CN103076035A - Sensor measuring method based on double support vector machines - Google Patents

Sensor measuring method based on double support vector machines Download PDF

Info

Publication number
CN103076035A
CN103076035A CN2012105834636A CN201210583463A CN103076035A CN 103076035 A CN103076035 A CN 103076035A CN 2012105834636 A CN2012105834636 A CN 2012105834636A CN 201210583463 A CN201210583463 A CN 201210583463A CN 103076035 A CN103076035 A CN 103076035A
Authority
CN
China
Prior art keywords
sensor
model
error
support vector
output
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN2012105834636A
Other languages
Chinese (zh)
Other versions
CN103076035B (en
Inventor
黄为勇
高玉芹
田秀玲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xuzhou University of Technology
Original Assignee
Xuzhou University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xuzhou University of Technology filed Critical Xuzhou University of Technology
Priority to CN201210583463.6A priority Critical patent/CN103076035B/en
Publication of CN103076035A publication Critical patent/CN103076035A/en
Application granted granted Critical
Publication of CN103076035B publication Critical patent/CN103076035B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention discloses a sensor measuring method based on double support vector machines, belonging to the field of sensor information processing. The method comprises the following steps: establishing a data sample by sensor calibration data under the working condition environment; constructing a sensor inverse model based on the support vector machines by applying the data sample; establishing a second data sample by applying error data of sensor output and sensor inverse model output to construct an error model based on the support vector machines; selecting and optimizing the parameters of the sensor inverse model and the error model through a quantum-behaved particle swarm optimization algorithm, as well as minimum standards of root-mean-square error of the model and maximum absolute error; and taking the difference of the sensor inverse model output and the error model output as a measured true valve. The sensor measuring method has the advantages that the influence of non-linearity property of the sensor on the measuring result can be effectively lowered; the measured high-precision measurement is realized; the modeling working amount is low; and the sensor measuring method based on the double support vector machines can be widely applied to the field of high-precision measurement of various sensors.

