CN102254227A - Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things - Google Patents
Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things Download PDFInfo
- Publication number
- CN102254227A CN102254227A CN201110187826XA CN201110187826A CN102254227A CN 102254227 A CN102254227 A CN 102254227A CN 201110187826X A CN201110187826X A CN 201110187826XA CN 201110187826 A CN201110187826 A CN 201110187826A CN 102254227 A CN102254227 A CN 102254227A
- Authority
- CN
- China
- Prior art keywords
- neural network
- decision table
- rough set
- sample
- things
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of the Internet of things. The method comprises the following steps of: firstly, performing heuristic reduction on a sample space according to the significance of attributes by using a rough set so as to obtain a reduced decision table after eliminating the same samples; and secondly, performing training, by using a BP neural network, on the decision table until convergence. With the technical proposal of the invention, the purposes of simplifying the network structure, speeding up the network convergence and improving the system instantaneity can be achieved.
Description
Technical field
The present invention relates to Internet of Things Multi-sensor Fusion algorithm, relate in particular to a kind of method that merges based on the BP neural network of rough set of Internet of Things multisensor.
Background technology
The normal multisensor that uses is monitored (as vibration, temperature, humidity, pressure, flow etc.) to the various features amount in actual Internet of things system, and the information of these sensors is merged, and obtains compatibility of goals and explains and description.Its core is to select suitable blending algorithm, requires to have the ability of robustness, parallel processing, also will guarantee arithmetic speed and given precision.And artificial neural network is to be interconnected by a large amount of basic neurons to form, can carry out distributed parallel handles and non-linear conversion, have powerful study and sum up the function of concluding, processing speed is fast, the advantage that data space is little exactly can satisfy the requirement of multisensor syste to blending algorithm.When using neural network as data fusion model, the input information of network is the various measurement parameters of multisensor to target, and the output of network is to the pattern-recognition of target or classification results or other response results.
But traditional neural network also has certain limitation aspect fusion, and it does not possess the preprocessing function to the input sample space.When input feature vector amount dimension was big, neural network is complex structure not only, and the training time prolongs greatly, and real-time is also bad.
Summary of the invention
The object of the present invention is to provide a kind of method that merges based on the BP neural network of rough set of Internet of Things multisensor, can simplify network structure, accelerate network convergence speed, improve the real-time of system.
For achieving the above object, the present invention is achieved through the following technical solutions:
A kind of method that merges based on the BP neural network of rough set of Internet of Things multisensor, 1, original training sample is carried out discretize, draw decision table;
2, utilize the heuritic approach of the attribute importance degree of rough set that decision table is carried out yojan, eliminate identical sample, draw the yojan decision table;
3, the proper vector of each sample of yojan decision table input BP neural metwork training, refreshing weight and threshold value successively are till satisfying given accuracy requirement;
4, with the BP neural network that trains unbred sample subclass in the same sample set is tested again, draw recognition result.
Adopt technical scheme of the present invention, rough set is incorporated in the Fusion Model of neural network, utilize rough set attribute reduction and do not change the characteristics of classification capacity and the advantage of neural network concurrent processing power and powerful fault-tolerant ability, earlier the input feature vector amount is carried out the dimensionality reduction operation, eliminate redundant attributes,, reach and simplify network structure again via neural metwork training, accelerate network convergence speed, improve the purpose of real-time.
Description of drawings
According to embodiment and accompanying drawing the present invention is described in further detail below.
Fig. 1 uses original decision table for the training of the method that merges based on the BP neural network of rough set of a kind of Internet of Things multisensor of the embodiment of the invention;
Fig. 2 is that the training of the method that merges based on the BP neural network of rough set of a kind of Internet of Things multisensor of the embodiment of the invention is used and simplified decision table;
Fig. 3 is the original decision table verify error curve of the method that merges based on the BP neural network of rough set of a kind of Internet of Things multisensor of the embodiment of the invention;
Fig. 4 is a verify error curve after the Rough Set Reduction of the method that merges based on the BP neural network of rough set of a kind of Internet of Things multisensor of the embodiment of the invention.
Embodiment
With CTR (the Car Test Results) database that is published on " Popular Science " is example, uses the recognition effect that blending algorithm checking that trifle proposes increases the pretreated BP network of rough set.This database conditional attribute has 9, uses x1 respectively, x2 ..., x9 represents; Decision attribute has only one, represents with y, sees Table 1.Wherein each Column Properties implication is as follows:
x1:size,overall?length;
x2:number?of?cylinders;
x3:presence?of?a?turbocharger;
x4:type?of?fuel?system;
x5:engine?displacement;
x6:compression?ratio;
x7:power;
x8:type?of?transmission;
x9:weight,
y:mileage。
Each property value implication is:
c:compact;s:subcompact;sm:small;
y:yes; n:no; E:EFI;
B:2-BBL; m:medium; ma:manual;h:high;
he:heavy; l:light;
lo:low; a:auto。
Total sample number is 21, in 7: 3 ratios sample set is divided into training sample and test samples, promptly preceding 15 kinds as training sample, back 6 kinds as test samples.
1 discretize Discretization
Because rough set can only be handled discrete data, so need earlier the continuous type knowledge base to be carried out discretize.The discrete logarithm kind is a lot, equidistant method, equifrequent method, experience split plot design is arranged, clustering procedure, greedy algorithm and improved greedy algorithm etc.Property value in the table 1 is the logical type discrete variable, only needs when using matlab to handle its digitizing is got final product.At first Column Properties, replace c with numeral 1, numeral 0 replaces s.Use the same method and handle other attribute column, obtain decision table 2 at last.
Table 1 CTR database
Table 2 CTR decision table after digitalized
2 ask the relative yojan get the minimal set of the attributes of decision table
After the discretize, adopt Pawlak attribute importance degree algorithm that decision table is carried out attribute reduction.
(1) at first obtains the relative nuclear of decision table.Obtain the relative nuclear B:B=CORE of decision table 2 according to the definition of front
C(D)={ x4, x9}={type of fuel system, weight}, redundant attributes collection R={x1 then, x2, x3, x4, x5, x7, x8}.
(2) then, judge whether relative nuclear is the relative yojan of decision table.
Pos
C(D)={ u1} ∪ { u2} ∪ ... ∪ { u21}=U, pos
B(D)={ u4, u11} ∪ (u6} ∪ { u7} ∪ { u8}, obviously posB (D)
1PosC (D) is so examine the relative yojan that B is not a decision table relatively.
(3), according to each redundant attributes the importance degree of relative nuclear B is added single attribute successively so and obtain new combinations of attributes and be used to judge relative yojan if relatively nuclear is not the relative yojan of decision table.In this example, the yojan of decision attribute has P1={x1, x4, x5, x9}, P2={x1, x4, x6, x7, x9}, P3={x2, x4, x6, x7, x9} relatively.
(4) last, set up decision table again by the attribute column and the decision attribute of relative yojan P1 correspondence.Delete wherein identical row, obtain final decision table, as table 3.
Table 3 The Decision Table after Attribute Reduction
3 use the BP neural metwork training
This example makes up two-layer BP neural network, with the sample of table 3 it is trained.Wherein, attribute x1, x4, x5, each row sample (property value) of x9 correspondence will be imported the BP network successively and train, and the input vector of network is reduced to 4 present dimensions by 9 original dimensions.Because the decision attribute value of table 3 is 1,2,3, represents three kinds of ranks of speedometer for automobile respectively, so the output layer of network is selected " pureline " linear function for use, the node number is 1.Hidden layer uses " sigmoid " transition function.The node number substitution experimental formula of input vector number and output layer is tried to achieve this layer neuron number between 3~12.Design a BP network that the node number is variable, to such as table 4, the hidden layer node number is 10 time error minimums through error, and only needs for 131 steps just can restrain.
Table 4 the Error of the network trained
|
3 | 4 | 5 | 6 | 7 |
error | 0.1223 | 0.1666 | 0.1225 | 0.1219 | 0.1215 |
nodes | 8 | 9 | 10 | 11 | 12 |
error | 0.1200 | 0.1216 | 0.1166 | 0.1195 | 0.1172 |
Trained by the BP neural network after using original decision table to train and utilize Rough Set Reduction by the BP neural network, the training process that draws such as Fig. 1 are shown in 2 again.
After the BP network training is good, network is tested by 6 test samples.Because this moment, the input vector of network was four-dimensional, only needed the property value fan-in network of attribute x1, the x4 of test samples, x5, x9 correspondence is got final product.Through experimental verification, the BP network after Rough Set Reduction has good recognition capability, and discrimination reaches 83%, as shown in Figure 4.Discrimination without the BP network of Rough Set Reduction is 67%, as shown in Figure 3.As can be seen, the BP network integration model that rough set is preposition, not only network structure has become simply, and has reached the better recognition effect.
Claims (1)
1. the method that merges based on the BP neural network of rough set of an Internet of Things multisensor is characterized in that:
A. original training sample is carried out discretize, draw decision table;
B. utilize the heuritic approach of the attribute importance degree of rough set that decision table is carried out yojan, eliminate identical sample, draw the yojan decision table;
C. the proper vector of each sample of yojan decision table input BP neural metwork training, refreshing weight and threshold value successively are till satisfying given accuracy requirement;
D. with the BP neural network that trains unbred sample subclass in the same sample set is tested again, draw recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110187826XA CN102254227A (en) | 2011-07-06 | 2011-07-06 | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110187826XA CN102254227A (en) | 2011-07-06 | 2011-07-06 | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102254227A true CN102254227A (en) | 2011-11-23 |
Family
ID=44981475
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201110187826XA Pending CN102254227A (en) | 2011-07-06 | 2011-07-06 | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102254227A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103778479A (en) * | 2014-01-10 | 2014-05-07 | 国网上海市电力公司 | Adaptive information fault-tolerant protection method |
CN106503798A (en) * | 2015-09-08 | 2017-03-15 | 中国科学院沈阳自动化研究所 | Based on rough set and the method for diagnosing faults of the pump of BP neural network |
CN108921342A (en) * | 2018-06-26 | 2018-11-30 | 圆通速递有限公司 | A kind of logistics customer churn prediction method, medium and system |
CN109512390A (en) * | 2018-10-26 | 2019-03-26 | 武汉大学 | Sleep stage method and wearable device based on EEG time domain various dimensions feature and M-WSVM |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1389710A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-sensor and multiple-object information fusing method |
CN1474177A (en) * | 2003-08-08 | 2004-02-11 | 王耀南 | Multiple sensor integrated intelligent transparent container detecting device and method |
CN101110106A (en) * | 2007-06-21 | 2008-01-23 | 上海交通大学 | Multiple sensor information amalgamation method combining rough set and neural network |
CN101430309A (en) * | 2008-11-14 | 2009-05-13 | 西安建筑科技大学 | Environmental quality evaluation method based on rough set-RBF neural network |
DE102007055639A1 (en) * | 2007-11-21 | 2009-05-28 | Robert Bosch Gmbh | Actual vehicle condition determining device, has evaluation and control unit evaluating and combining signals of sensor units for determining motion variables, where sensor units detect movement signals |
-
2011
- 2011-07-06 CN CN201110187826XA patent/CN102254227A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1389710A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-sensor and multiple-object information fusing method |
CN1474177A (en) * | 2003-08-08 | 2004-02-11 | 王耀南 | Multiple sensor integrated intelligent transparent container detecting device and method |
CN101110106A (en) * | 2007-06-21 | 2008-01-23 | 上海交通大学 | Multiple sensor information amalgamation method combining rough set and neural network |
DE102007055639A1 (en) * | 2007-11-21 | 2009-05-28 | Robert Bosch Gmbh | Actual vehicle condition determining device, has evaluation and control unit evaluating and combining signals of sensor units for determining motion variables, where sensor units detect movement signals |
CN101430309A (en) * | 2008-11-14 | 2009-05-13 | 西安建筑科技大学 | Environmental quality evaluation method based on rough set-RBF neural network |
Non-Patent Citations (2)
Title |
---|
张兆礼等: "粗神经网络及其在数据融合中的应用", 《控制与决策》 * |
汪凌等: "基于粗糙集的决策系统知识获取算法及实证分析", 《情报杂志》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103778479A (en) * | 2014-01-10 | 2014-05-07 | 国网上海市电力公司 | Adaptive information fault-tolerant protection method |
CN106503798A (en) * | 2015-09-08 | 2017-03-15 | 中国科学院沈阳自动化研究所 | Based on rough set and the method for diagnosing faults of the pump of BP neural network |
CN108921342A (en) * | 2018-06-26 | 2018-11-30 | 圆通速递有限公司 | A kind of logistics customer churn prediction method, medium and system |
CN108921342B (en) * | 2018-06-26 | 2022-07-12 | 圆通速递有限公司 | Logistics customer loss prediction method, medium and system |
CN109512390A (en) * | 2018-10-26 | 2019-03-26 | 武汉大学 | Sleep stage method and wearable device based on EEG time domain various dimensions feature and M-WSVM |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Shao et al. | Rolling bearing fault diagnosis using adaptive deep belief network with dual-tree complex wavelet packet | |
Li et al. | A deep transfer nonnegativity-constraint sparse autoencoder for rolling bearing fault diagnosis with few labeled data | |
CN105224872A (en) | A kind of user's anomaly detection method based on neural network clustering | |
CN113869208B (en) | Rolling bearing fault diagnosis method based on SA-ACWGAN-GP | |
CN102254227A (en) | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things | |
CN109406118A (en) | A kind of mechanical failure prediction method based on level convolutional neural networks | |
WO2022040972A1 (en) | Product information visualization processing method and apparatus, and computer device | |
Rai et al. | Prediction models for performance and emissions of a dual fuel CI engine using ANFIS | |
CN104680233A (en) | Wavelet neural network-based engine failure diagnosing method | |
CN110243590A (en) | A kind of Fault Diagnosis Approach For Rotor Systems learnt based on principal component analysis and width | |
Hu et al. | Driver identification using 1D convolutional neural networks with vehicular CAN signals | |
CN111860775A (en) | Ship fault real-time diagnosis method based on CNN and RNN fusion | |
Montazeri et al. | Driving segment simulation for determination of the most effective driving features for HEV intelligent control | |
CN116337449A (en) | Sparse self-coding fault diagnosis method and system based on information fusion | |
CN109341989B (en) | Bridge influence line identification method capable of eliminating vehicle power effect | |
Wang et al. | Predictability of vehicle fuel consumption using LSTM: Findings from field experiments | |
Esser et al. | Enabling a privacy-preserving synthesis of representative driving cycles from fleet data using data aggregation | |
Cheng et al. | A rolling bearing fault diagnosis model based on WCNN-BiGRU | |
Zhang et al. | Novel method of flatness pattern recognition via cloud neural network | |
Zhang et al. | Research on fault diagnosis and signal reconstruction technology of diesel engine NOx sensor based on deep learning algorithm | |
Liu et al. | The optimal road grade design for minimizing ground vehicle energy consumption | |
Do et al. | A comparative study of machine learning techniques in prediction of exhaust emissions and performance of a diesel engine fuelled with biodiesel blends | |
Abouel-Seoud et al. | Real world evaluation and control of vehicle engine exhaust air pollution using adaptive neural fuzzy inference system | |
CN114998075A (en) | Transportation carbon emission calculation method and system and storable medium | |
Yan et al. | Fault diagnosis based on a stacked sparse auto-encoder network and KNN classifier |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20111123 |