map carries out inferring to obtain complete signal map, but this method not only calculates higher cost but also to original
The structure of beginning signal map has also require when having to seek common ground to obtain before primary complete signal map, therefore its applicability by
Certain limitation is arrived.
Summary of the invention
Goal of the invention: overcome under the scene for constructing signal map in a manner of crowdsourcing it is an object of the present invention to provide a kind of because collecting
To signal map imperfect the problem of not being available so as to cause crowdsourcing signal map the crowdsourcing signal based on self-encoding encoder
Map constructing method, this method is adaptable, and precision is high.
Technical solution: the present invention provides a kind of crowdsourcing signal map constructing method based on self-encoding encoder, including walks as follows
It is rapid:
(1) it off-line training step: is instructed off-line manner with the imperfect historical signal map being collected under crowdsourcing mode
Practice model;
(2) online phase of regeneration: the imperfect historical signal that the model completed using off-line training arrives current collection into
Row missing values infer to rebuild complete signal map.
Further, the off-line training of the step (1) includes the following steps:
(1) according to the range of collected signal map carry out geographic grid division, and will be in same time interval on
The history crowdsourcing signal of biography is divided in different grids by GPS coordinate and then forms the imperfect signal ground under multiple different moments
Figure;
(2) by the signal map under different moments it is open and flat be one-dimensional vector, and form the behavior moment, be classified as going through for grid number
History signal map is input in self-encoding encoder and is trained then using the historical signal map as input;
(3) network structure, parameter and the smallest model of test error of self-encoding encoder are selected by cross validation method.
The online reconstruction of the step (2) includes the following steps:
(1) signal in different grids is calculated for the importance of the signal map accuracy of reconstruction and is ranked up;
(2) according to the size of estimated cost, the quantity of crowdsourcing signal needed for being arranged simultaneously formulates corresponding incentive mechanism, preferentially
It selects grid of high importance to be collected signal and uploads, to form incomplete signal map;
(3) using the imperfect signal map being collected into as input, it is entered into the self-encoding encoder of off-line training completion
It is finely adjusted in model, obtains complete signal map;
(4) its reconstruction precision calculated using cross validation method to the signal being collected into, and with pre-set threshold value
It is compared, if being unsatisfactory for threshold requirement, repeats the above process until reaching threshold requirement.
Further, the calculation method of step (1) importance includes the following steps:
(1) input for the self-encoding encoder for completing the preset value of missing values in signal map as off-line training,
Export the signal map as default;
(2) in the case where no crowdsourcing signal, the average value in historical signal is compared with default signal map, it is poor
Importance of the off course degree as next round collecting signal;
(3) the case where collecting a number of signal, the more adjacent signal map rebuild twice, and its difference degree is made
For the importance of next round collecting signal.
Further, in the step (3) when only collecting the signal of a wheel, then compared with default signal map.
Further, the network structure is the hidden layer number of plies and every layer of neuron number.
Further, the incentive mechanism is game mechanism or cash mechanism.
The present invention makes full use of existing a large amount of incomplete historical signal maps under crowdsourcing mode and utilizes self-encoding encoder
Extract temporal characteristics therein instructs the collection process of this signal map to infer in turn effectively to carry out missing values whereby
Rebuild complete signal map.Method meaning of the invention is signal map constructing method different from the past, and this method is borrowed
Self-encoding encoder is helped to learn the inherent law being present in a large amount of history incompleteness signal maps, the incomplete signal map obtain to this
Missing values are carried out to infer to obtain complete signal map.So crowdsourcing signal map constructing method of the present invention does not need it
Preceding complete signal map does not need a large amount of crowdsourcing participants simultaneously.
The utility model has the advantages that the method for the present invention compared with traditional signal map collection mode, does not need the survey of professional scene
Amount;And compared with common crowdsourcing mode, which does not need numerous crowdsourcing participants;And with missing values estimating method before
It compares, which has better reconstruction precision.
Detailed description of the invention
Fig. 1 is system framework figure of the invention;
Fig. 2 is flow chart of the invention;
Fig. 3 is self-encoding encoder structure chart of the invention.
Specific embodiment
As shown in Fig. 1, the present embodiment includes two stages, off-line training step and online phase of regeneration altogether.