Description

A kind of sensor measurement based on two support vector machine
Technical field
The invention belongs to the sensor information process field, specifically a kind of sensor measurement based on two support vector machine.
Background technology
Sensor is widely used in various fields such as industry, agricultural, national defence, science and technology, has become the basis of advanced information society.Because the impact of the many factors such as sensor sensing element characteristic, applied environment, sensor are aging so that the output of sensor with input between be a kind of nonlinear relationship of complexity, thereby in practical engineering application, bring larger measuring error.
Sensor inverse modeling method based on machine learning is effective technology means that realize the sensor nonlinear characteristic compensation and improve the sensor measurement precision at present.Neural network has very strong non-linear mapping capability and powerful self-learning capability, can match Nonlinear Mapping and need not to set up its analytic model arbitrarily, become an important method of sensor inverse modeling.Need the defectives such as large sample and generalization ability be poor yet the method exists, to have affected the raising of sensor measurement precision.
Support vector machine (support vector machine, SVM) is to be based upon the VC dimension theory of Statistical Learning Theory and a kind of Novel learning method on the structural risk minimization basis.Than neural net method, support vector machine has solved the problems such as small sample, non-linear and local minimum point effectively, has preferably Generalization Capability, is obtaining application aspect the sensor inverse modeling.
Because the sensor input-output characteristic has the non-linear of height, adopt based on the contrary modeling of support vector machine match sensor inverse characteristic accurately, still there is suitable measuring error in employing the method.Along with the development of science and technology, people are more and more higher to the requirement of sensor measurement precision, realize that the sensor high-acruracy survey is significant.
Summary of the invention
In order to overcome above-mentioned the deficiencies in the prior art, the invention provides a kind of sensor measurement based on two support vector machine, realize the effective compensation of sensor nonlinear characteristic, reach the purpose of carrying out high-acruracy survey to measured.
The present invention realizes with following technical scheme: a kind of sensor measurement based on two support vector machine, and utilize the transducer calibration data to set up data sample, to use support vector machine and make up respectively sensor inverse model and error model, concrete steps are as follows:
Step 1: sensor is demarcated under work condition environment, obtained n sensor input-output data (y i, x i), wherein: y iWith x iBe respectively sensor output value and input value (i=1,2 ..., n); And it is divided into training sample set S1 and test sample book collection S2 two parts;
Step 2: set up one based on the sensor inverse model of support vector machine by training sample set S1, the parameter of sensor inverse model is carried out selection and optimization by the criterion of intelligent optimization algorithm and test sample book collection S2 error minimum;
Step 3: by sensor output y iWith sensor inverse model output error e iData (y i, e i) set up data sample (i=1,2 ..., n), and it is divided into training sample set S3 and test sample book collection S4 two parts;
Step 4: make up the error model based on support vector machine by training sample set S3, the parameter of error model is carried out selection and optimization by the criterion of intelligent optimization algorithm and test sample book collection S4 error minimum;
Step 5: when actual measurement, the output valve x* of sensor inverse model and the output valve e* of error model are subtracted each other, its difference z=x*-e* is as measured true value.
It further is: the sensor inverse model is made up by training sample set S1, and its model parameter adopts the minimum criterion of quantum particle swarm optimization and test sample book collection S2 De Zhuo root error RMSE and maximum absolute error MAE while to be optimized;
The expression formula of RMSE and MAE is respectively:
Figure 2012105834636100002DEST_PATH_IMAGE001
(1)
Figure 742174DEST_PATH_IMAGE002
(2)
Wherein,
Figure 2012105834636100002DEST_PATH_IMAGE003
Be the model output valve,
Figure 568792DEST_PATH_IMAGE004
Be the model desired output, n is the sample number of test sample book collection.
Error model is made up by training sample set S3, and its model parameter adopts the minimum criterion of quantum particle swarm optimization and test sample book collection S4 De Zhuo root error RMSE and maximum absolute error MAE while to be optimized;
The expression formula of RMSE and MAE is respectively:
Figure 879818DEST_PATH_IMAGE001
(1)
Figure 422706DEST_PATH_IMAGE002
(2)
Wherein,
Figure 248710DEST_PATH_IMAGE003
Be the model output valve,
Figure 944265DEST_PATH_IMAGE004
Be the model desired output, n is the sample number of test sample book collection.
The invention has the beneficial effects as follows: adopt support vector machine to make up respectively sensor inverse model and error model, utilize the output of error model that the Output rusults of sensor inverse model is proofreaied and correct, the parameter of two models all adopts quantum particle swarm optimization to carry out selection and optimization, effectively reduce the nonlinear characteristic of sensor to the impact of measurement result, realization is to measured high-acruracy survey, and the modeling workload is less, can be widely used in various sensor high-acruracy surveys field.Because support vector machine has the advantage of small-sample learning, quantum particle swarm optimization have calculating fast, be easy to realize and control the few characteristics of parameter, the present invention can realize the high-acruracy survey of sensor in the Small samples modeling situation.
Description of drawings
The invention will be further described below in conjunction with drawings and Examples.
Fig. 1 is the measuring system structured flowchart;
Fig. 2 is the sensor samples data and curves;
Fig. 3 is based on the sensor inverse model modeling process flow diagram of support vector machine;
Fig. 4 is based on the error model modeling process flow diagram of support vector machine;
Fig. 5 is the input-output curve of test data.
Embodiment
As shown in Figure 1, a kind of sensor measurement based on two support vector machine, utilize the transducer calibration data to set up data sample, use support vector machine and make up respectively sensor inverse model and error model, the parameter of two supporting vector machine models is carried out selection and optimization by the root-mean-square error of quantum particle swarm optimization and model and the criterion of maximum absolute error while minimum, the difference of exporting with sensor inverse model output and error model is as measured true value, realize the effective compensation of sensor nonlinear characteristic, reach the purpose of carrying out high-acruracy survey to measured.
Present embodiment describes this method with temperature sensor, and sensor inverse model and error model all adopt quantum particle swarm optimization to carry out selection and optimization.
Step 1: sensor is demarcated under work condition environment, obtained the output data of 23 sensors of input temp between 210 ℃ ~ 430 ℃, it is consisted of 23 sensor input-output data (y i, x i), wherein: y iWith x iBe respectively the output of sensor and input value (i=1,2 ..., 23).And it is divided into training sample set S1 and test sample book collection S2 two parts, among the embodiment, the sample number of S1 is that the sample number of 18, S2 is 5.
Step 2: by the sensor inverse model of training sample set S1 foundation based on support vector machine.The kernel function of support vector machine is got the RBF kernel function among the embodiment, and its formula is:
Figure DEST_PATH_IMAGE005
(3)
Wherein:
Figure 982235DEST_PATH_IMAGE006
,
Figure DEST_PATH_IMAGE007
Be the mode input parameter;
Figure 547952DEST_PATH_IMAGE008
Be spread factor.
The parameter of model comprises insensitive loss coefficient ε, punishment coefficient C and kernel function spread factor
Figure 798936DEST_PATH_IMAGE008
Deng 3 parameters, these 3 parameters are carried out selection and optimization by the root-mean-square error RMSE of quantum particle swarm optimization and test sample book collection S2 and the criterion of maximum absolute error MAE minimum.
The fitness function expression formula of quanta particle swarm optimization is:
Figure DEST_PATH_IMAGE009
(4)
Wherein:
Figure 967355DEST_PATH_IMAGE003
The model output valve,
Figure 128341DEST_PATH_IMAGE004
Be the model desired output, n is the sample number of test sample book collection, gets 5 among the embodiment.
Contraction-the flare factor of quantum particle swarm optimization
Figure 51910DEST_PATH_IMAGE010
Adopt linearity to be decreased to 0.5 from 1.0.Use the parameter (ε that obtains the sensor inverse model among the embodiment based on the sensor inverse model modeling flow process of support vector machine shown in Figure 3 1, C 1,
Figure 790190DEST_PATH_IMAGE008
1)=(376.9372,0.05046,0.54564).
Step 3: by sensor output y iWith sensor inverse model output error e iData (y i, e i) set up data sample, and it is divided into training sample set S3 and test sample book collection S4 two parts, among the embodiment, the sample number of S3 is that the sample number of 18, S2 is 5.
Step 4: by the error model of training sample set S3 structure based on support vector machine.In an embodiment, adopt the RBF kernel function during support vector machine error model.
Error model insensitive loss coefficient ε, penalty coefficient C and kernel function spread factor
Figure 827547DEST_PATH_IMAGE008
Adjusted by the root-mean-square error RMSE of quantum particle swarm optimization and test set S4 and the criterion of maximum absolute error MAE equally Deng 3 parameters.The fitness function expression formula of quantum particle swarm optimization is:
Figure 761481DEST_PATH_IMAGE009
(5)
Wherein:
Figure 921198DEST_PATH_IMAGE003
The model output valve,
Figure 209091DEST_PATH_IMAGE004
Be the model desired output, n is the number of test sample book, gets 5 among the embodiment.
Contraction-the flare factor of quantum particle swarm optimization
Figure 112456DEST_PATH_IMAGE010
Adopt linearity to be decreased to 0.5 from 1.0.Use the error model modeling flow process based on support vector machine shown in Figure 4 and obtain error model parameters (ε among the embodiment 2, C 2,
Figure 41841DEST_PATH_IMAGE008
2)=(1000,1.1880,1.1303).
Step 5: when actual measurement, make up sensor measuring system by Fig. 1, the output valve x* of sensor inverse model and the output valve e* of error model are subtracted each other, z=x*-e* is as final measurement for its difference.
The input-output curve of 5 test sample books among the embodiment as shown in Figure 5, its absolute error and relative error are as shown in table 1.By Fig. 5 with as seen from Table 1, the present invention has realized the highly linear of sensor characteristic preferably, has reached the purpose to measured high-acruracy survey, table 1 is measurement result and the measuring error of test data.
The sensor input/℃ 240 280 320 360 400
Measurement result/℃ 239.3071 279.1226 320.7675 360.2137 400.1033
Absolute error/℃ -0.6929 -0.8774 0.7675 0.2137 0.1033
Relative error/% -0.2887 -0.3134 0.2398 0.0594 0.0258
Table 1
Above-mentioned by reference to the accompanying drawings embodiment has been described in detail during to this; any those skilled in the art of being familiar with can make the many variations example to the present invention according to the above description; thereby the details among the embodiment should not consist of limitation of the invention, and the scope that the present invention will define with appended claims is as protection scope of the present invention.

Claims (3)

1. sensor measurement based on two support vector machine is characterized in that: utilize the transducer calibration data to set up data sample, use support vector machine and make up respectively sensor inverse model and error model, concrete steps are as follows:
Step 1: sensor is demarcated under work condition environment, obtained n sensor input-output data (y i, x i), wherein: y iWith x iBe respectively sensor output value and input value (i=1,2 ..., n); And it is divided into training sample set S1 and test sample book collection S2 two parts;
Step 2: set up one based on the sensor inverse model of support vector machine by training sample set S1, the parameter of sensor inverse model is carried out selection and optimization by the criterion of intelligent optimization algorithm and test sample book collection S2 error minimum;
Step 3: by sensor output y iWith sensor inverse model output error e iData (y i, e i) set up data sample (i=1,2 ..., n), and it is divided into training sample set S3 and test sample book collection S4 two parts;
Step 4: make up the error model based on support vector machine by training sample set S3, the parameter of error model is carried out selection and optimization by the criterion of intelligent optimization algorithm and test sample book collection S4 error minimum;
Step 5: when actual measurement, the output valve x* of sensor inverse model and the output valve e* of error model are subtracted each other, its difference z=x*-e* is as measured true value.
2. a kind of sensor measurement based on two support vector machine according to claim 1, be characterised in that: the sensor inverse model is made up by training sample set S1, and its model parameter adopts the minimum criterion of quantum particle swarm optimization and test sample book collection S2 De Zhuo root error RMSE and maximum absolute error MAE while to be optimized;
The expression formula of RMSE and MAE is respectively:
Figure 2012105834636100001DEST_PATH_IMAGE001
(1)
Figure 588631DEST_PATH_IMAGE002
(2)
Wherein,
Figure DEST_PATH_IMAGE003
Be the model output valve,
Figure 173939DEST_PATH_IMAGE004
Be the model desired output, n is the sample number of test sample book collection.
3. a kind of sensor measurement based on two support vector machine according to claim 1, be characterised in that: error model is made up by training sample set S3, and its model parameter adopts the minimum criterion of quantum particle swarm optimization and test sample book collection S4 De Zhuo root error RMSE and maximum absolute error MAE while to be optimized;
The expression formula of RMSE and MAE is respectively:
Figure 553099DEST_PATH_IMAGE001
(1)
Figure 730134DEST_PATH_IMAGE002
(2)
Wherein, Be the model output valve,
Figure 225630DEST_PATH_IMAGE004
Be the model desired output, n is the sample number of test sample book collection.
CN201210583463.6A 2012-12-28 2012-12-28 A kind of sensor measurement based on two support vector machine Expired - Fee Related CN103076035B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210583463.6A CN103076035B (en) 2012-12-28 2012-12-28 A kind of sensor measurement based on two support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210583463.6A CN103076035B (en) 2012-12-28 2012-12-28 A kind of sensor measurement based on two support vector machine

Publications (2)

Publication Number Publication Date
CN103076035A true CN103076035A (en) 2013-05-01
CN103076035B CN103076035B (en) 2015-10-07

Family

ID=48152645

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210583463.6A Expired - Fee Related CN103076035B (en) 2012-12-28 2012-12-28 A kind of sensor measurement based on two support vector machine

Country Status (1)

Country Link
CN (1) CN103076035B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104504407A (en) * 2014-12-17 2015-04-08 西南大学 Electronic nose feature selection optimization method on basis of multiple Fisher kernel discriminant analysis
CN104880216A (en) * 2015-06-17 2015-09-02 北京理工大学 Method for sensor fault identification based on cross usage of different error correction codes
CN107192690A (en) * 2017-05-19 2017-09-22 重庆大学 Near infrared spectrum Noninvasive Blood Glucose Detection Methods and its detection network model training method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009192312A (en) * 2008-02-13 2009-08-27 Nippon Telegr & Teleph Corp <Ntt> Device for predicting generation of thunder, and method therefor, and program
CN102162733A (en) * 2011-01-11 2011-08-24 哈尔滨工程大学 Method for correcting autonomous underwater vehicle (AUV) dead reckoning navigation error in real time based on space vector modulation (SVM)
KR20120080270A (en) * 2011-01-07 2012-07-17 동명대학교산학협력단 System for processing biological signal and portable instrumnet for processing biological signal
CN102778538A (en) * 2012-07-06 2012-11-14 广东电网公司电力科学研究院 Soft measuring method based on improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009192312A (en) * 2008-02-13 2009-08-27 Nippon Telegr & Teleph Corp <Ntt> Device for predicting generation of thunder, and method therefor, and program
KR20120080270A (en) * 2011-01-07 2012-07-17 동명대학교산학협력단 System for processing biological signal and portable instrumnet for processing biological signal
CN102162733A (en) * 2011-01-11 2011-08-24 哈尔滨工程大学 Method for correcting autonomous underwater vehicle (AUV) dead reckoning navigation error in real time based on space vector modulation (SVM)
CN102778538A (en) * 2012-07-06 2012-11-14 广东电网公司电力科学研究院 Soft measuring method based on improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
李亚楠，等: "基于支持向量机的传感器故障诊断研究", 《中国造船工程学会2005舰船电子装备维修理论与应用研讨会》 *
黄为勇，等: "矿井CO 浓度的支持向量机数据融合检测方法", 《计算机工程与应用》 *
黄为勇: "基于支持向量机数据融合的矿井瓦斯预警技术研究", 《中国优秀博士论文全文库工程科技Ⅰ辑》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104504407A (en) * 2014-12-17 2015-04-08 西南大学 Electronic nose feature selection optimization method on basis of multiple Fisher kernel discriminant analysis
CN104504407B (en) * 2014-12-17 2018-02-13 西南大学 Electronic nose feature selecting optimization method based on more kernel Fisher discriminant analysis
CN104880216A (en) * 2015-06-17 2015-09-02 北京理工大学 Method for sensor fault identification based on cross usage of different error correction codes
CN107192690A (en) * 2017-05-19 2017-09-22 重庆大学 Near infrared spectrum Noninvasive Blood Glucose Detection Methods and its detection network model training method
CN107192690B (en) * 2017-05-19 2019-04-23 重庆大学 Near infrared spectrum Noninvasive Blood Glucose Detection Methods and its detection network model training method

Also Published As

Publication number Publication date
CN103076035B (en) 2015-10-07

Similar Documents

Publication Publication Date Title
US11150231B2 (en) Method for measuring a mudflat elevation by remotely sensed water content
CN102353481B (en) Method and device for complementing temperature and pressure of pressure sensor based on two-dimensional orthogonal function
CN101858811B (en) Method for compensating signal of high-precision pressure sensor
CN104992225B (en) A kind of compensation method of large deformation flexible body dynamic measuring data temperature and humidity
CN108332837A (en) It is a kind of applied to electronic scale based on BP neural network compensation temperature gain method
CN109255160A (en) Unit time delay prediction method and unit time-delay sensitivity calculation method neural network based
CN103335814B (en) Correction method for inclination angle measurement error data of experimental model in wind tunnel
CN104331591A (en) Granary grain storage quantity detection method based on support vector regression
CN113176393B (en) HASM model-based three-dimensional estimation method and system for soil organic carbon reserves
CN103063233B (en) A kind of method that adopts multisensor to reduce measure error
CN103076035A (en) Sensor measuring method based on double support vector machines
Santhosh et al. Online implementation of an adaptive calibration technique for displacement measurement using LVDT
CN104122031A (en) Silicon pressure sensor temperature compensation method based on extreme learning machine
CN106354016B (en) A kind of control method of metal tube compacting height
Li et al. Soil moisture dynamics and implications for irrigation of farmland with a deep groundwater table
CN104535257B (en) A kind of silicon pressure drag temperature-compensating appraisal procedure
CN115147012A (en) Carbon emission accounting method and device based on neural network model
Zhao et al. A comprehensive compensation method for piezoresistive pressure sensor based on surface fitting and improved grey wolf algorithm
CN103940453B (en) A kind of raising sensor accuracy class method
CN103344740B (en) Based on the glutamic acid production concentration online soft sensor method of multi input Wiener model
CN103925939B (en) The calibration compensation method of a kind of photoelectric coding device gamut
CN116296047B (en) Temperature compensation improvement method of monocrystalline silicon pressure transmitter
CN107730045A (en) A kind of baseline load thermal inertia modification method based on discrete inertia force system
CN104703277B (en) One-dimensional linearity region wireless sensor network node locating method
Potapovs et al. Continuous crop weight measurement sensor calibration algorithm for industrial greenhouse

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20151007

Termination date: 20151228

EXPY Termination of patent right or utility model