CN109709497A - A kind of high-accuracy position system and method - Google Patents
A kind of high-accuracy position system and method Download PDFInfo
- Publication number
- CN109709497A CN109709497A CN201910118352.XA CN201910118352A CN109709497A CN 109709497 A CN109709497 A CN 109709497A CN 201910118352 A CN201910118352 A CN 201910118352A CN 109709497 A CN109709497 A CN 109709497A
- Authority
- CN
- China
- Prior art keywords
- error
- layer
- magnetic
- output layer
- threshold value
- 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
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000005291 magnetic effect Effects 0.000 claims abstract description 127
- 238000005457 optimization Methods 0.000 claims abstract description 15
- 238000005520 cutting process Methods 0.000 claims abstract description 14
- 230000004913 activation Effects 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 29
- 150000001875 compounds Chemical class 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 230000033001 locomotion Effects 0.000 claims description 10
- 238000002945 steepest descent method Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004807 localization Effects 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 abstract description 40
- 238000000605 extraction Methods 0.000 abstract 1
- 238000004422 calculation algorithm Methods 0.000 description 25
- 230000006870 function Effects 0.000 description 21
- 238000011161 development Methods 0.000 description 12
- 239000002245 particle Substances 0.000 description 9
- 230000003287 optical effect Effects 0.000 description 6
- 230000005294 ferromagnetic effect Effects 0.000 description 5
- 239000000203 mixture Substances 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 239000000428 dust Substances 0.000 description 3
- 230000006698 induction Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000035515 penetration Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000013011 mating Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 239000002893 slag Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000005641 tunneling Effects 0.000 description 1
Landscapes
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention discloses a kind of high precision positioning system and methods, the positioning for cutting head of roadheader during ore extraction.The method includes Magnetic oriented methods, Magnetic oriented method is positioned using machine learning method, machine learning method is optimized, choose optimal performance parameter, to optimize machine learning method, optimization Magnetic oriented model is obtained using the machine learning method of optimization, to improve positioning accuracy, can be realized the tracking and positioning for carrying out real-time high-precision to cutterhead.
Description
Technical field
The present invention relates to a kind of high-accuracy position system and methods.Specifically, the present invention relates to one kind by based on infrared
Positioning and the compound positioning system of cutting head of roadheader and method constituted based on Magnetic oriented.
Background technique
With the development of mining industry, automation, intelligent driving and coal mining are trends of the times, and to cutting head of roadheader
Positioning in real time is the premise of intelligent driving.But a large amount of dust during the work time due to development machine, can be generated, and cutting
Head is buried under slag sometimes, and traditional optical positioning method can not adapt to such adverse circumstances.Utilize infrared positioning
Method can complete high accuracy positioning in the lower situation of dust, but when cutterhead is buried and is blocked by machine body of boring machine, red
Outer localization method failure, the stronger Magnetic oriented of penetration power is exactly a kind of selection.
Although Magnetic oriented has penetration power strong, the advantages of influence by visibility, because of development machine generally iron
Magnetic substance, ferromagnetic environment is very complicated under mine, so classical magnetic field positioning accuracy is very poor.Early stage Magnetic oriented method can return mostly
Become and magnetic source is equivalent to magnetic dipole display or ellipsoid display by each point measured value establishes system of linear equations, then instead solves
The information such as the magnetic moment of magnetic source, position out.But such method is not particularly suited for the environment that irregular magnetic medium is contained in space, and
It is very common for containing irregular magnetic medium in actual environment.Therefore, problems ten are solved in Magnetic oriented practical application
Divide urgent.And combine Magnetic oriented with machine learning algorithm, it can be very good the influence for eliminating ferromagnetics, significantly mention
High position precision.
The Magnetic oriented method combined in the prior art using Magnetic oriented with machine learning algorithm, positioning accuracy is not
Height, and the factor for influencing positioning accuracy is numerous, wherein the selection for the parameter being in the machine learning algorithm used is a need
The factor to be considered, it is improper that parameter is chosen, and just will affect the speed and precision of algorithm, and then influences to calculate by machine learning
The location model structure that method is established, and the location model is used to directly affect the positioning accuracy of magnetic field object when Magnetic oriented.
Summary of the invention
The invention discloses a kind of based on infrared with Magnetic oriented cutting head of roadheader combined positioning method, wherein magnetic field
Positioning optimizes in conjunction with machine learning algorithm, and to machine learning algorithm, obtains high-precision Magnetic oriented model, from
And the precision of Magnetic oriented is improved, and solve the cutterhead orientation problem in the development machine course of work, it is automatic to further realize
Driving provides technical foundation.Positioning system includes the infrared locating module being made of multiple groups thermal camera, by multiple three axis
The Magnetic oriented module of magnetic field sensor composition, the magnetic source for being securable to cutterhead rear, the terminal containing compound location model
Processor.Due to the influence of ferromagnetics development machine body, traditional Magnetic oriented based on Biot-Savart law is simultaneously uncomfortable
With.Magnetic oriented part in this method is first by the cutterhead with fixed permanent magnet in traversal working space, by infrared
Locating module records three-dimensional space position in real time, by the magnetic induction intensity of magnetic field sensor record corresponding position, to obtain training
Then data construct Magnetic oriented model by machine learning algorithm.By the combination of infrared positioning and Magnetic oriented, overcome
Low visibility environment caused by high concentrate dust when development machine work influences and the influence of development machine body ferromagnetics,
Solves the real-time orientation problem of cutting head of roadheader.
Particular technique content provided by the invention is:
A kind of high-precision Magnetic oriented method, the positioning for cutting head of roadheader;The following steps are included:
1), building positioning network, which includes input layer, hidden layer and output layer;According to magnetic field sensor
Number and dimension determine output layer node number, output layer node number;
2), input sample data choose suitable training function, are trained to positioning network, are using by comparing
The size of output layer error, chooses suitable hidden layer section when different hidden layer activation primitives and different node in hidden layer
Point number;
3), after determining output layer node number, output layer node number and hidden layer node number, it is first determined defeated
Layer activation primitive out reuses different hidden layer activation primitives and instructs to sample data after determining output layer activation primitive
Practice, compare using output layer error size when different hidden layer activation primitives and stability, it is small and stable to choose output layer error
When corresponding activation primitive as hidden layer activation primitive;
4), above-mentioned positioning network is trained using above-mentioned trained function, obtains the connection weight w of hidden layerin, output
The connection weight w of layerout, the threshold value b of each node of hidden layerin, the threshold value b of each node of output layerout;Detailed process includes: to calculate
The error of output layer;The error is mean square error;Judge whether the error is less than predictive error;If it is not, then according to this
The new weight of error calculation and threshold value;Utilize the new weight and threshold value, the weight and threshold value of more new definition network;It is described
According to the new weight of the error calculation and threshold value, comprising: obtained new using the error of output layer according to steepest descent method
Output layer weight and threshold value;Output layer error and new weight and threshold value are conducted forward, according to steepest descent
Method obtains the weight and threshold value of new hidden layer using above-mentioned output layer error and the weight and threshold value of new output layer.
5) step 1) -4, is utilized) each node layer number, activation primitive, connection weight and the threshold value that obtain form optimization
Positioning network model, which is used for Magnetic oriented.
A kind of high-accuracy position system, the positioning system include infrared locating module, Magnetic oriented module, are securable to
The magnetic source at cutterhead rear, the terminal handler containing compound location model.The infrared locating module be used for cutterhead into
The infrared positioning of row;The Magnetic oriented module is used to carry out Magnetic oriented to cutterhead;The magnetic source is used for the position of cutterhead
It sets data and passes to Magnetic oriented module.Wherein, the Magnetic oriented module is using above-mentioned Magnetic oriented method to cutterhead
Carry out Magnetic oriented.
Preferably, the infrared locating module includes at least two infrared cameras and matched data collector, institute
It states at least two infrared cameras and is located at cutterhead back upper place, and be placed in parallel, optical center spacing is 25cm.
Preferably, the Magnetic oriented module includes at least two 3 number of axle word magnetic field sensors, is respectively placed in driving
Machine fuselage two sides.
Preferably, magnetic source is fixed on cutterhead rear, moves together with cutterhead, influences the magnetic by the movement of magnetic source
The magnetic field strength that field sensor obtains.The magnetic source is permanent magnet or electromagnet.
Localization method disclosed by the invention are as follows: Magnetic oriented part is traversed by the cutterhead with fixed permanent magnet first
Working space records three-dimensional space position by infrared locating module in real time, by the magnetic induction of magnetic field sensor record corresponding position
Then intensity constructs Magnetic oriented model by machine learning algorithm to obtain training data.Infrared positioning method is based on several
What optical principle, is not provided with infrared light supply, to because of friction heat production and then generating the cutterhead of infra-red radiation in the course of work and carry out
Positioning.By the combination of infrared positioning and Magnetic oriented, the anchor point that infrared positioning is lost, solution are filled up with Magnetic oriented result
The certainly real-time orientation problem of cutting head of roadheader.
For the effective range for guaranteeing fixed-field location model, the effective position range of infrared locating module O is larger than pre-
Phase Magnetic oriented range;Infrared locating module O is made of the infrared camera for being no less than two groups, passes through contour detecting algorithm, mesh
Mark light source decision algorithm, minimum circumscribed circle and the center of circle determine that algorithm and the three-dimensional location based on geometric optics obtain target
Three-dimensional coordinate.
To guarantee enough positioning accuracy, Magnetic oriented module S, by being fixed on development machine independently of cutterhead motion parts
On outer fuselage or follow development machine mobile litter on no less than two triaxial magnetic field sensors, and by infrared fixed
Bit model and magnetic field sensor obtain training data, the Magnetic oriented obtained using machine learning optimization algorithm by training data
Model composition.
Preferably, the optimization algorithm the following steps are included:
Step 1 determines each layer structure of machine learning positioning network.The machine learning positioning network of building includes input
Layer, hidden layer and output layer.
Step 2 initializes the speed of particle, position and individual history is optimal and global optimum.
Step 3 updates particle rapidity and position, calculates particle adaptive value and determines that individual history is optimal and global optimum,
Judge whether global optimum's adaptive value is less than setting accuracy, if so, executing step 5;If it is not, then executing step 4.
Step 4, judges whether the number of iterations is greater than maximum number of iterations;If it is not, then repeating step 2;If
It is then to export global optimum's particle position.
Step 5 utilizes the initial weight and threshold value of above-mentioned calculated result initialization machine learning positioning network.
Step 6 inputs P group sample, calculates each layer output and error.
Step 7 calculates the error of output layer;The error is mean square error;Judge whether the error is less than predetermined mistake
Difference;If it is not, then according to the new weight of the error calculation and threshold value;Using the new weight and threshold value, machine learning is updated
The weight and threshold value of network are positioned, return step six recalculates each layer output and error;It is described new according to the error calculation
Weight and threshold value, comprising: 1) weight of new output layer is obtained according to steepest descent method using the error of output layer
And threshold value;2) output layer error and new weight and threshold value are conducted forward, according to steepest descent method, using above-mentioned defeated
The weight and threshold value of layer error and new output layer out, obtains the weight and threshold value of new hidden layer;In above-mentioned new power
In the calculating of value and threshold value, weight and threshold value are mutually indepedent.
Step 8, judges whether the error is less than predictive error;If it is, saving current machine learning position network
Weight and threshold value;
Step 9, all training of judgement sample whether complete by training, if it is not, the number of iterations adds 1, return step six,
Carry out the training of next sample;If it is, terminating training;To obtain required machine learning location model, it to be used for magnetic
Field positioning.
To guarantee Magnetic oriented precision and effective time, be fixed on cutterhead rear magnetic source M can be used permanent magnet or
Electromagnet generates stronger magnetic field, and can be fixed on cutterhead rear under the premise of not influencing cutterhead work, with cutterhead
It moves together.
To handle each module data, and positioning result is visualized, the terminal handler C containing compound location model is by terminal
Hardware components that host or embedded system and display are constituted and by Magnetic oriented model, infrared location model, infrared fixed
Bit loss decision algorithm, positioning result show that the software section that program is constituted is constituted.
Cutterhead is positioned using compound positioning system of the invention and method, accurately positioning knot can be obtained
Fruit;Referring now to the positioning result of the prior art, positioning accuracy is significantly increased.And do not influenced by operating condition, it can be in complexity
Operating condition under work, be not in cutterhead track lose the case where.
Detailed description of the invention
Fig. 1 positioning system schematic diagram;
Fig. 2 camera coordinates and image coordinate relationship;
The infrared positioning principle schematic diagram of Fig. 3;
Fig. 4 optimization algorithm flow chart;
The mean square error figure of output quantity when Fig. 5 hidden layer activation primitive is y=logsig (x);
The mean square error figure of output quantity when Fig. 6 hidden layer activation primitive is y=tansig (x);
The compound positioning schematic diagram of Fig. 7;
The compound positioning coordinate of Fig. 8 and actual coordinate comparison diagram.
Specific embodiment
With reference to the accompanying drawing, by case study on implementation, the present invention will be described in further detail, but does not limit in any way
The scope of the present invention processed.
Positioning system in the present invention in case study on implementation include the infrared locating module being made of multiple groups thermal camera, by
The Magnetic oriented module of multiple triaxial magnetic field sensor compositions, contains compound positioning mould at the magnetic source for being securable to cutterhead rear
The terminal handler of type.
As shown in Figure 1, infrared locating module is made of two infrared cameras 3 and mating data collector, it is placed in pick
Into machine body upper;Magnetic oriented module is made of two three number of axle word magnetic field sensors 6, is placed in machine body of boring machine two sides;Magnetic
Source 2 is circle D100X20mmN35 neodium magnet, is fixed on the rear of cutting head of roadheader 1;Terminal handler uses desktop computer
7, display screen is placed in front of development machine operator seat.
The present invention provides a kind of based on the infrared and Magnetic oriented compound positioning system of cutting head of roadheader, positioning system packet
Include infrared locating module, Magnetic oriented module and the terminal handler containing compound location model.
In the infrared locating module O, two infrared cameras 3 are placed in parallel, optical center spacing 25cm.By Fig. 2, Fig. 3 institute
Show, it, can be in the hope of object three-dimensional coordinate according to geometric optical theory.
In the Magnetic oriented module S, two three number of axle word magnetic field sensors 6 are respectively placed in machine body of boring machine two sides,
Position does not specially require, but needs to guarantee in training process and position fixing process, and 6 position of magnetic field sensor remains unchanged, can be with
By conventional fixing means, magnetic field sensor 6 is fixed on to the two sides of machine body of boring machine, as development machine moves together,
And it keeps invariable with the relative position of fuselage.
Magnetic source M is circle D100X20mmN35 neodium magnet 2, is fixed on cutterhead rear.Notice that magnetic source M cannot influence cutting
Head work.And in the training process and in position fixing process, magnetic source M need to be consistent.Due to the movement of magnetic source M (i.e. magnet 2)
The movement of cutterhead 1 is represented, i.e. magnetic source M is moved with cutterhead 1, and the motion profile of magnetic source M is exactly the fortune of cutterhead 1
Dynamic rail mark (in cutterhead movement here, it should using cutterhead as a particle, rather than turn), therefore magnetic source M is fixed
Position need to be remained unchanged relative to cutterhead.
For terminal handler for handling position data, the position data includes the cutting that infrared camera obtains
The magnetic field position data for the cutterhead that the infrared position data and Magnetic oriented module of head obtain.In terminal handler,
The position that processing obtains cutterhead is carried out to infrared position data by infrared positioning method, by Magnetic oriented method to magnetic
Field position data carries out processing and obtains the position of cutterhead, and the position of cutterhead is shown.
It is the concrete composition of tunneling machine cutting head positioning system of the invention above.It is described in detail below and of the invention determines
Position method.
Localization method is divided into Magnetic oriented, infrared positioning and infrared and three parts of the compound positioning in magnetic field.
Infrared positioning method is introduced first.
Infrared positioning is based on geometric optical theory, can be in the hope of target by the aberration of analysis two infrared cameras imaging
Object three-dimensional coordinate, as shown in Figure 2,3.Above-mentioned infrared positioning method the following steps are included:
Step 1: establishing camera coordinates system and image coordinate system.For a camera, from camera coordinates system to image
Coordinate system is considered as perspective projection relationship, as shown in Figure 2;Camera coordinates system is denoted as OC-XCYCZC, image coordinate system is denoted as o-
xyz;
Step 2: indicating image coordinate with camera coordinates and camera focus according to similar triangles knowledge.Such as Fig. 2, according to
Known to similar triangles knowledge: AB/oC=AOC/oOC=PB/pc=XC/ x=ZC/ f=Yc/y;It can thus be concluded that:
X=f*XC/ZC, (1)
Y=f*YC/ZC, (2)
Step 3: world coordinate system is established, in world coordinate system using the midpoint of the two image center lines in left and right as origin
The middle left camera coordinates of determination, right camera coordinates, left image coordinate and right image coordinate;And the world is indicated with left and right camera coordinates
Point coordinate in coordinate system.
Two video camera relationships of left and right are as shown in Figure 3.If two camera spacing b, world coordinate system origin is in two camera centers
Line center, the world coordinates of measurement point are denoted as (X, Y, Z).The direction XYZ and XCYCZCIt is identical.Left camera coordinates are denoted as (XCl,
YCl, ZCl), right camera coordinates are denoted as (XCr, YCr, ZCr).Left image coordinate is denoted as (xl, yl, zl), right image coordinate is denoted as (xr,
yr, zr);The coordinate of above-mentioned measurement point and the coordinate of left and right camera have following relationship:
X=XCl+ b/2=XCr- b/2, (3)
Y=YCl=YCr, (4)
Z=ZCl=ZCr, (5)
Step 4: combining the space bit for acquiring measurement point by the point coordinate in the image coordinate and step 4 in step 2
Coordinate is set, thus to obtain the spatial position of measurement point.
By (1), (3) Shi Ke get:
xl*Z-f*(XCr- b)=0, (6)
xr*Z-f*XCr=0, (7)
It can be solved by (6) (7) formula:
Z=-f*b/ (xl-xr), (8)
XCr=xr*(-f*b/(xl-xr))/f, (9)
It can be obtained by (3) (9)
X=xr*(-f*b/(xl-xr))/f-b/2 (10)
By (2) (4) (8) Shi Ke get:
Y=yr*(-f*b/(xl-xr))/f (11)
Next Magnetic oriented method is introduced.
Magnetic oriented needs training location model in advance, passes through the cutterhead with fixed permanent magnet first in the present embodiment
In traversal working space, three-dimensional space position is constantly recorded by infrared locating module, by magnetic field sensor record corresponding position
Magnetic induction intensity, to obtain training data.Because the ferromagnetics such as machine body of boring machine influence, magnetic field is distributed in space and injustice
It is sliding, so regular machinery learning position network algorithm convergence rate is very slow, and it is easily trapped into local minimum.Therefore, of the invention
Network is positioned to machine learning by following optimization algorithm to optimize, and Magnetic oriented model is constructed by training data.
Firstly, introducing the mathematical model and algorithm principle of machine learning network.First variable is defined: three-decker
Machine learning network include input layer, hidden layer and output layer, the variable which is related to is as follows:
Input variable x (x1,x2,…xn);
Hidden layer input variable hin=(hin,1,hin,2,…,hin,p);Hidden layer output variable hout=(hout,1,hout
,2,…,hout,p);
Output layer input variable yin=(yin,1,yin,2,…,yin,m);Output layer output variable yout=(yout,1,
yout,2,…,yout,m);
Desired output vector do=(d1,d2,…,dm)
The connection weight w of input layer and hidden layerin;The threshold value b of each node of hidden layerin;
The connection weight w of hidden layer and output layerout;The threshold value b of each node of output layerout;
Error functionActivation primitive f ()
Wherein n, m, p, k are positive integer.
1, then need to carry out netinit by calculating: computer is respectively in one section (- 1,1) of each connection weight
Random number, set error function E, give computational accuracy value ε and maximum study number M.
Next it randomly selects calculating sample: it is (the number of iterations) and corresponding to randomly select the t times input sample training
Desired output:
X=(x1,x2,…xn)
do=(d1,d2,…,dm)
2, hidden layer and output layer data are calculated:
1) outputting and inputting for each network node of hidden layer is calculated:
2) outputting and inputting for each network node of output layer is calculated:
3, followed by amendment weight:
The error function of whole network is
Connection weight, the improvement formula of the connection weight are improved according to the negative gradient of error function E are as follows:
Specifically: the connection weight more new formula for k-th of network node of output layer is
In above formula, η indicates learning rate, δkThe referred to as learning error of k-th of network node of output layer.
To the more new formula of hidden layer network node connection weight are as follows:
In above formula, σkReferred to as the learning error of each node of hidden layer of k-th of node of output layer;Machine learning net
Network be it is a kind of according to error backpropagation algorithm training Multi-layered Feedforward Networks, by information forward-propagating and error it is reversed
Propagate two process compositions.Wherein, the error back propagation thinking of machine learning network are as follows: estimated using the error of output layer
The error for counting output layer weight updates output layer weight according to steepest descent principle, then output layer error and newly
Weight conduct forward, calculate upper one layer of error, and update weight until output data and expected data meet error requirements,
Then deconditioning.
Next the step of introducing the optimization algorithm that the present invention utilizes.
Above-mentioned optimization algorithm flow chart is as shown in Fig. 4.The optimization algorithm the following steps are included:
Step 1 determines each layer structure of machine learning positioning network.The machine learning positioning network of building includes input
Layer, hidden layer and output layer.
Step 2 initializes the speed of particle, position and individual history is optimal and global optimum.
Step 3 updates particle rapidity and position, calculates particle adaptive value and determines that individual history is optimal and global optimum,
Judge whether global optimum's adaptive value is less than setting accuracy, if so, executing step 5;If it is not, then executing step 4.
Step 4, judges whether the number of iterations is greater than maximum number of iterations;If it is not, then repeating step 2;If
It is then to export global optimum's particle position.
Step 5 utilizes the initial weight and threshold value of above-mentioned calculated result initialization machine learning positioning network.
Step 6 inputs P group sample, calculates each layer output and error.
Step 7 calculates the error of output layer;The error is mean square error;Judge whether the error is less than predetermined mistake
Difference;If it is not, then according to the new weight of the error calculation and threshold value;Using the new weight and threshold value, machine learning is updated
The weight and threshold value of network are positioned, return step six recalculates each layer output and error;It is described new according to the error calculation
Weight and threshold value, comprising: 1) weight of new output layer is obtained according to steepest descent method using the error of output layer
And threshold value;2) output layer error and new weight and threshold value are conducted forward, according to steepest descent method, using above-mentioned defeated
The weight and threshold value of layer error and new output layer out, obtains the weight and threshold value of new hidden layer;In above-mentioned new power
In the calculating of value and threshold value, weight and threshold value are mutually indepedent.
Step 8, judges whether the error is less than predictive error;If it is, saving current machine learning position network
Weight and threshold value;
Step 9, all training of judgement sample whether complete by training, if it is not, the number of iterations adds 1, return step six,
Carry out the training of next sample;If it is, terminating training;To obtain required machine learning location model, it to be used for magnetic
Field positioning.
It is the detailed process of Magnetic oriented method above, network is positioned by machine learning, using optimization algorithm, thus
The position that target object can accurately be obtained can be used for the positioning of target object when infrared locating effect is poor.
It is described above and machine learning positioning network is optimized using optimization algorithm.Below to the machine learning of optimization
Each important parameter is determined in positioning network, to form stable machine learning location model, is used for Magnetic oriented, is obtained
Obtain stable high-precision locating effect.The forward-propagating mathematical expression of machine learning positioning network are as follows:
Y=F(Xwin+bin)wout+bout
Wherein, Y is output quantity, and F is training function, and X is input variable, win,wout, bin,boutIt is as defined above text.On
Stating machine learning positioning network includes input layer, hidden layer and output layer, and input variable inputs hidden layer and forms hidden layer input
Variable, as shown in formula (12), input variable is by obtaining hidden layer input variable after hidden layer connection weight and threshold function;
Hidden layer input variable forms hidden layer output variable by way of formula (13), specifically, hidden layer input variable passes through
Hidden layer output variable is formed after the conduction of hidden layer activation primitive;Similarly, hidden layer output variable input and output layer forms output
Layer input variable, as shown in formula (14), hidden layer output variable after output layer connection weight and threshold function by being exported
Layer input variable;Output layer input variable forms output layer output variable by way of formula (15), specifically, output layer
Input variable forms output layer output variable after conducting by output layer activation primitive;In conclusion above-mentioned machine learning positioning
Connection is established by connection weight and threshold value between each layer of network, and the input variable and output variable in same layer are by being somebody's turn to do
The activation primitive of layer establishes connection.By analyzing above, in above-mentioned machine learning positioning network, the ginseng of positioning accuracy is influenced
Number includes: the connection weight w of input layer and hidden layerin;The threshold value b of each node of hidden layerin;The connection of hidden layer and output layer
Weight wout;The threshold value b of each node of output layerout;The activation primitive f of hidden layer and output layer, and training function F.Wherein,
Weight win,woutWith threshold value bin,boutIt is constantly modified by the error negative feedback process of network, until the output of output layer
The error of variable and desired value is less than predictive error, to stop the training of machine learning positioning network, the positioning network of acquisition
Model is used for Magnetic oriented;Above-mentioned weight win,woutWith threshold value bin,boutOptimization algorithm process described above,
This is repeated no more.
The activation primitive f of hidden layer and output layer, and the determination of training function F is described in detail below, and thereby determines that
Input layer, hidden layer, output layer number of nodes optimal number.Input layer and output node layer are tieed up by the input of magnetic field sensor
Degree and output dimension determine, for example acquire data using 4 sensors, to obtain two-dimensional coordinate of the magnetic source on experimental bench,
So input number of nodes is 12, and output node number is 2.The most commonly used is trial and error procedures for the number of hidden nodes at present, and are arranged how many hidden
Node layer is related to complex relationship in the quantity of training sample and sample.It is common hidden in machine learning network theory knowledge
Activation primitive containing layer includes mainly function y=logsig (x) and function y=tansig (x);Firstly, with function y=logsig
(x) and function y=tansig (x) is respectively hidden layer activation primitive, chooses different node in hidden layer, real by simulation
It tests, when calculating different number of nodes, whole network output layer error, the error is mean square error MSE, experimental result such as Fig. 5, Fig. 6
The corresponding mean square error MSE of different number of nodes shown, that Fig. 5 is hidden layer activation primitive when being y=logsig (x);Fig. 6 is
The corresponding mean square error MSE of different number of nodes when hidden layer activation primitive is y=tansig (x);It can be obtained from Fig. 5, Fig. 6
Out, when node in hidden layer is 12, mean square error is minimum, it is possible thereby to which it is 12 that the number of nodes of determining hidden layer is optimal.Hidden
After number of nodes containing layer determines, then determine hidden layer activation primitive.Under different node in hidden layer, by comparing different hidden
In the case where activation primitive containing layer, the size of output layer error, it is possible thereby to determine that discovery is based on activation primitive y=logsig
(x) network model established is more stable under different hidden nodes, and therefore, optimal hidden layer activation primitive is function y=
logsig(x).It is 12 in input layer number, node in hidden layer 12, output layer number of nodes is 2, hidden layer activation primitive
In the case where y=logsig (x), simulation calculates the error amount that different output layer activation primitives obtain, when choosing error minimum
Activation primitive of the corresponding activation primitive as output layer, can be obtained by simulated experiment, and output layer activation primitive is purelin
Function.Wherein, network training function uses Levenberg-Marquardt function, which overcomes base to a certain extent
The problems such as this network convergence rate is slow and is easily trapped into local minimum points.The training function is the training in MATLAB software
Training function in function tool box, details are not described herein.
Combined positioning method includes combining infrared positioning with Magnetic oriented as a result, as shown in fig. 7, when infrared positioning result is lost
When mistake or track are obviously discontinuous, infrared positioning result is replaced to be output to display screen using Magnetic oriented result.
The present invention development machine simulation practical work process in positioning result as shown in figure 8, as shown in Figure 8, utilization is above-mentioned
The motion profile of target object that combined positioning method obtains coincide well with its practical motion profile, and error rate can be with
Control is in the error range of setting.
It should be noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but the skill of this field
Art personnel, which are understood that, not to be departed from the present invention and spirit and scope of the appended claims, and various substitutions and modifications are all
It is possible.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is wanted with right
Subject to the range for asking book to define.
Claims (5)
1. a kind of high-precision Magnetic oriented method, the positioning for cutting head of roadheader;It is characterized by comprising following steps:
1), building positioning network, which includes input layer, hidden layer and output layer;According to the number of magnetic field sensor
And dimension determines output layer node number, output layer node number;
2), input sample data are chosen suitable training function, are trained to positioning network, by comparing using different
The size of output layer error, chooses suitable hidden layer node when hidden layer activation primitive and different node in hidden layer
Number;
3), after determining output layer node number, output layer node number and hidden layer node number, it is first determined output layer
Activation primitive reuses different hidden layer activation primitives and is trained to sample data after determining output layer activation primitive, than
Compared with output layer error size and stability when using different hidden layer activation primitives, selection output layer error is small and corresponding when stablizing
Activation primitive as hidden layer activation primitive;
4), above-mentioned positioning network is trained using above-mentioned trained function, obtains the connection weight w of hidden layerin, output layer
Connection weight wout, the threshold value b of each node of hidden layerin, the threshold value b of each node of output layerout;Detailed process includes: to calculate output
The error of layer;The error is mean square error;Judge whether the error is less than predictive error;If it is not, then according to the error
Calculate new weight and threshold value;Utilize the new weight and threshold value, the weight and threshold value of more new definition network;The basis should
The new weight of error calculation and threshold value, comprising: new output is obtained according to steepest descent method using the error of output layer
The weight and threshold value of layer;Output layer error and new weight and threshold value are conducted forward, according to steepest descent method, in utilization
The weight and threshold value for stating output layer error and new output layer obtain the weight and threshold value of new hidden layer.
5) step 1) -4, is utilized) each node layer number, activation primitive, connection weight and the threshold value that obtain form determining for optimization
Position network model, is used for Magnetic oriented for the positioning network model.
2. a kind of high-accuracy position system, it is characterised in that: the positioning system includes infrared locating module, Magnetic oriented mould
Block, the magnetic source for being securable to cutterhead rear, the terminal handler containing compound location model.The infrared locating module is used for
Infrared positioning is carried out to cutterhead;The Magnetic oriented module is used to carry out Magnetic oriented to cutterhead;The magnetic source is used for will
The position data of cutterhead passes to Magnetic oriented module.Wherein, the Magnetic oriented module utilizes magnetic described in claim 1
Field localization method carries out Magnetic oriented to cutterhead.
3. positioning system according to claim 2, which is characterized in that the Magnetic oriented module includes at least two 3 axis
Magnetic field sensor is respectively placed in machine body of boring machine two sides.
4. positioning system according to claim 2 or 3, which is characterized in that the magnetic source is fixed on cutterhead rear, with cut
First movement is cut, the magnetic field strength that the magnetic field sensor obtains is influenced by the movement of magnetic source.
5. positioning system according to claim 3 or 4, which is characterized in that the magnetic source is permanent magnet or electromagnet.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910118352.XA CN109709497A (en) | 2019-02-16 | 2019-02-16 | A kind of high-accuracy position system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910118352.XA CN109709497A (en) | 2019-02-16 | 2019-02-16 | A kind of high-accuracy position system and method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109709497A true CN109709497A (en) | 2019-05-03 |
Family
ID=66264480
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910118352.XA Pending CN109709497A (en) | 2019-02-16 | 2019-02-16 | A kind of high-accuracy position system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109709497A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110513120A (en) * | 2019-08-17 | 2019-11-29 | 冀中能源峰峰集团有限公司 | A kind of cutting head of roadheader adaptive location system and method |
CN112036073A (en) * | 2020-07-16 | 2020-12-04 | 成都飞机工业(集团)有限责任公司 | 3D printing part measurement result correction method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4317078A (en) * | 1979-10-15 | 1982-02-23 | Ohio State University Research Foundation | Remote position and orientation detection employing magnetic flux linkage |
CN101695190A (en) * | 2009-10-20 | 2010-04-14 | 北京航空航天大学 | Three-dimensional wireless sensor network node self-locating method based on neural network |
CN102741653A (en) * | 2009-11-24 | 2012-10-17 | 诺基亚公司 | Installation of magnetic signal sources for positioning |
CN103728431A (en) * | 2014-01-09 | 2014-04-16 | 重庆科技学院 | Industrial sewage COD (chemical oxygen demand) online soft measurement method based on ELM (extreme learning machine) |
CN105919595A (en) * | 2016-05-17 | 2016-09-07 | 浙江大学宁波理工学院 | System and method for tracking miniature device with magnetic signals in body of moving object |
CN108871375A (en) * | 2018-04-24 | 2018-11-23 | 北京大学 | A kind of calibration system and method for three-dimensional space magnetic orientation system |
-
2019
- 2019-02-16 CN CN201910118352.XA patent/CN109709497A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4317078A (en) * | 1979-10-15 | 1982-02-23 | Ohio State University Research Foundation | Remote position and orientation detection employing magnetic flux linkage |
CN101695190A (en) * | 2009-10-20 | 2010-04-14 | 北京航空航天大学 | Three-dimensional wireless sensor network node self-locating method based on neural network |
CN102741653A (en) * | 2009-11-24 | 2012-10-17 | 诺基亚公司 | Installation of magnetic signal sources for positioning |
CN103728431A (en) * | 2014-01-09 | 2014-04-16 | 重庆科技学院 | Industrial sewage COD (chemical oxygen demand) online soft measurement method based on ELM (extreme learning machine) |
CN105919595A (en) * | 2016-05-17 | 2016-09-07 | 浙江大学宁波理工学院 | System and method for tracking miniature device with magnetic signals in body of moving object |
CN108871375A (en) * | 2018-04-24 | 2018-11-23 | 北京大学 | A kind of calibration system and method for three-dimensional space magnetic orientation system |
Non-Patent Citations (1)
Title |
---|
王杰等: "一种基于粒子群优化的极限学习机", 《郑州大学学报(理学版)》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110513120A (en) * | 2019-08-17 | 2019-11-29 | 冀中能源峰峰集团有限公司 | A kind of cutting head of roadheader adaptive location system and method |
CN110513120B (en) * | 2019-08-17 | 2020-12-29 | 冀中能源峰峰集团有限公司 | Self-adaptive positioning system and method for cutting head of heading machine |
CN112036073A (en) * | 2020-07-16 | 2020-12-04 | 成都飞机工业(集团)有限责任公司 | 3D printing part measurement result correction method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106557844B (en) | Path planning method for welding robot | |
Bonsignorio et al. | Toward replicable and measurable robotics research [from the guest editors] | |
CN109767477A (en) | A kind of Precise Position System and method | |
CN109163722B (en) | Humanoid robot path planning method and device | |
CN103414853B (en) | Support the sequence of video images real-time stabilization apparatus and method of multifreedom motion | |
CN109709497A (en) | A kind of high-accuracy position system and method | |
CN108319293A (en) | A kind of UUV Realtime collision free planing methods based on LSTM networks | |
Huang et al. | FAEL: fast autonomous exploration for large-scale environments with a mobile robot | |
CN210033427U (en) | Compound positioning system of heading machine cutterhead | |
Zhang et al. | ST-VIO: Visual-inertial odometry combined with image segmentation and tracking | |
CN112182955A (en) | Construction method and application of shield general type segment point selection model | |
Tang et al. | Hierarchical and controlled advancement for continuous collision detectionof rigid and articulated models | |
CN109555521A (en) | A kind of cutting head of roadheader combined positioning method | |
CN109538208A (en) | A kind of compound positioning system of cutting head of roadheader and method | |
Du et al. | Industrial robot digital twin system motion simulation and collision detection | |
Liu et al. | Evaluation and comparison of Gmapping and Karto SLAM systems | |
Susanto et al. | XNOR-YOLO: the high precision of the ball and goal detecting on the Barelang-FC robot soccer | |
CN113656918B (en) | Four-rotor simulation test method applied to finished product overhead warehouse scene | |
Gao et al. | A novel local path planning method considering both robot posture and path smoothness | |
CN113848750A (en) | Two-wheeled robot simulation system and robot system | |
CN111435536A (en) | Accurate positioning system | |
Chen et al. | What should be the input: Investigating the environment representations in sim-to-real transfer for navigation tasks | |
Wang et al. | Unmanned Aerial Vehicle Autonomous Visual Landing through Visual Attention-Based Deep Reinforcement Learning | |
Lei et al. | An Improved Multi-objective Path Planning Algorithm and its Application to Tennis Ball Collection Robots | |
Gabas et al. | Ape: A more practical approach to 6-dof pose estimation |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190503 |
|
WD01 | Invention patent application deemed withdrawn after publication |