1, off-line training step is divided into following steps (attached drawing 2):
(1) determine that reasonable geographic grid divides according to the range of collected signal map, in the present embodiment using imitative
True data, simulation space size are 24 × 20m2, grid division size is 20 × 20cm2, 120 × 100 grids are formed altogether, so
Emulation signal is divided into different grids by coordinate range afterwards, the signal map that symbiosis is 120 × 100 at 5000 Zhang great little,
Grid Signal for the imperfection for simulating historical signal map, therefore every signal map random selection 50% is set as -150
(the signal default value not being collected into, value selection are less than the maximum integer of signal map minimum value), therefore final symbiosis
At 5000 × 120 × 100 dimension historical signal maps;
(2) it is open and flat to tie up imperfect historical signal maps by 5000 × 120 × 100 is 5000 × 12000 dimensional vectors, forms row
For the moment, it is classified as the historical signal map matrix of grid number.Then using the matrix as input, it is input to self-encoding encoder (attached drawing
3) it is trained in, self-encoding encoder structure such as attached drawing 3, wherein black circle represents the signal not being collected into;
Self-encoding encoder is a kind of unsupervised neural network model, and it includes two parts: encoder, which may learn, to be deposited
The hidden feature being in input data;Decoder reconstructs original input data using the hidden feature learnt.In this reality
It applies in example, encoder is made of a monolayer neural networks, i.e., by input layer (imperfect signal graph) and single layer hidden layer
Composition;What decoder was equally also made of a monolayer neural networks, i.e., complete signal (is rebuild by single layer hidden layer and output layer
Figure) composition.Wherein the number of neuron should be not less than the quantity of known element in input data in hidden layer, can be by cross validation
Method is determined.
(3) nearest self-encoding encoder model, that is, the smallest model of test error is selected using the cross validation method of 80-20,
1 hidden layer of middle setting, then loss function are as follows:
Wherein, M is line number, Ω be oriental matrix (if the signal of corresponding grid is collected, value 1;Otherwise it is
0) element product, is indicated, | | |FThe F norm of expression, xiIndicate corresponding line of input, σ(E), W(E), b(E)Respectively indicate volume
Activation primitive (tanh () function), weight matrix and bias (default random initializtion) in code layer, σ(D), W(D), b(D)
Respectively indicate activation primitive (identity () function), the weight matrix (W in decoding layer(E)Transposed matrix) and bias
(default random initializtion), λ is hyper parameter (being selected by cross validation method), generates best self-encoding encoder model, i, j are respectively
Row, column number.
2, online phase of regeneration is divided into following steps (attached drawing 2):
(1) signal in different grids is calculated for the importance of the signal map accuracy of reconstruction and is ranked up.?
When crowdsourcing task starts, by historical signal average value (Wherein xI, jIndicate i-th in input data
Row, jth column element.) (generated the signal default value not being collected into as the input of self-encoding encoder with default signal map
Complete signal map) it compares, difference degree (absolute value of corresponding element difference) is as the important of next round collecting signal
Property, the more adjacent signal map rebuild twice in the case where collecting certain amount signal, and by its difference degree (corresponding element
The absolute value of plain difference) as next round collecting signal importance (when only collect one wheel signal when, then with default signal
Figure compares);(2) according to the size of estimated cost, quantity (in the present embodiment, the estimated cost of crowdsourcing signal needed for being arranged
It is set as crowdsourcing number of signals, defaults 500) and formulate corresponding incentive mechanism (such as game, cash) preferentially selecting importance
Higher grid is collected signal and uploads, to form incomplete signal map x;(3) this is collected into endless
Entire signal map is entered into the self-encoding encoder model of off-line training completion and is finely adjusted as input, obtains complete
Signal map
(4) its reconstruction errors is calculated using cross validation method to the signal being collected intoAnd with it is pre-
The threshold value first set is compared, if being unsatisfactory for threshold requirement, is repeated the above process up to reaching threshold requirement,Table
Show the average value of every column element in historical signal;It indicates to rebuild the signal map completed by self-encoding encoder.
