CN105242536A - Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network - Google Patents

Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network Download PDF

Info

Publication number
CN105242536A
CN105242536A CN201510613438.1A CN201510613438A CN105242536A CN 105242536 A CN105242536 A CN 105242536A CN 201510613438 A CN201510613438 A CN 201510613438A CN 105242536 A CN105242536 A CN 105242536A
Authority
CN
China
Prior art keywords
destination
neural network
represent
bp neural
aerial vehicle
Prior art date
Application number
CN201510613438.1A
Other languages
Chinese (zh)
Inventor
苏寒松
张永振
刘高华
Original Assignee
天津大学
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 天津大学 filed Critical 天津大学
Priority to CN201510613438.1A priority Critical patent/CN105242536A/en
Publication of CN105242536A publication Critical patent/CN105242536A/en

Links

Abstract

The invention discloses an unmanned aerial vehicle driving route waypoint calibration method based on a BP nerve network. The method comprises the following steps: randomly generating 1000 three-dimensional coordinate points; establishing a matrix P of a BP nerve network input layer; listing a BP nerve network output end matrix T; establishing the BP nerve network; training the BP nerve network; and fitting a waypoint route of an unmanned aerial vehicle. Compared to the prior art, the delimited mark quantity of the unmanned aerial vehicle is increased, the mark route of the unmanned aerial vehicle is directive, an unmanned aerial vehicle automatic pilot system can be improved, a BP nerve network technology is innovatively applied, the design of the waypoint route at a ground station of the unmanned aerial vehicle can be rapidly and efficiently realized based on this, and the method provided by the invention can be applied to an autopilot; and as a new nerve network based autopilot waypoint route scheme, the method has great development prospect in research on a future unmanned aerial vehicle autopilot.

Description

Based on the unmanned plane drive route destination scaling method of BP neural network

Technical field

The present invention relates to unmanned plane automatic Pilot technical field, particularly relate to a kind of destination scaling method of unmanned plane drive route.

Background technology

In recent years, the development of unmanned plane is swift and violent especially.Unmanned plane usable range is expanded to the civilian and multiple field of scientific research from military affairs.In military affairs, unmanned plane can be used for investigation supervision, trunking traffic, electronic countermeasure, combat success assessment, (sea) attack, early warning etc. over the ground; On civilian, for the experimental verification etc. of atmospheric research, meteorological observation and new technology and equipment.

In mid-April, 2010, AUS formally discloses the UAS way for development line chart of 2010 to 2035: the immediate objective of ground force is realize helicopter unmanned, namely novel carrier-borne vertical helicopter landing TUAV is equipped, to fill up current urgent war demand.About the research of unmanned plane robot pilot, also grow in intensity.The HcliAP depopulated helicopter robot pilot of the Canada's depopulated helicopter robot pilot MP2128HELI of MicroPilot company, the wcPilot1000 Miniature Unmanned Helicopter robot pilot of WcControl company of Switzerland and California, USA science and engineering research and development is all typical representative.

At present, domestic unmanned plane robot pilot research level is also in the imitated stage, and overall research and development are still in the starting stage.

BP neural network belongs to feedforward network, is the core of neural network, is also the elite in whole neural network system, and it is different from multilayer perceptron simultaneously.

Summary of the invention

For above-mentioned prior art and Problems existing, the present invention proposes a kind of unmanned plane drive route destination scaling method based on BP neural network, apply BP nerual network technique innovatively in this field, the unmanned plane during flying route realizing standing in for unmanned aerial vehicle destination on map is demarcated.

The present invention proposes a kind of unmanned plane drive route destination scaling method based on BP neural network, comprise the following steps:

Step 1: random generation 1000 three-dimensional coordinate points, for simulating 1000 destinations on land station's map;

Step 2: the matrix P setting up BP neural network input layer;

P = a 11 a 12 a 13 ... a 1999 a 11000 a 21 a 22 a 23 ... a 2999 a 21000 a 31 a 32 a 33 ... a 3999 a 31000

Wherein, (a 11, a 21, a 31) represent the 1st destination, (a 21, a 22, a 32) represent the 2nd destination, by that analogy, (a 1999, a 2999, a 3999) represent the 999th destination, (a 11000, a 21000, a 31000) represent the 1000th destination.

Step 3: set up BP neural network output terminal matrix T;

T = b 11 b 12 b 13 ... b 1999 b 11000 b 21 b 22 b 23 ... b 2999 b 21000 b 31 b 32 b 33 ... b 3999 b 31000

Wherein, (b 11, b 21, b 31) represent that system exports the 1st destination of process, (b 12, b 22, b 32) represent that system exports the 2nd destination of process, by that analogy, (b 1999, b 2999, b 3999) represent that system exports the 999th destination of process, (b 11000, b 21000, b 31000) represent that system exports the 1000th destination of process.Meanwhile, matrix P is equal with matrix T, i.e. P=T.

Step 4: set up BP neural network, net=netff (P, T, 3), wherein, P, T represent BP neural network input and output matrix respectively, and 3 represent that the BP neural network of design has a hidden layer, and hidden layer neuron number is 3;

Step 5: train this BP neural network, [net, tr]=train (net, P, T), use input, output matrix are trained BP neural network net, obtain new neural network net, tr is for recording step number epoch and the performance perf of training simultaneously;

Step 6: for the Output matrix of BP neural network, uses plot3 function to carry out line to the way point representated by matrix, generates the destination route of unmanned plane.

Compared with prior art, unmanned plane can be delimited the increase of punctuate quantity by the present invention, unmanned plane punctuate route has directivity and can improve unmanned plane autopilot system, apply BP nerual network technique innovatively, and fast and effeciently realize the design of unmanned aerial vehicle station to destination route on this basis, can be applied on the robot pilot of unmanned plane;

As the robot pilot destination route plan based on neural network of novelty, in unmanned plane robot pilot research afterwards, there is great development prospect.

Accompanying drawing explanation

Fig. 1 is the demarcation route schematic diagram under 100 destinations;

Fig. 2 is the error performance curve synoptic diagram under 100 destinations between actual flight route and design route;

Wherein, Train, Validation, Test represent result performance, the result performance of inspection, the result performance of checking of training respectively.Performance represents system performance, and Epochs represents the step number of systematic training.As can be seen from the figure, three curves are when 1000 step, and error all have decreased to 10 -5(percent value).

Fig. 3 is the demarcation route schematic diagram under 1000 destinations;

Fig. 4 is the error performance curve synoptic diagram under 1000 destinations between actual flight route and design route;

Wherein, Train, Validation, Test, Best represent the performance under the result performance of the result performance of training, the result performance verification of inspection, system optimum condition respectively.Performance represents system performance, and Epochs represents the step number of systematic training.BestValidationPerformanceis0.00046723atepoch1000 represent training to best test performance during 1000 step be 0.00046723.Meanwhile, system performance least mean-square error represents MSE (MeanSquaredError).Can see, in figure 1. 2. 3. bar curve have overlap, illustrate that training, inspection, checking, best result performance are similar.

Fig. 5 is overall flow figure of the present invention.

Embodiment

Below in conjunction with the drawings and the specific embodiments, be described in further detail technical scheme of the present invention.

The route destination of common unmanned plane automatic Pilot is demarcated all within 100, the present invention is based on BP neural network (ErrorBackPropagation, i.e. error backpropagation algorithm neural network) and destination quantity can be brought up to 1000.Although use BP neural network to carry out the training need regular hour of flight path, the present invention utilizes based on this innovative point of BP neural network, and demarcating degree of accuracy can reach 10 -4the order of magnitude (the ratio size of error amount and actual value).Meanwhile, when using BP neural network, the direction of destination can be demarcated, thus judge that whether unmanned plane during flying direction is correct.When unmanned plane flies between two destinations, utilize calibration result of the present invention can provide error between actual flight route and design route, if error is excessive, so system will provide alarm.Like this, be conducive to improving unmanned plane robot pilot further.

Emulation of the present invention is carried out under MATLAB:

First the randi function in MATLAB is utilized to generate random three-dimensional destination data (have employed one group of 100 destination respectively and 1000 destinations of the present invention contrast); Limit data area, simulate aerial three-dimensional point, and make destination reach 1000.

Secondly, set up BP neural network and (adopt the neural network of three layers, it comprises input layer, output layer and a hidden layer), hidden layer adopts tansig function, output layer adopts purelin function, weights/threshold learning the function of BP network is learngdm, and its performance function is MSE (MeanSquareError), namely uses mean square error function as the index of performance evaluation;

Finally, carry out curve fitting for above destination, form the flight path of unmanned plane.

BP neural network is used for training by 60% of input data, and 20% for inspection, and 20% for checking, and have employed the strategy of premature termination, prevent the situation of over-fitting from occurring, degree of accuracy increases.As shown in Figure 2 and Figure 4, sets forth the error performance curve synoptic diagram between actual flight route and design route under error performance curve synoptic diagram under 00 destination between actual flight route and design route and 1000 destinations, can see, error is all 10 -4under.

When using MATLAB emulation, adopt " > " with the direction of the trend pointed the route with each destination.Like this, unmanned plane is unlikely to when flying and advancing along route flight in the other direction.In the process using BP neural fusion unmanned plane punctuate route, can direction between given any two punctuates, namely which punctuate unmanned plane first passes through, after through which punctuate.

When unmanned plane during flying, its real-time flight data can be observed in land station, by calculating the distance in actual flight route between current location point and two destination straight lines, i.e. BP neural fusion unmanned plane punctuate route, when unmanned plane practical flight position and the vertical range of the route gone out through neural metwork training over-fitting reach setting value, system can provide alarm, the error of unmanned plane current flight state can be passed judgment on, to make unmanned plane robot pilot make rapid adjustment, reduce error.This is a kind of judgment of error method, can also have the method for other equivalence.

Such as, unmanned plane is in the airflight of three dimensions sky, and corresponding unmanned plane any point is as destination respectively, sets the height of its three-dimensional coordinate (x, y, z), latitude, longitude and distance reference plane.These reference planes can set according to actual conditions.Latitude scope is (-90 °, 90 °), and negative value represents south latitude, on the occasion of expression north latitude; Longitude range is (-180 °, 180 °), and negative value represents west longitude, on the occasion of expression east longitude; Altitude range is (0,10000), and unit is rice, and when generating altitude information, employing maximal value is ten thousand metres.At present, unmanned plane on the market supports 50,100 destinations mostly.The present invention trains BP neural network on the basis of 1000 destinations, reaches the object of setting-out.

For above 1000 destinations, set up BP neural network, hidden layer adopts tansig function, output layer adopts purelin function, the training function of BP network is trainlm, weights/threshold learning the function of BP network is learngdm, and its performance function is MSE (MeanSquareError), and namely square error is as error performance function.

On the basis of trained rear neural network, input 1000 destinations, make network correspondence export 1000 and sit destination.

Sit destination for these 1000, carry out curve fitting, form the flight path of unmanned plane.

Under 1000 destinations as shown in Figure 3 demarcation route schematic diagram, its scaling method comprises the following steps:

Step 1: random generation 1000 three-dimensional coordinate points, for simulating 1000 destinations on land station's map;

Step (2), sets up the matrix P of BP neural network input layer;

P = a 11 a 12 a 13 ... a 1999 a 11000 a 21 a 22 a 23 ... a 2999 a 21000 a 31 a 32 a 33 ... a 3999 a 31000

Wherein, (a 11, a 21, a 31) represent the 1st destination, (a 21, a 22, a 32) represent the 2nd destination, by that analogy, (a 1999, a 2999, a 3999) represent the 999th destination, (a 11000, a 21000, a 31000) represent the 1000th destination.

Step (3), lists BP neural network output terminal matrix T;

T = b 11 b 12 b 13 ... b 1999 b 11000 b 21 b 22 b 23 ... b 2999 b 21000 b 31 b 32 b 33 ... b 3999 b 31000

Wherein, (b 11, b 21, b 31) represent that system exports the 1st destination of process, (b 12, b 22, b 32) represent that system exports the 2nd destination of process, by that analogy, (b 1999, b 2999, b 3999) represent that system exports the 999th destination of process, (b 11000, b 21000, b 31000) represent that system exports the 1000th destination of process.Meanwhile, matrix P is equal with matrix T, i.e. P=T.

Step (4), sets up BP neural network, net=netff (P, T, 3), and wherein, P, T represent BP neural network input and output matrix respectively, and 3 represent that the BP neural network of design has a hidden layer, and hidden layer neuron number is 3;

Step (5), trains this BP neural network, [net, tr]=train (net, P, T), uses input, output matrix pair

BP neural network net trains, and obtains new neural network net simultaneously, and tr is for recording step number epoch and the performance perf of training;

Step (6), for the Output matrix of BP neural network, uses plot3 function to the way point representated by matrix

Carry out line, generate the destination route of unmanned plane.

As shown in Figure 4, be the error performance under 1000 destinations, said method comprising the steps of:

Step (7): on the basis of step (1), step (2), step (3), step (4), step (5), step (6), use MSE method, estimating system error performance, namely uses plotperf () function.

Demarcation route schematic diagram as shown in Figure 1, its scaling method and error performance computing method the same, random produce 100 three-dimensional destinations, for simulating 100 destinations on land station's map; Actual flight route under 100 destinations as shown in Figure 2 and the error performance between design route.

Claims (1)

1., based on a unmanned plane drive route destination scaling method for BP neural network, it is characterized in that, the method comprises the following steps:
Step (1), random generation 1000 three-dimensional coordinate points, for simulating 1000 destinations on land station's map;
Step (2), sets up the matrix P of BP neural network input layer;
P = a 11 a 12 a 13 ... a 1999 a 11000 a 21 a 22 a 23 ... a 2999 a 21000 a 31 a 32 a 33 ... a 3999 a 31000
Wherein, (a 11, a 21, a 31) represent the 1st destination, (a 21, a 22, a 32) represent the 2nd destination, by that analogy, (a 1999, a 2999, a 3999) represent the 999th destination, (a 11000, a 21000, a 31000) represent the 1000th destination.
Step (3), lists BP neural network output terminal matrix T;
T = b 11 b 12 b 13 ... b 1999 b 11000 b 21 b 22 b 23 ... b 2999 b 21000 b 31 b 32 b 33 ... b 3999 b 31000
Wherein, (b 11, b 21, b 31) represent that system exports the 1st destination of process, (b 12, b 22, b 32) represent that system exports the 2nd destination of process, by that analogy, (b 1999, b 2999, b 3999) represent that system exports the 999th destination of process, (b 11000, b 21000, b 31000) represent that system exports the 1000th destination of process.Meanwhile, matrix P is equal with matrix T, i.e. P=T.
Step (4), sets up BP neural network, net=netff (P, T, 3), and wherein, P, T represent BP neural network input and output matrix respectively, and 3 represent that the BP neural network of design has a hidden layer, and hidden layer neuron number is 3;
Step (5), trains this BP neural network, [net, tr]=train (net, P, T), use input, output matrix are trained BP neural network net, and obtain new neural network net, tr is for recording step number epoch and the performance perf of training simultaneously;
Step (6), for the Output matrix of BP neural network, uses plot3 function to carry out line to the way point representated by matrix, generates the destination route of unmanned plane.
CN201510613438.1A 2015-09-22 2015-09-22 Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network CN105242536A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510613438.1A CN105242536A (en) 2015-09-22 2015-09-22 Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510613438.1A CN105242536A (en) 2015-09-22 2015-09-22 Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network

Publications (1)

Publication Number Publication Date
CN105242536A true CN105242536A (en) 2016-01-13

Family

ID=55040217

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510613438.1A CN105242536A (en) 2015-09-22 2015-09-22 Unmanned aerial vehicle driving route waypoint calibration method based on BP nerve network

Country Status (1)

Country Link
CN (1) CN105242536A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1691251A2 (en) * 2005-02-10 2006-08-16 Northrop Grumman Corporation Synchronization of multiple operational flight programs
CN104268625A (en) * 2014-10-09 2015-01-07 哈尔滨工程大学 Autonomous underwater vehicle track predicating method based on marine environment information
CN104331085A (en) * 2014-11-03 2015-02-04 东华大学 Unmanned aerial vehicle control method based on PID (Proportion Integration Differentiation) neural network
CN104700076A (en) * 2015-02-13 2015-06-10 电子科技大学 Face image virtual sample generating method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1691251A2 (en) * 2005-02-10 2006-08-16 Northrop Grumman Corporation Synchronization of multiple operational flight programs
CN104268625A (en) * 2014-10-09 2015-01-07 哈尔滨工程大学 Autonomous underwater vehicle track predicating method based on marine environment information
CN104331085A (en) * 2014-11-03 2015-02-04 东华大学 Unmanned aerial vehicle control method based on PID (Proportion Integration Differentiation) neural network
CN104700076A (en) * 2015-02-13 2015-06-10 电子科技大学 Face image virtual sample generating method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
徐婷婷 等: "基于BP 神经网络的船舶航迹实时预测", 《大连海事大学学报》 *
段海庆 等: "基于反步自适应神经网络的船舶航迹控制", 《智能系统学报》 *
王薇 等: "基于神经网络的移动机器人路径规划", 《北京工业大学学报》 *
郭文刚: "基于BP 神经网络的船舶航迹控制技术", 《舰船科学技术》 *

Similar Documents

Publication Publication Date Title
CN105389988B (en) A kind of express highway intelligent cruising inspection system of multiple no-manned plane collaboration
JP6370040B2 (en) Aviation forest survey system
CN104808682B (en) Small-sized rotor wing unmanned aerial vehicle automatic obstacle avoiding flight control method
CN102859569B (en) Determine the emergency condition landing point of aircraft
CN104029825B (en) On-the-spot actual situation coupling Unmanned Aircraft Systems (UAS)
US7982659B2 (en) Method and system for reducing light pollution
Hüppop et al. Bird migration studies and potential collision risk with offshore wind turbines
CN103941747B (en) The control method of unmanned aerial vehicle group and system
Davis et al. Gliders measure western boundary current transport from the South Pacific to the equator
CN100583185C (en) Flight simulator
Schofield et al. Slocum gliders: Robust and ready
Cardone et al. Hindcasting the directional spectra of hurricane-generated waves
Abdelkader et al. A UAV based system for real time flash flood monitoring in desert environments using Lagrangian microsensors
US8774989B1 (en) Aircraft energy state display
CN102915465B (en) Multi-robot combined team-organizing method based on mobile biostimulation nerve network
CN102591358B (en) Multi-UAV (unmanned aerial vehicle) dynamic formation control method
CN103365299B (en) A kind of barrier-avoiding method of unmanned plane and device thereof
CN102645931B (en) For selecting meteorological data for the method updating aerial vehicle trajectory
Ferretti et al. Overview of the first HyMeX Special Observation Period over Italy: observations and model results
US9441985B2 (en) Future scene generating system, device, and method for a remotely-operated vehicle
CN104618689B (en) Coastal waters spilled oil monitoring method and system based on unmanned plane
CN105206108B (en) A kind of vehicle collision prewarning method based on electronic map
Mannarini et al. A prototype of ship routing decision support system for an operational oceanographic service
US7945355B2 (en) Flight control method using wind data from airplane trajectory
CN106020237B (en) The multimachine operation flight course planning and its spraying operation method and system of plant protection drone

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20160113

WD01 Invention patent application deemed withdrawn after publication