CN109556609A - A kind of collision prevention method and device based on artificial intelligence - Google Patents
A kind of collision prevention method and device based on artificial intelligence Download PDFInfo
- Publication number
- CN109556609A CN109556609A CN201811358583.XA CN201811358583A CN109556609A CN 109556609 A CN109556609 A CN 109556609A CN 201811358583 A CN201811358583 A CN 201811358583A CN 109556609 A CN109556609 A CN 109556609A
- Authority
- CN
- China
- Prior art keywords
- path
- collision prevention
- network model
- path point
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
- G01C21/203—Specially adapted for sailing ships
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Automation & Control Theory (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The collision prevention method and device based on artificial intelligence that the invention discloses a kind of.Wherein, this method comprises: establishing collision prevention Policy model；The collision prevention Policy model is sent by current motion data, obtains preferred solution space；Monte Carlo tree search algorithm is applied in the preferred solution space, adjusting parameter is calculated, the motion model of analytical form is not needed to establish not only, and the low defect with poor robustness of control precision that accurate manipulation mathematical model can not be established under the conditions of the large disturbances such as stormy waves stream is also overcomed, it realizes and improves intelligence, the technical effect of stability and applicability.
Description
Technical field
The present invention relates to artificial intelligence and collision prevention technical field more particularly to a kind of collision prevention method based on artificial intelligence and
Device.
Background technique
Research of the country in terms of water surface unmanned boat control technology is started late, but from initial conceptual phase by
Gradual transition is to the practice stage.Due to the complexity and non-intellectual of water surface unmanned boat working environment, need to continuously improve with it is complete
It is apt to existing smart system structure, is promoted to following predictive ability, the independent learning ability of strengthen the system makes intelligence system more
It is forward-looking.
There are many information content of required processing when ship collision prevention, and operation is considerably complicated, the research of relevant avoidance algorithm technology
Need to be goed deep into.Since current avoidance algorithm has limitation, lead to ship collision prevention mode not intelligence, and most of algorithms
Just for the collision prevention under broad sea area, the less environment of dynamic barrier.
Therefore, a kind of temporarily good without intelligence, stability is strong, suitable for the collision prevention method under Various Complex sea situation.
Summary of the invention
The present invention solves intelligent in the prior art by providing a kind of collision prevention method and device based on artificial intelligence
It is low, stability is poor, can not be suitable for the technical issues of Various Complex sea situation, realize raising intelligence, stability and applicability
Technical effect.
The collision prevention method based on artificial intelligence that the present invention provides a kind of, comprising:
Establish collision prevention Policy model；
The collision prevention Policy model is sent by current motion data, obtains preferred solution space；
Monte Carlo tree search algorithm is applied in the preferred solution space, adjusting parameter is calculated.
Further, described to send the collision prevention Policy model for current motion data, it obtains preferred solution space, wraps
It includes:
The collision prevention Policy model is sent by the current motion data, obtain the preferred solution space and estimates movement
Data；
Further include:
Establish evaluation network model relevant to automatic obstacle avoiding performance indicator；
The exercise data of estimating is sent in the evaluation network model, obtains assessed value；
It is described to apply to Monte Carlo tree search algorithm in the preferred solution space, adjusting parameter is calculated, comprising:
The Monte Carlo tree search algorithm is applied in the preferred solution space, the adjusting parameter is calculated；
The adjusting parameter is adjusted based on the assessed value, obtains optimal correction parameter.
It is further, described to establish evaluation network model relevant to automatic obstacle avoiding performance indicator, comprising:
Pass through formulaIt establishes and path length f1Relevant evaluation network model；
Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi, yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1,yi+1)。
It is further, described to establish evaluation network model relevant to automatic obstacle avoiding performance indicator, comprising:
Pass through formulaN > 0 is established and path smooth degree f2Relevant evaluation network model；
Wherein, αiFor the path point P at i momentiThe turning at place, expression formula areIn formula,
PiFor the path point at i moment, PiCoordinate be (xi,yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1, yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| it is vectorLength；For from path
Point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength；K is the number at the turning more than or equal to pi/2.
It is further, described to establish evaluation network model relevant to automatic obstacle avoiding performance indicator, comprising:
Pass through formulaIt establishes and path safety f3Relevant evaluation network model；
Wherein, λ is weight regulation coefficient；M is the second punishment parameter；D is all path points being averaged most apart from barrier
Short distance, expression formula areIn formula, n is the path point number in path in addition to starting point and target point；diFor
The shortest distance between path point and barrier, expression formula are WithRespectively
For path point PiThe two-end-point of place linkage lines.
The present invention also provides a kind of anticollision devices based on artificial intelligence, comprising:
Collision prevention Policy model establishes module, for establishing collision prevention Policy model；
Collision prevention Policy model computing module obtains excellent for sending the collision prevention Policy model for current motion data
Select solution space；
Adjustment is calculated for applying to Monte Carlo tree search algorithm in the preferred solution space in computing module
Parameter.
Further, the collision prevention Policy model computing module, comprising:
First model arithmetic unit obtains institute for sending the collision prevention Policy model for the current motion data
State preferred solution space；
Second model arithmetic unit obtains pre- for sending the collision prevention Policy model for the current motion data
Estimate exercise data；
Further include:
Evaluation network model establishes module, for establishing evaluation network model relevant to automatic obstacle avoiding performance indicator；
Network model computing module is evaluated, for sending the exercise data of estimating in the evaluation network model, is obtained
Assessed value out；
The computing module, comprising:
Computing unit is calculated for applying to the Monte Carlo tree search algorithm in the preferred solution space
The adjusting parameter；
Adjustment unit obtains optimal correction parameter for being adjusted based on the assessed value to the adjusting parameter.
Further, the evaluation network model establishes module, comprising:
First evaluation network model establishes unit, for passing through formulaIt establishes and path length f1It is related
Evaluation network model；Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi, yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1,yi+1)。
Further, the evaluation network model establishes module, comprising:
Second evaluation network model establishes unit, for passing through formulaN > 0 is established and path
Smoothness f2Relevant evaluation network model；Wherein, αiFor the path point P at i momentiThe turning at place, expression formula areIn formula, PiFor the path point at i moment, PiCoordinate be (xi,yi)；Pi+1For the i+1 moment
Path point, Pi+1Coordinate be (xi+1,yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| for
AmountLength；For from path point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength
Degree；K is the number at the turning more than or equal to pi/2.
Further, the evaluation network model establishes module, comprising:
Third evaluation network model establishes unit, for passing through formulaIt establishes and path safety f3It is related
Evaluation network model；Wherein, λ is weight regulation coefficient；M is the second punishment parameter；D is all path points apart from barrier
Average shortest distance, expression formula areIn formula, n is the path point in path in addition to starting point and target point
Number；diThe shortest distance between path point and barrier, expression formula are WithRespectively path point PiThe two-end-point of place linkage lines.
One or more technical solution provided in the present invention, has at least the following technical effects or advantages:
The present invention establishes the collision prevention Policy model based on deep learning (convolutional neural networks), and by current motion data
It is sent to collision prevention Policy model, obtains preferred solution space；Monte Carlo tree search algorithm is applied in preferred solution space again, is counted
Calculation is adjusted parameter, does not need the motion model for establishing analytical form not only, but also also overcomes in large disturbances such as stormy waves streams
Under the conditions of can not establish accurate manipulation mathematical model control precision is low and the defect of poor robustness, realize raising intelligence,
The technical effect of stability and applicability.
Detailed description of the invention
Fig. 1 is the flow chart of the collision prevention method provided in an embodiment of the present invention based on artificial intelligence；
Fig. 2 is the schematic illustration of the collision prevention method provided in an embodiment of the present invention based on artificial intelligence；
Fig. 3 is the process that Monte Carlo tree is established in the collision prevention method provided in an embodiment of the present invention based on artificial intelligence
Figure；
Fig. 4 is the turning schematic diagram in the collision prevention method provided in an embodiment of the present invention based on artificial intelligence；
Fig. 5 is the path point in the collision prevention method provided in an embodiment of the present invention based on artificial intelligence from barrier most short distance
From schematic diagram；
Fig. 6 is the module map of the anticollision device provided in an embodiment of the present invention based on artificial intelligence.
Specific embodiment
The embodiment of the present invention solves in the prior art by providing a kind of collision prevention method and device based on artificial intelligence
Intelligence it is low, stability is poor, can not be suitable for Various Complex sea situation the technical issues of, realize raising intelligence, stability and
The technical effect of applicability.
Technical solution in the embodiment of the present invention is to solve the above problems, general thought is as follows:
The embodiment of the present invention establishes the collision prevention Policy model based on deep learning (convolutional neural networks), and will currently transport
Dynamic data are sent to collision prevention Policy model, obtain preferred solution space；It is empty that Monte Carlo tree search algorithm is applied into preferred solution again
Between in, adjusting parameter is calculated, does not need the motion model for establishing analytical form not only, but also also overcomes in stormy waves stream etc.
The low defect with poor robustness of control precision that accurate manipulation mathematical model can not be established under the conditions of large disturbances, realizes raising intelligence
It can property, the technical effect of stability and applicability.
Above-mentioned technical proposal in order to better understand, in conjunction with appended figures and specific embodiments to upper
Technical solution is stated to be described in detail.
Referring to Fig. 1 and Fig. 2, the collision prevention method provided in an embodiment of the present invention based on artificial intelligence, comprising:
Step S110: collision prevention Policy model is established；
This step is illustrated:
Receive original aeronautical data；
Collision prevention Policy model is established based on original aeronautical data.
In the present embodiment, original aeronautical data be at least it is below any one:
The speed of a ship or plane of aircraft, the course of aircraft, wind speed, wind direction, the position of static obstruction, static obstruction and navigation
The relative bearing of device, static obstruction and the relative velocity of aircraft, the position of dynamic obstruction, dynamic obstruction and aircraft
Relative bearing, the quantity of the relative velocity of dynamic obstruction and aircraft, obstruction.
In order to improve the operation accuracy of collision prevention Policy model, the collision prevention Policy model of foundation is learnt.
Specifically learning process includes:
History collision prevention case data is input in collision prevention Policy model, with determine weight optimal inside prototype network and
Excitation function.Collision prevention Policy model at this time is the network of a deep learning.
In the present embodiment, history collision prevention case data be at least it is below any one:
Course during the longitude and latitude of obstruction, the longitude and latitude of collision prevention route, collision prevention, the speed of a ship or plane during collision prevention.
What needs to be explained here is that expert's steering data, Collision Accidents of Ships case data, International Maritime can also be collected
The data such as collision regulation construct data sample library, for exercising supervision study to collision prevention Policy model.
In order to realize the scalability of the embodiment of the present invention, in data sample library spare interface, other rule can be accessed at any time
Then it is used as sample.
Collision prevention Policy model setting constraint in order to further increase the operation accuracy of collision prevention Policy model, to succeeding in school
Condition.
In the present embodiment, constraint condition is the course of aircraft and/or the solution space of the speed of a ship or plane.
Step S120: collision prevention Policy model is sent by current motion data, obtains preferred solution space；
Specifically, collision prevention Policy model is sent by current motion data, autonomous learning is carried out based on constraint condition, is obtained
It is preferred that solution space.
Step S130: Monte Carlo tree search algorithm is applied in preferred solution space, adjusting parameter is calculated.
This step is illustrated:
Referring to Fig. 3, rectangle indicates root node, and the foundation of tree is extended downwards by root node.The state of the node is usually
Refer to that aircraft encounters obstruction and need to carry out collision prevention.Ellipse indicates that child node, child node are generating state transfers during collision prevention
General node.The transfer between node is generated when aircraft selection movement, the node is in store to be taken for a period of time at certain
Collision prevention strategy (i.e. solution space).Triangle indicate leaf node, represent search tree reach aircraft movement boundary or
Uncertain environment, the state of the node are divided into collision prevention success and collision prevention two kinds of situations of failure.Monte Carlo tree search algorithm is general
It is divided into 4 stages, i.e. choice phase, extension phase, dummy run phase and backtracking more new stage.Algorithm can be repeatedly carried out this 4
Stage, until meeting some specific condition of collision prevention.
In order to keep the adjusting parameter finally obtained optimal, step S120 is specifically included:
Collision prevention Policy model is sent by current motion data, obtain preferred solution space and estimates exercise data；
The embodiment of the invention also includes:
Establish evaluation network model relevant to automatic obstacle avoiding performance indicator；
Exercise data will be estimated to send in evaluation network model, obtain assessed value；
Then step S130 is specifically included:
Monte Carlo tree search algorithm is applied in preferred solution space, adjusting parameter is calculated；
Adjusting parameter is adjusted based on assessed value, obtains optimal correction parameter.
It should be noted that evaluation network model can be established simultaneously with collision prevention Policy model, it can also be in collision prevention strategy
It establishes, can also be established after collision prevention Policy model, the embodiment of the present invention is not particularly limited this before model.
In the present embodiment, automatic obstacle avoiding performance indicator includes at least: path length, path smooth degree and path safety
Property.
Specifically, the detailed process of evaluation network model relevant to path length is established are as follows:
Pass through formulaIt establishes and path length f1Relevant evaluation network model；
Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi, yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1,yi+1)。
In the same circumstances, navigation path length is shorter, and hours underway is shorter, evaluates higher.
Establish the detailed process of evaluation network model relevant to path smooth degree are as follows:
Pass through formulaN > 0 is established and path smooth degree f2Relevant evaluation network model；
Wherein, referring to fig. 4, αiFor the path point P at i momentiThe turning at place, expression formula are In formula, PiFor the path point at i moment, PiCoordinate be (xi,yi)；Pi+1For the path at i+1 moment
Point, Pi+1Coordinate be (xi+1,yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| it is vectorLength；For from path point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength；k
For the number at the turning more than or equal to pi/2, also referred to as the first punishment parameter, i.e., when a certain turning is greater than or equal to pi/2,
Target value is punished；As n=0, path is line of the starting point to target point, path smooth degree f2Value be 0.
Path smooth degree f2It is indicated with turning average value, turning average value is smaller, turns more smooth, then path is more flat
It is sliding, it evaluates higher.
Establish the detailed process of evaluation network model relevant to path safety are as follows:
Pass through formulaIt establishes and path safety f3Relevant evaluation network model；
Wherein, referring to Fig. 5, λ is weight regulation coefficient, for solving to seek average distance what the value is too small after inverse to ask
Topic, λ random value in 0-1；M is the second punishment parameter；In the present embodiment, m is zero for the shortest distance to barrier
The number of path point.D is average shortest distance of all path points apart from barrier, and expression formula isIn formula, n
For the path point number in path in addition to starting point and target point；diThe shortest distance between path point and barrier,
Expression formula isTakeWithSmaller value in two distances；WithRespectively path point PiThe two-end-point of place linkage lines.
In the same circumstances, average shortest distance of all path points apart from barrier is longer, then path is safer, evaluation
It is higher.
It should be noted that can will estimate exercise data be sent to evaluation network model relevant to path length, with
These three evaluation network models of the relevant evaluation network model of path smooth degree, evaluation network model relevant to path safety
In any one or any two or three in.Correspondingly, an assessed value or two assessed values or three are based on
A assessed value is adjusted adjusting parameter, obtains optimal correction parameter.It follows that adjusted based on three assessed values
Adjusting parameter is optimal.
Referring to Fig. 6, the anticollision device provided in an embodiment of the present invention based on artificial intelligence, comprising:
Collision prevention Policy model establishes module 100, for establishing collision prevention Policy model；
Specifically, collision prevention Policy model establishes module 100, comprising:
Data receipt unit, for receiving original aeronautical data；
Execution unit is established, for establishing collision prevention Policy model based on original aeronautical data.
In the present embodiment, original aeronautical data be at least it is below any one:
The speed of a ship or plane of aircraft, the course of aircraft, wind speed, wind direction, the position of static obstruction, static obstruction and navigation
The relative bearing of device, static obstruction and the relative velocity of aircraft, the position of dynamic obstruction, dynamic obstruction and aircraft
Relative bearing, the quantity of the relative velocity of dynamic obstruction and aircraft, obstruction.
In order to improve the operation accuracy of collision prevention Policy model, further includes:
Model learning module, for learning to the collision prevention Policy model of foundation.
In the present embodiment, model learning module, specifically for history collision prevention case data is input to collision prevention strategy mould
In type, to determine weight and excitation function optimal inside prototype network.Collision prevention Policy model at this time is a deep learning
Network.
Specifically, history collision prevention case data be at least it is below any one:
Course during the longitude and latitude of obstruction, the longitude and latitude of collision prevention route, collision prevention, the speed of a ship or plane during collision prevention.
In order to exercise supervision study to collision prevention Policy model, further includes:
Data collection module, for collecting expert's steering data, Collision Accidents of Ships case data, International Maritime collision prevention rule
The data such as then；
Sample database constructs module, for constructing data sample library.
In order to realize the scalability of the embodiment of the present invention, in data sample library spare interface, other rule can be accessed at any time
Then it is used as sample.
In order to further increase the operation accuracy of collision prevention Policy model, further includes:
Constraint condition setting module, for setting constraint condition to the collision prevention Policy model succeeded in school.
In the present embodiment, constraint condition is the course of aircraft and/or the solution space of the speed of a ship or plane.
Collision prevention Policy model computing module 200 obtains preferably for sending collision prevention Policy model for current motion data
Solution space；
In the present embodiment, collision prevention Policy model computing module 200, specifically for sending collision prevention for current motion data
Policy model carries out autonomous learning based on constraint condition, obtains preferred solution space.
Adjustment ginseng is calculated for applying to Monte Carlo tree search algorithm in preferred solution space in computing module 300
Number.
This step is illustrated:
Referring to Fig. 3, rectangle indicates root node, and the foundation of tree is extended downwards by root node.The state of the node is usually
Refer to that aircraft encounters obstruction and need to carry out collision prevention.Ellipse indicates that child node, child node are generating state transfers during collision prevention
General node.The transfer between node is generated when aircraft selection movement, the node is in store to be taken for a period of time at certain
Collision prevention strategy (i.e. solution space).Triangle indicate leaf node, represent search tree reach aircraft movement boundary or
Uncertain environment, the state of the node are divided into collision prevention success and collision prevention two kinds of situations of failure.Monte Carlo tree search algorithm is general
It is divided into 4 stages, i.e. choice phase, extension phase, dummy run phase and backtracking more new stage.Algorithm can be repeatedly carried out this 4
Stage, until meeting some specific condition of collision prevention.
In the present embodiment, collision prevention Policy model computing module 200, comprising:
First model arithmetic unit obtains preferred solution space for sending collision prevention Policy model for current motion data；
Second model arithmetic unit obtains for sending collision prevention Policy model for current motion data and estimates movement number
According to；
In order to keep the adjusting parameter finally obtained optimal, the embodiment of the invention also includes:
Evaluation network model establishes module, for establishing evaluation network model relevant to automatic obstacle avoiding performance indicator；
Network model computing module is evaluated, is sent in evaluation network model for exercise data will to be estimated, obtains assessed value；
Then computing module 300, comprising:
Adjusting parameter is calculated for applying to Monte Carlo tree search algorithm in preferred solution space in computing unit；
Adjustment unit obtains optimal correction parameter for being adjusted based on assessed value to adjusting parameter.
It should be noted that evaluation network model can be established simultaneously with collision prevention Policy model, it can also be in collision prevention strategy
It establishes, can also be established after collision prevention Policy model, the embodiment of the present invention is not particularly limited this before model.
In the present embodiment, automatic obstacle avoiding performance indicator includes at least: path length, path smooth degree and path safety
Property.
Specifically, evaluation network model establishes module, comprising:
First evaluation network model establishes unit, for passing through formulaIt establishes and path length f1It is related
Evaluation network model；Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi,yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1, yi+1).In the same circumstances, navigation path length is got over
Short, hours underway is shorter, evaluates higher.
Second evaluation network model establishes unit, for passing through formulaN > 0 is established and path
Smoothness f2Relevant evaluation network model；Wherein, αiFor the path point P at i momentiThe turning at place, expression formula areIn formula, PiFor the path point at i moment, PiCoordinate be (xi,yi)；Pi+1For the i+1 moment
Path point, Pi+1Coordinate be (xi+1, yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| for
AmountLength；For from path point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength
Degree；K be more than or equal to pi/2 turning number, also referred to as the first punishment parameter, i.e., when a certain turning be greater than or equal to π/
When 2, target value is punished；As n=0, path is line of the starting point to target point, path smooth degree f2Value be 0.
In the present embodiment, path smooth degree f2It is indicated with turning average value, turning average value is smaller, turns more smooth, then path
It is more smooth, it evaluates higher.
Third evaluation network model establishes unit, for passing through formulaIt establishes and path safety f3It is related
Evaluation network model；Wherein, λ is weight regulation coefficient, and for solving the problems, such as to seek average distance, the value is too small after inverse,
λ random value in 0-1；M is the second punishment parameter；In the present embodiment, m is the path for being zero to the shortest distance of barrier
The number of point.D is average shortest distance of all path points apart from barrier, and expression formula isIn formula, n is road
Path point number in diameter in addition to starting point and target point；diThe shortest distance between path point and barrier, expression
Formula isIt takesWithSmaller value in two distances；WithRespectively
For path point PiThe two-end-point of place linkage lines.In the same circumstances, average shortest distance of all path points apart from barrier
Longer, then path is safer, evaluates higher.
It establishes unit it should be noted that can will estimate exercise data and be sent to the first evaluation network model, second comment
Valence network model establish unit, third evaluation network model establish unit these three evaluation network models in any one or
In any two or three.Correspondingly, adjustment is joined based on an assessed value or two assessed values or three assessed values
Number is adjusted, and obtains optimal correction parameter.It follows that optimal based on the adjusting parameter that three assessed values adjust.
[technical effect]
1, the embodiment of the present invention establishes the collision prevention Policy model based on deep learning (convolutional neural networks), and will be current
Exercise data is sent to collision prevention Policy model, obtains preferred solution space；Monte Carlo tree search algorithm is applied into preferred solution again
In space, adjusting parameter is calculated, does not need the motion model for establishing analytical form not only, but also also overcomes in stormy waves stream
The low defect with poor robustness of control precision that accurate manipulation mathematical model can not be established under the conditions of equal large disturbances, realizes raising
Intelligence, the technical effect of stability and applicability.
2, the assessed value that Utilization assessment network model provides assesses the movement selection of aircraft, in conjunction with Monte Carlo
Tree search algorithm updates the evaluation of estimate in entire path under the movement, and the course until searching and speed of a ship or plane value restrain, and as most
Excellent adjusting parameter is given to executing agency, improves the validity of collision prevention.
3, expert's steering data (including real ship sea examination, historical experience etc.), accident case, International Regulations for Preventing Collisions at Sea are based on
Etc. data establish training sample database, establish the collision prevention Policy model of basic depth enhancing study, not only sufficiently combine expert behaviour
The relevant history information of rudder experience and accident case improves the feasibility of Decision of Collision Avoidance；And have also combined International Maritime
Collision regulation is more in line with practical collision prevention situation, further improves the applicability of collision prevention.
4, in data sample library spare interface, other rules can be accessed at any time as sample, realize the embodiment of the present invention
Scalability, further improve the applicability of the embodiment of the present invention.
The embodiment of the present invention is applicable not only to broad sea area, but also suitable for the complicated ocean such as high sea condition, narrow water
Environment, the effective guarantee navigation safety of ship.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (10)
1. a kind of collision prevention method based on artificial intelligence characterized by comprising
Establish collision prevention Policy model；
The collision prevention Policy model is sent by current motion data, obtains preferred solution space；
Monte Carlo tree search algorithm is applied in the preferred solution space, adjusting parameter is calculated.
2. the method as described in claim 1, which is characterized in that described to send the collision prevention strategy mould for current motion data
Type obtains preferred solution space, comprising:
The collision prevention Policy model is sent by the current motion data, obtain the preferred solution space and estimates movement number
According to；
Further include:
Establish evaluation network model relevant to automatic obstacle avoiding performance indicator；
The exercise data of estimating is sent in the evaluation network model, obtains assessed value；
It is described to apply to Monte Carlo tree search algorithm in the preferred solution space, adjusting parameter is calculated, comprising:
The Monte Carlo tree search algorithm is applied in the preferred solution space, the adjusting parameter is calculated；
The adjusting parameter is adjusted based on the assessed value, obtains optimal correction parameter.
3. method according to claim 2, which is characterized in that described to establish evaluation net relevant to automatic obstacle avoiding performance indicator
Network model, comprising:
Pass through formulaIt establishes and path length f1Relevant evaluation network model；
Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi,yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1, yi+1)。
4. method according to claim 2, which is characterized in that described to establish evaluation net relevant to automatic obstacle avoiding performance indicator
Network model, comprising:
Pass through formulaN > 0 is established and path smooth degree f2Relevant evaluation network model；
Wherein, αiFor the path point P at i momentiThe turning at place, expression formula areIn formula, PiFor i
The path point at moment, PiCoordinate be (xi,yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1, yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| it is vectorLength；For from path
Point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength；K is the number at the turning more than or equal to pi/2.
5. method according to claim 2, which is characterized in that described to establish evaluation net relevant to automatic obstacle avoiding performance indicator
Network model, comprising:
Pass through formulaIt establishes and path safety f3Relevant evaluation network model；
Wherein, λ is weight regulation coefficient；M is the second punishment parameter；D is average most short distance of all path points apart from barrier
From expression formula isIn formula, n is the path point number in path in addition to starting point and target point；diFor path
The shortest distance between point and barrier, expression formula are WithRespectively road
Diameter point PiThe two-end-point of place linkage lines.
6. a kind of anticollision device based on artificial intelligence characterized by comprising
Collision prevention Policy model establishes module, for establishing collision prevention Policy model；
Collision prevention Policy model computing module obtains preferred solution for sending the collision prevention Policy model for current motion data
Space；
Adjusting parameter is calculated for applying to Monte Carlo tree search algorithm in the preferred solution space in computing module.
7. device as claimed in claim 6, which is characterized in that the collision prevention Policy model computing module, comprising:
First model arithmetic unit obtains described excellent for sending the collision prevention Policy model for the current motion data
Select solution space；
Second model arithmetic unit obtains for sending the collision prevention Policy model for the current motion data and estimates fortune
Dynamic data；
Further include:
Evaluation network model establishes module, for establishing evaluation network model relevant to automatic obstacle avoiding performance indicator；
Network model computing module is evaluated, for sending the exercise data of estimating in the evaluation network model, obtains and comments
Valuation；
The computing module, comprising:
Computing unit is calculated described for applying to the Monte Carlo tree search algorithm in the preferred solution space
Adjusting parameter；
Adjustment unit obtains optimal correction parameter for being adjusted based on the assessed value to the adjusting parameter.
8. device as claimed in claim 7, which is characterized in that the evaluation network model establishes module, comprising:
First evaluation network model establishes unit, for passing through formulaIt establishes and path length f1It is relevant to comment
Valence network model；Wherein, diFor the path point at i moment and the distance between the path point at i+1 moment, expression formula isIn formula, PiFor the path point at i moment, PiCoordinate be
(xi,yi)；Pi+1For the path point at i+1 moment, Pi+1Coordinate be (xi+1,yi+1)。
9. device as claimed in claim 7, which is characterized in that the evaluation network model establishes module, comprising:
Second evaluation network model establishes unit, for passing through formulaN > 0 is established and path smooth
Spend f2Relevant evaluation network model；Wherein, αiFor the path point P at i momentiThe turning at place, expression formula areIn formula, PiFor the path point at i moment, PiCoordinate be (xi,yi)；Pi+1For the i+1 moment
Path point, Pi+1Coordinate be (xi+1, yi+1)；For from path point Pi-1To path point PiVector, | Pi-1Pi| for
AmountLength；For from path point PiTo path point Pi+1Vector, | PiPi+1| it is vectorLength；
K is the number at the turning more than or equal to pi/2.
10. device as claimed in claim 7, which is characterized in that the evaluation network model establishes module, comprising:
Third evaluation network model establishes unit, for passing through formulaIt establishes and path safety f3Relevant evaluation
Network model；Wherein, λ is weight regulation coefficient；M is the second punishment parameter；D is all path points being averaged most apart from barrier
Short distance, expression formula areIn formula, n is the path point number in path in addition to starting point and target point；diFor
The shortest distance between path point and barrier, expression formula are WithRespectively
For path point PiThe two-end-point of place linkage lines.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811358583.XA CN109556609B (en) | 2018-11-15 | 2018-11-15 | Artificial intelligence-based collision avoidance method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811358583.XA CN109556609B (en) | 2018-11-15 | 2018-11-15 | Artificial intelligence-based collision avoidance method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109556609A true CN109556609A (en) | 2019-04-02 |
| CN109556609B CN109556609B (en) | 2020-11-17 |
Family
ID=65866434
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811358583.XA Active CN109556609B (en) | 2018-11-15 | 2018-11-15 | Artificial intelligence-based collision avoidance method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109556609B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113076338A (en) * | 2021-02-04 | 2021-07-06 | 中国船级社 | Rule-based intelligent ship collision avoidance automatic test scene generation method and system |
| CN113917929A (en) * | 2021-11-11 | 2022-01-11 | 中国船舶重工集团公司第七一九研究所 | Unmanned ship path optimization method and system based on hybrid particle swarm algorithm |
| CN114111824A (en) * | 2021-11-11 | 2022-03-01 | 中国船舶重工集团公司第七一九研究所 | Unmanned ship path optimization method and system based on multi-objective optimization |
| CN114313142A (en) * | 2022-03-04 | 2022-04-12 | 中国船级社 | Evaluation method, device and equipment for ship collision avoidance algorithm |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794526A (en) * | 2015-05-13 | 2015-07-22 | 哈尔滨工程大学 | Automatic ship anti-collision method optimized by wolf colony search algorithm |
| CN106548645A (en) * | 2016-11-03 | 2017-03-29 | 济南博图信息技术有限公司 | Vehicle route optimization method and system based on deep learning |
| CN106980317A (en) * | 2017-03-31 | 2017-07-25 | 大鹏高科（武汉）智能装备有限公司 | A kind of obstacle avoidance method and system under water |
| CN107037809A (en) * | 2016-11-02 | 2017-08-11 | 哈尔滨工程大学 | A kind of unmanned boat collision prevention method based on improvement ant group algorithm |
| CN107272705A (en) * | 2017-07-31 | 2017-10-20 | 中南大学 | A kind of multiple neural network controlling planning method of robot path under intelligent environment |
| US20170308836A1 (en) * | 2016-04-22 | 2017-10-26 | Accenture Global Solutions Limited | Hierarchical visualization for decision review systems |
| CN108320051A (en) * | 2018-01-17 | 2018-07-24 | 哈尔滨工程大学 | A kind of mobile robot dynamic collision-free planning method based on GRU network models |
| CN108399201A (en) * | 2018-01-30 | 2018-08-14 | 武汉大学 | A kind of Web user access path prediction technique based on Recognition with Recurrent Neural Network |
| CN108772835A (en) * | 2018-05-03 | 2018-11-09 | 华南理工大学 | A kind of barrier hides method with physics limit |
-
2018
- 2018-11-15 CN CN201811358583.XA patent/CN109556609B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794526A (en) * | 2015-05-13 | 2015-07-22 | 哈尔滨工程大学 | Automatic ship anti-collision method optimized by wolf colony search algorithm |
| US20170308836A1 (en) * | 2016-04-22 | 2017-10-26 | Accenture Global Solutions Limited | Hierarchical visualization for decision review systems |
| CN107037809A (en) * | 2016-11-02 | 2017-08-11 | 哈尔滨工程大学 | A kind of unmanned boat collision prevention method based on improvement ant group algorithm |
| CN106548645A (en) * | 2016-11-03 | 2017-03-29 | 济南博图信息技术有限公司 | Vehicle route optimization method and system based on deep learning |
| CN106980317A (en) * | 2017-03-31 | 2017-07-25 | 大鹏高科（武汉）智能装备有限公司 | A kind of obstacle avoidance method and system under water |
| CN107272705A (en) * | 2017-07-31 | 2017-10-20 | 中南大学 | A kind of multiple neural network controlling planning method of robot path under intelligent environment |
| CN108320051A (en) * | 2018-01-17 | 2018-07-24 | 哈尔滨工程大学 | A kind of mobile robot dynamic collision-free planning method based on GRU network models |
| CN108399201A (en) * | 2018-01-30 | 2018-08-14 | 武汉大学 | A kind of Web user access path prediction technique based on Recognition with Recurrent Neural Network |
| CN108772835A (en) * | 2018-05-03 | 2018-11-09 | 华南理工大学 | A kind of barrier hides method with physics limit |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113076338A (en) * | 2021-02-04 | 2021-07-06 | 中国船级社 | Rule-based intelligent ship collision avoidance automatic test scene generation method and system |
| CN113917929A (en) * | 2021-11-11 | 2022-01-11 | 中国船舶重工集团公司第七一九研究所 | Unmanned ship path optimization method and system based on hybrid particle swarm algorithm |
| CN114111824A (en) * | 2021-11-11 | 2022-03-01 | 中国船舶重工集团公司第七一九研究所 | Unmanned ship path optimization method and system based on multi-objective optimization |
| CN114313142A (en) * | 2022-03-04 | 2022-04-12 | 中国船级社 | Evaluation method, device and equipment for ship collision avoidance algorithm |
| CN114313142B (en) * | 2022-03-04 | 2022-06-07 | 中国船级社 | Evaluation method, device and equipment for ship collision avoidance algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109556609B (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111694365B (en) | Unmanned ship formation path tracking method based on deep reinforcement learning | |
| CN106845716B (en) | Navigation error constraint-based water surface unmanned ship local hierarchical path planning method | |
| CN109556609A (en) | A kind of collision prevention method and device based on artificial intelligence | |
| CN112099496B (en) | Automatic driving training method, device, equipment and medium | |
| CN110850873B (en) | Unmanned ship path planning method, device, equipment and storage medium | |
| CN110956853B (en) | Multi-ship collision prediction method, system and storage medium | |
| Havenstrøm et al. | Deep reinforcement learning controller for 3D path following and collision avoidance by autonomous underwater vehicles | |
| Wang et al. | Ship route planning based on double-cycling genetic algorithm considering ship maneuverability constraint | |
| Zheng et al. | A Decision‐Making Method for Ship Collision Avoidance Based on Improved Cultural Particle Swarm | |
| Wang et al. | Local path optimization method for unmanned ship based on particle swarm acceleration calculation and dynamic optimal control | |
| CN110906935A (en) | Unmanned ship path planning method | |
| CN113821035A (en) | Unmanned ship trajectory tracking control method and device | |
| CN116476863A (en) | Automatic driving transverse and longitudinal integrated decision-making method based on deep reinforcement learning | |
| Wang et al. | A collision avoidance method for intelligent ship based on the improved bacterial foraging optimization algorithm | |
| CN113538973B (en) | Automatic ship collision avoidance method based on improved particle swarm optimization | |
| Sawada | Automatic collision avoidance using deep reinforcement learning with grid sensor | |
| Yang et al. | Improved reinforcement learning for collision-free local path planning of dynamic obstacle | |
| Wen et al. | Online heuristically planning for relative optimal paths using a stochastic algorithm for USVs | |
| US12020577B2 (en) | Ship maneuvering calculation device | |
| Yuan et al. | EMPMR berthing scheme: A novel event-triggered motion planning and motion replanning scheme for unmanned surface vessels | |
| CN114609925B (en) | Training method of underwater exploration strategy model and underwater exploration method of bionic machine fish | |
| Nguyen et al. | Automatic collision avoiding support system for ships in congested waters and at open sea | |
| Uçan et al. | Using genetic algorithms for navigation planning in dynamic environments | |
| CN116300913A (en) | Unmanned ship multi-constraint local path planning method based on visual information | |
| Jose et al. | Navigating the Ocean with DRL: Path following for marine vessels |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20190402 Assignee: Wuhan Financial Leasing Co.,Ltd. Assignor: WUHAN NANHUA INDUSTRIAL EQUIPMENT ENGINEERING Co.,Ltd. Contract record no.: X2023420000146 Denomination of invention: A collision avoidance method and device based on artificial intelligence Granted publication date: 20201117 License type: Exclusive License Record date: 20230607 |
|
| EE01 | Entry into force of recordation of patent licensing contract |