Claims (7)
1. a kind of crowdsourcing signal map constructing method based on self-encoding encoder, characterized by the following steps:
(1) off-line training step: mould is trained off-line manner with the imperfect historical signal map being collected under crowdsourcing mode
Type;
(2) online phase of regeneration: using off-line training complete model to current collection to imperfect historical signal carry out it is scarce
Mistake value is inferred to rebuild complete signal map.
2. the crowdsourcing signal map constructing method according to claim 1 based on self-encoding encoder, it is characterised in that: the step
Suddenly the off-line training of (1) includes the following steps:
(1) geographic grid division is carried out according to the range of collected signal map, and will be uploaded in same time interval
History crowdsourcing signal is divided in different grids by GPS coordinate and then forms the imperfect signal map under multiple different moments;
(2) by the signal map under different moments it is open and flat be one-dimensional vector, and form the behavior moment, be classified as the history letter of grid number
Number map is input in self-encoding encoder and is trained then using the historical signal map as input;
(3) network structure, parameter and the smallest model of test error of self-encoding encoder are selected by cross validation method.
3. the crowdsourcing signal map constructing method according to claim 2 based on self-encoding encoder, it is characterised in that: the step
Suddenly the online reconstruction of (2) includes the following steps:
(1) signal in different grids is calculated for the importance of the signal map accuracy of reconstruction and is ranked up;
(2) according to the size of estimated cost, the quantity of crowdsourcing signal needed for being arranged simultaneously formulates corresponding incentive mechanism, preferential to select
Grid of high importance is collected signal and uploads, to form incomplete signal map;
(3) using the imperfect signal map being collected into as input, it is entered into the self-encoding encoder model of off-line training completion
In be finely adjusted, obtain complete signal map;
(4) its reconstruction precision is calculated using cross validation method to the signal being collected into, and is carried out with pre-set threshold value
Compare, if being unsatisfactory for threshold requirement, repeats the above process until reaching threshold requirement.
4. the crowdsourcing signal map constructing method according to claim 3 based on self-encoding encoder, it is characterised in that: the step
Suddenly the calculation method of (1) importance includes the following steps:
(1) input for the self-encoding encoder for completing the preset value of missing values in signal map as off-line training, output
Signal map as default;
(2) in the case where no crowdsourcing signal, the average value in historical signal is compared with default signal map, difference journey
Spend the importance as next round collecting signal;
(3) the case where collecting a number of signal, the more adjacent signal map rebuild twice, and using its difference degree as under
The importance of one wheel collecting signal.
5. the crowdsourcing signal map constructing method according to claim 4 based on self-encoding encoder, it is characterised in that: the step
Suddenly in (3) when only collecting the signal of a wheel, then compared with default signal map.
6. the crowdsourcing signal map constructing method according to claim 2 based on self-encoding encoder, it is characterised in that: the net
Network structure is the hidden layer number of plies and every layer of neuron number.
7. the crowdsourcing signal map constructing method according to claim 3 based on self-encoding encoder, it is characterised in that: described to swash
Encouraging mechanism is game mechanism or cash mechanism.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454208.3A CN110264154B (en) | 2019-05-28 | 2019-05-28 | Crowd-sourced signal map construction method based on self-encoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454208.3A CN110264154B (en) | 2019-05-28 | 2019-05-28 | Crowd-sourced signal map construction method based on self-encoder |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110264154A true CN110264154A (en) | 2019-09-20 |
CN110264154B CN110264154B (en) | 2023-06-09 |
Family
ID=67915724
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910454208.3A Active CN110264154B (en) | 2019-05-28 | 2019-05-28 | Crowd-sourced signal map construction method based on self-encoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110264154B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112261719A (en) * | 2020-07-24 | 2021-01-22 | 大连理智科技有限公司 | Area positioning method combining SLAM technology with deep learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8350758B1 (en) * | 2009-10-01 | 2013-01-08 | Lighthouse Signal Systems LLC | Systems and methods for indoor geolocation based on yield of RF signals |
CN103745489A (en) * | 2014-01-07 | 2014-04-23 | 电子科技大学 | Method for constructing base station signal field intensity map based on compressed sensing |
CN105263113A (en) * | 2015-10-26 | 2016-01-20 | 深圳大学 | Wi-Fi location fingerprint map building method and system based on crowd-sourcing |
CN106714110A (en) * | 2017-01-19 | 2017-05-24 | 深圳大学 | Auto building method and system of Wi-Fi position fingerprint map |
CN106792465A (en) * | 2016-12-26 | 2017-05-31 | 华中科技大学 | A kind of indoor fingerprint map constructing method based on mass-rent fingerprint |
-
2019
- 2019-05-28 CN CN201910454208.3A patent/CN110264154B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8350758B1 (en) * | 2009-10-01 | 2013-01-08 | Lighthouse Signal Systems LLC | Systems and methods for indoor geolocation based on yield of RF signals |
CN103745489A (en) * | 2014-01-07 | 2014-04-23 | 电子科技大学 | Method for constructing base station signal field intensity map based on compressed sensing |
CN105263113A (en) * | 2015-10-26 | 2016-01-20 | 深圳大学 | Wi-Fi location fingerprint map building method and system based on crowd-sourcing |
CN106792465A (en) * | 2016-12-26 | 2017-05-31 | 华中科技大学 | A kind of indoor fingerprint map constructing method based on mass-rent fingerprint |
CN106714110A (en) * | 2017-01-19 | 2017-05-24 | 深圳大学 | Auto building method and system of Wi-Fi position fingerprint map |
Non-Patent Citations (2)
Title |
---|
XIAOCHEN FAN: "iMap: A Crowdsensing Based System for Outdoor Radio Signal Strength Map", 《2016 IEEE TRUSTCOM-BIGDATASE-ISPA》 * |
周牧 等: "基于图像边缘检测的WLAN室内用户运动地图构建", 《电子科技大学学报》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112261719A (en) * | 2020-07-24 | 2021-01-22 | 大连理智科技有限公司 | Area positioning method combining SLAM technology with deep learning |
CN112261719B (en) * | 2020-07-24 | 2022-02-11 | 大连理智科技有限公司 | Area positioning method combining SLAM technology with deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN110264154B (en) | 2023-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109285346B (en) | Urban road network traffic state prediction method based on key road sections | |
CN111161535B (en) | Attention mechanism-based graph neural network traffic flow prediction method and system | |
CN109492830B (en) | Mobile pollution source emission concentration prediction method based on time-space deep learning | |
CN106815971B (en) | A kind of landslide disaster forecasting procedure based on RBF neural | |
CN103700255B (en) | A kind of traffic flow forecasting method based on spacetime correlation data mining | |
Roodposhti et al. | Towards automatic calibration of neighbourhood influence in cellular automata land-use models | |
CN103853786B (en) | The optimization method and system of database parameter | |
CN107301387A (en) | A kind of image Dense crowd method of counting based on deep learning | |
CN104900063B (en) | A kind of short distance running time Forecasting Methodology | |
CN104899431B (en) | Based on ant colony and swarm of particles into LSSVM fluctuating wind speed Forecasting Methodologies | |
CN107784380A (en) | The optimization method and optimization system of a kind of inspection shortest path | |
CN112905379B (en) | Traffic big data restoration method of graph self-encoder based on self-attention mechanism | |
CN115951014A (en) | CNN-LSTM-BP multi-mode air pollutant prediction method combining meteorological features | |
CN112149887A (en) | PM2.5 concentration prediction method based on data space-time characteristics | |
CN113610286A (en) | PM accounting for spatio-temporal correlations and meteorological factors2.5Concentration prediction method and device | |
CN110781595A (en) | Energy use efficiency PUE prediction method, device, terminal and medium | |
CN113516304A (en) | Space-time joint prediction method and device for regional pollutants based on space-time graph network | |
CN116011684A (en) | Traffic flow prediction method based on space-time diagram convolutional network | |
CN110264154A (en) | A kind of crowdsourcing signal map constructing method based on self-encoding encoder | |
CN101146313B (en) | A correction method of radio transmission model | |
Wang et al. | Two-dimension monthly river flow simulation using hierarchical network-copula conditional models | |
CN116976227B (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
CN113515798A (en) | Urban three-dimensional space expansion simulation method and device | |
CN104573216A (en) | Antenna performance optimizing method and device | |
CN111242266A (en) | Operation data management system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |